the games of july
the games of july Explaining the Great War
frank c. zagare
the university of michigan press ann a...
87 downloads
769 Views
2MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
the games of july
the games of july Explaining the Great War
frank c. zagare
the university of michigan press ann arbor
Copyright © by the University of Michigan 2011 All rights reserved Published in the United States of America by The University of Michigan Press Manufactured in the United States of America Printed on acid-free paper 2014 2013 2012 2011
4 3 2 1
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, or otherwise, without the written permission of the publisher. A CIP catalog record for this book is available from the British Library. Library of Congress Cataloging-in-Publication Data Zagare, Frank C. The games of July : explaining the Great War / Frank C. Zagare. p.
cm.
Includes bibliographical references and index. isbn 978-0-472-07116-6 (cloth : alk. paper) — isbn 978-0-47205116-8 (pbk. : alk. paper) — isbn 978-0-472-02370-7 (e-book) 1. World War, 1914-1918—Causes.
2. World politics—1900–1918.
3. World politics—19th century. d511.z34
I. Title.
2010
940.3'1—dc22
2010014035
To Sefa
Contents
Preface
ix
part one. theoretical underpinnings chapter one. Purpose and Method
1.1. 1.2. 1.3. 1.4. 1.5. 1.6.
Bringing Theory to Bear on Data Bringing Data to Bear on Theory A Possible Objection and a Rejoinder Blueprint Scylla and Charybdis Coda
chapter two. Theories and Explanations
2.1. The Realist Paradigm and Classical Deterrence Theory 2.2. Coda chapter three. Perfect Deterrence Theory: An Overview
3.1. Axioms and General Theoretical Characteristics 3.2. Explaining the Absence of War 3.3. Theoretical Propositions, Empirical Expectations, and Policy Implications 3.4. Coda
3 6 7 9 10 12 14 19 21 37 39 40 44 49 56
part two. explaining the great war chapter four. Bismarck’s System
4.1. The Tripartite Crisis Game 4.2. Outcomes and Preferences 4.3. Analysis 4.4. Explaining the Austro-German Alliance of 1879 4.5. Coda
61 64 67 72 87 89
chapter five. Vienna, Berlin, and the Blank Check
5.1. 5.2. 5.3. 5.4.
Background The Tripartite Crisis Game Redux Analysis Coda
chapter six. La Guerre Européenne
6.1. Asymmetric Escalation Game 6.2. Preferences 6.3. Some Caveats 6.4. Analysis 6.5. Discussion 6.6. Coda chapter seven. Britain’s Strategic Dilemma
7.1. 7.2. 7.3. 7.4. 7.5. 7.6.
Modeling the Deterrence versus Restraint Dilemma Preference Assumptions Information and Utility Assumptions Analysis Explaining Britain’s Foreign Policy in 1914 Coda
90 91 96 100 106 110 113 118 120 123 128 143 144 147 148 151 152 158 166
part three. endgame 8.1. Explanation 8.2. Theoretical Assessment 8.3. Implications and Final Thoughts
171 172 181 186
References Index
195 209
chapter eight. Questions, Answers, Implications
In Flanders ‹elds the poppies blow Between the crosses, row on row john mccray
Preface
The First World War is doubtless the seminal event of the twentieth century. It is safe to say that had it not occurred, the history of the past hundred years would have been dramatically different. Had major-power war been avoided in 1914, it is likely that there would have been no Soviet Union, no Hitler, no Second World War, no Cold War, no nuclear arms race, and no post–Cold War period, to offer just a few possible counterfactuals.1 In this book, I explain to the reader and to myself why this cataclysmic war occurred. Implied in my purpose is dissatisfaction with existing historical explanations and with two in particular. The ‹rst is the revisionist explanation that the participants mindlessly “slithered” into the con›ict. In this view, the war that “no one wanted” was just a big mistake. To explain away the war as an unfortunate accident, however, is to beg the question and to offer no explanation at all. As Israel’s former prime minister, Menachem Begin, once asked rhetorically, “If both sides don’t want war, how can war break out?” In the pages that follow, I show why the “inadvertent war theory” fails to suf‹ce. The second historical explanation that falls short of the mark is the thesis that the First World War was simply willed by Germany. If accepted, the explanation is straightforward. Fearing a rising Russia, Germany instigated a preemptive war in 1914. Interpreted in this way, deterrence failed because, collectively, Russia, France, and possibly Great Britain lacked a capable retaliatory threat. As I demonstrate, this explanation, while logically unassailable, is empirically dubious. Having rejected these two case-speci‹c explanations, I next sought insight into the outbreak of the First World War by examining the causes of war in general. But the most prominent theoretical explanations were also found wanting. 1. For additional conjectures, see Cowley 2001. For commentary on the theoretical issues raised by counterfactual argumentation, see Tetlock and Belkin 1996.
x
preface
If anything, the First World War seemed to refute the liberal hypothesis that high levels of economic interdependence all but eliminate the likelihood of major-power war. I also judged various realist explanations to be de‹cient, some because they are logically incoherent, others because they are empirically questionable, and still others because they are theoretically degenerative. To explain the First World War, therefore, I turned to perfect deterrence theory, a general theory of international con›ict initiation, escalation, and war avoidance that I have developed in collaboration with D. Marc Kilgour. But, as originally speci‹ed, perfect deterrence theory was not immediately applicable to certain aspects of the con›ict. Accordingly, in a pair of articles, Kilgour and I expanded the theory to capture more fully the dynamics of extended deterrence relationships such as those that existed just prior to the outbreak of war in 1914. These axiomatically compatible extensions of our original theoretical framework form the basis of chapters 4 and 7. I am deeply grateful to Marc Kilgour for permission to draw on material from our jointly authored works for use in this book and more generally for what I have learned from him over the many years we have worked together. I also acknowledge several other individuals who have helped me, either directly or indirectly, write this book. Stephen Quackenbush read several chapters, delivered sage advice, and greatly improved the ‹nal product. I could not ask for a more helpful editor than Melody Herr at the University of Michigan Press. Her great patience and support have gone above and beyond the call of duty. Others who have offered criticism, insightful comments, and/or encouragement include Steven Brams, Bruce Bueno de Mesquita, Claudio Ciof‹-Revilla, Ray Dacey, Vesna Danilovic, Bill Geasor, Pat James, Jacek Kugler, Douglas Lemke, Paul Senese, Randy Siverson, and John Vasquez. My three daughters, Catherine, Ann, and Elizabeth Zagare, have also been a source of much inspiration, each in her own way. (This is, after all, a book about deterrence failure.) I especially thank my favorite reference librarian, Patricia Zagare, who was on the constant lookout for relevant source material and who provided me with a number of invaluable leads. More than anyone, she has helped me survive this project. Finally, I thank my mother, to whom this book is dedicated. To her I wish to say, “I did it for you.”
part one
Theoretical Underpinnings
The science of politics . . . explains the mystery of how social choice evolves out of individual values. william h. riker and peter c. ordeshook
chapter one
Purpose and Method
Until Germany invaded Poland in 1939, it was known simply as the Great War. The label was an understatement. The carnage, both human and political, was unprecedented. According to one estimate, more than 9,000,000 lives were lost: 1,800,000 Germans and 1,700,000 Russians died; France lost 1,350,000 of its citizens, Austria-Hungary more than 1,200,000; Great Britain lost almost 1,000,000 soldiers, and Italy more than 650,000. Lesser but still substantial losses were borne by the United States, Serbia, Belgium, Turkey, and Rumania. Japan, Bulgaria, Portugal, and Greece also sustained casualties (Singer and Small 1982: 89). As horri‹c as the loss of human capital was, it tells only part of the story. In the aftermath of the war, four great empires—the Russian, the Ottoman, the German, and the Austro-Hungarian—were obliterated. Another, the British, was placed on the critical list. More signi‹cantly, the war set loose a variety of forces that would, in short order, lead to the establishment of totalitarian regimes in Germany, Italy, Russia, Spain, and several other European countries. Within a generation, history would repeat itself, and all the major powers of the world would be at war once again. At this point, sadly, the Great War assumed another name, World War I. Although some have, it is dif‹cult to read the diplomatic record of World War I and conclude that it was inevitable. There were so many points at which it could easily have gone the other way (Mombauer 2001: 288). So how could such an unmitigated disaster have occurred? How could a war that “no one wanted” break out? These are, of course, loaded questions that immediately raise a number of vexing philosophical, ontological, and epistemological issues. Nonetheless, one purpose of this book is to address them by explaining: 3
4
the games of july
• why Germany entered into a defensive alliance with Austria in 1879, an “unlikely” alliance that seemed to offer Germany few tangible bene‹ts; • why Germany issued a blank check to Austria on July 5, 1914, and in so doing abrogated control of its foreign policy to another, lesser, power; • why a local war between Austria and Serbia escalated into a continental war that included Germany, Russia, and France; and • why British foreign policy failed to deter a German declaration of war against France on August 3, 1914, a failure that brought about a worldwide con›ict when Britain declared war on Germany the next day. Taken together, the answers to these questions form the basis of an explanation of the initiation, the escalation, and the expansion of a long-simmering dispute between a major and a minor power that would eventually include all of the great states in the international system. Explanation, therefore, rather than description is a primary objective of this book. In consequence, many of the details normally found in histories of the Great War are omitted from this analysis. The reader looking for a “thick” description of the events of July and early August 1914 will have to look elsewhere.1 Nonetheless, in addressing these questions, I hope to provide a coherent description of the chain of events that brought about World War I. To construct an explanation for the outbreak of the Great War, I take advantage of some recent advances in the theory of noncooperative games. More speci‹cally, I use a set of interrelated game-theoretic models called perfect deterrence theory (Zagare and Kilgour 2000) to answer the four central questions.2 My explanation therefore is unabashedly theoretical. For this I offer no apologies. As King, Keohane, and Verba (1994: 46) note, “The best scienti‹c way to organize facts is as observable implications of some theory or hypothesis.” Perfect deterrence theory is well suited for this purpose. As discussed in detail in chapter 3, perfect deterrence theory is a logically consistent and empirically plausible theory of interstate con›ict initiation, limitation, and escala1. Albertini’s (1952) three-volume account would be a good place to start. For an overview of the recent historiography relevant to the July Crisis, see Williamson and May 2007. A useful summary can be found in Meyer 2006: 1–87. Nomikos and North’s (1976) study, however, is the most judicious description of the diplomatic record. Finally, Trachtenberg’s (1990/91) analysis of the run-up to the war is an exemplar of rigorous historical research that should not be overlooked. 2. Basic game-theoretic concepts are introduced as they become relevant. For expository purposes, I assume that the reader is uninterested in certain technical details, including some threshold de‹nitions and all formal proofs. Section 1.6 provides a nontechnical overview of noncooperative game theory and its solution concepts.
Purpose and Method
5
tion.3 Because it is game-theoretic, it explicitly takes into account the interconnected decision-making process that is part and parcel of the web of facts that surround the Great War. Any explanation of a complex event such as the Great War that “assumes the absence of strategic interaction and anticipated reactions” is likely to be de‹cient (King, Keohane, and Verba 1994: 45–46).4 More important, though, is the fact that the equilibria of a (noncooperative) game model provide a solid foundation on which to construct explanations of real-world events and processes, such as those that preceded the outbreak of war in 1914. In section 1.6, I provide an informal discussion of the theoretical signi‹cance of game-theoretic equilibria and the precise sense in which they contribute to the development of an explanation. For some scholars (e.g., Rapoport 1958), game theory is strictly a normative theory that prescribes behavior to (instrumentally) rational decision makers (players) who are involved in an interactive situation (a game) whose outcome partly depends on the choices made by at least one other player.5 Game theory has long been used in the ‹eld of security studies for strictly normative purposes.6 Indeed, I use it in this way in chapter 7 to both analyze and assess British foreign policy in 1914. There is little doubt that game theory is a potentially powerful prescriptive tool. But Rapoport notwithstanding, there is no good reason why it cannot also be used descriptively to describe and explain decision making in the real world.7 As with many deductive models, game-theoretic models start with a small set of assumptions and rigorously de‹ned terms that permit the derivation of statements that, when they are operationalized, are subject to empirical application and testing. There is also a long tradition in the ‹eld of security studies of using game theory descriptively to explain interstate behavior. Riker (1962), for example, 3. Perfect deterrence theory is a general theory of interstate con›ict; its logical and empirical domain is not restricted to great-power wars or other types of armed con›ict. 4. F. Fischer’s (1967, 1975) explanation of the outbreak of the Great War has been criticized on precisely these grounds. Fischer’s strong assumptions about Germany’s preferences eliminate the necessity to take into account the policies of other states. It is highly likely, however, that Fischer’s failure to take into account strategic interaction led him to what many scholars consider questionable conclusions about the motivation of Germany’s leadership group. These conclusions are discussed in detail in chapter 6. 5. Instrumental rationality should be distinguished from the concept of procedural rationality, with which it is often con›ated. Instrumentally rational behavior should be understood as behavior that is no more and no less than purposeful behavior (Riker and Ordeshook 1973). The requirements of procedural rationality, by contrast, are much more demanding (Simon 1976). Procedurally rational actors are, in essence, omniscient actors who make “cool and clearheaded ends-means calculations” (Verba 1961: 95) based, inter alia, on a careful evaluation of all possible courses of action. For further discussion, see Zagare 1990. 6. See, e.g., Zagare 1983. 7. See Elster 2000 for the counterargument.
6
the games of july
used his theory of minimal winning coalitions to explain why the three known examples of grand coalitions forming in the international system fell apart so quickly.8 More recently, Bates et al. (1998) have used game theory to construct explanations (that they call analytic narratives) of a number of real-world processes. In so doing, their aim was “not only to bring theory to bear upon data but also to bring data to bear upon theory” (2000: 698). Their aim is also mine. In addition to providing a theoretically consistent explanation of the Great War, I use the events of July 1914 to evaluate perfect deterrence theory. Theoretical assessment, therefore, constitutes the second major purpose of this work. Although it is second, the goal of assessing perfect deterrence theory is in no sense secondary. Neither of this work’s two purposes should be understood to trump the other. Indeed, the development of a theory-based explanation of the Great War and the assessment of perfect deterrence theory within the context of the July Crisis are, in general, synergistic enterprises, and there are many advantages of treating them as such.
1.1. Bringing Theory to Bear on Data One very good reason to examine speci‹c events or processes within the con‹nes of a well-articulated theory is the additional organizing power that is acquired by doing so. From the in‹nite variety of observations about an event or a process that might be made, a theory singles some out for special consideration. In the case of a game-theoretic framework such as perfect deterrence theory, these categories include, but are not limited to, the identi‹cation of the players, the choices they face, the set of possible outcomes, the players’ preferences over the outcome set, the private information each player possesses, information that is common to all the players, and each player’s beliefs about what is likely to take place as the game is played out. At the same time, the explicit use of a theoretical framework makes analysis more tractable by suggesting what information can or should safely be ignored. Explorations of real-world political phenomena are also rendered less ad hoc when they are theoretically informed. Theoretical frameworks severely limit not only the number but also the cast of variables that can be called on to provide a coherent explanation. While this is true of all theories, it is especially true of formal theories such as perfect deterrence theory, since formalization 8. The three instances of overwhelming majorities in international politics are the Concert of Europe, an alliance composed of England, Prussia, Russia, France, and Austria that was institutionalized at the Congress of Vienna in 1815 following Napoleon’s ‹nal defeat, and the allied coalitions that came about when Germany collapsed after the two twentieth-century world wars.
Purpose and Method
7
requires an explicit statement of assumptions and arguments. Formal theories, in other words, are more transparent. In consequence, they are at once subject to more intense scrutiny and less amenable to even unintended manipulation (Snidal 2002: 80). Explanations derived from deductive methodologies such as game theory have the added bene‹t of clarity. The relationship between the premises and the conclusions of game-theoretic models explains why something must be the case. As Kaplan (1964: 339) pointed out long ago, “The explanation shows that, on the basis of what we know, the something cannot be otherwise. Whatever provides this element of necessity serves as an explanation. The great power of the deductive model is the clear and simple way in which necessity is accounted for.” Theoretically based explanations of real-world events are also more compelling than atheoretical accounts. When a seemingly unique event can be identi‹ed as an instance of a more general category that is part of some theory’s empirical domain (Rosenau and Durfee 2000: 3) or, to use Hempel’s (1965: 345) terminology, when its dynamic is placed under a covering law, a deeper understanding of an apparently singular event is achieved (Riker 1990: 168). For example, in chapter 6, I explain one aspect of the July Crisis ‹rst by identifying it as a speci‹c representation of a wider category of events (i.e., con›icts that escalate) and then by associating it with a speci‹c prediction of perfect deterrence theory (i.e., an equilibrium of an escalation model) and the conditions (i.e., the beliefs) that give rise to it. To summarize brie›y, one important goal of this book is to gain a deeper understanding of the outbreak of World War I by examining it from the vantage point of a well-articulated theory of con›ict initiation, escalation, and resolution—perfect deterrence theory. Explanations of the Great War are plentiful, but the explanation I construct here will be more coherent, less ad hoc, less opaque, and more forceful than other attempts to clarify the events of late July and early August 1914.
1.2. Bringing Data to Bear on Theory In addition to constructing a theoretically informed explanation of the Great War, this book subjects perfect deterrence theory to more intense empirical scrutiny.9 A number of good reasons exist for doing so. First, empirical examination provides an opportunity to illustrate the the9. For large-n studies, see Quackenbush 2003, 2010; Senese and Quackenbush 2003.
8
the games of july
ory in application or at work. Most theories are complexes of assumptions, concepts, and hypotheses. All things being equal, parsimonious theories are to be preferred to those that are not, but even parsimonious theories composed of a limited number of variables may be dif‹cult to appreciate or comprehend. At minimum, therefore, empirical exposition can enhance the understanding of a theory’s purpose and its underlying logic. The application of a theory to a speci‹c historical event can also serve as a “plausibility probe” of a theory’s explanatory and predictive (or, in this case, postdictive) power (Eckstein 1992: 147–52). A theory’s ability to explain a complex event and to anticipate its dynamic is a clear indicator of the theory’s continuing viability. A theory that survives its ‹rst plausibility probe should have greater standing than a theory that does not; moreover, a passing grade signals that further investment in the theory—in the form of elaboration, re‹nement, extension, or systematic testing—is warranted. Of course, this statement does not necessarily imply that a theory’s failure to account for all conceivable eventualities is a sign that it is degenerative in the sense of Lakatos (1970). The discovery of a lack of ‹t between theory and evidence may in fact be another reason for examining a theoretical structure in the context of a single event—if that term can be used to describe a case as complex as the July Crisis. The failure of a theory’s structure to conform to a particular feature of the real world provides an opportunity to adjust the theory to increase its scope. However, such adjustments should never seek simply to account for an empirical anomaly, for that kind of modi‹cation would indeed be evidence of degeneration (Vasquez 1997). Rather, one wholly defensible purpose of theory modi‹cation is to broaden the class of events to which it can be fruitfully applied. For example, when originally speci‹ed, perfect deterrence theory applied only to deterrence situations in which there were two principal players. The reason is straightforward: none of its component models were n-person games. In consequence, the theory’s implications for those extended deterrence situations in which the choices of a third actor, a pawn or a protégé, partly determine an outcome were circumscribed. To overcome this limitation, another game form—the Tripartite Crisis Game—was developed (Zagare and Kilgour 2003) to explore the impact of alignment patterns on the success or failure of extended deterrence relationships. Since the assumptions underlying the analysis of this model are derived from, and are consistent with, those initially used to cast the theory, the Tri-
Purpose and Method
9
partite Crisis Game can be considered not only an extension of, but also a critical component of, perfect deterrence theory. Indeed, it has proven indispensable to this analysis of the July Crisis. In part or in whole, in one variant or in another, it is called on to help answer three of the critical questions about the Great War. Put in a slightly different way, well-articulated theories always have clear boundary conditions. The assumptions on which a theory rests can be considered limiting conditions that de‹ne the set of circumstances to which the theory applies. When empirical scrutiny reveals these conditions to be too restrictive, their relaxation can be a proper analytic reaction. Indeed, it is the dominant scienti‹c response (Bates et al. 1998: 16).
1.3. A Possible Objection and a Rejoinder Not all analysts would agree that game-theoretic models and the theories derived from them should be used to develop an explanation of a real-world event, even if that event can be considered a member of a wider category.10 Morrow (1997: 29), for example, asserts that “individual case studies are poor tests of rational choice models.” He argues that “ex post reconstructions of historical events use information that the actors could not have at the time, subtly in›uencing us away from the strategic problems they faced[and] cannot explain the case precisely because the game cannot fail to ‹t the speci‹ed facts of the case.”11 Morrow’s position is extreme. It overlooks the fact that “critical case studies” can and do serve important theoretical purposes (Eckstein 1975). Still, Morrow is more correct than not in arguing that under most circumstances, even a theoretically informed single case study would fall short of a comprehensive test of a game-theoretic model or, for that matter, any theoretical construct. The reason is not dif‹cult to fathom: one cannot generalize from an n of one. And the July Crisis is clearly a “one.” Generalization, however, is not the motivation for this examination of the July Crisis. Explanation and theoretical illustration take pride of place, so Morrow’s observation is of little moment here. But even if it were pertinent, it does not necessarily follow that ex post reconstructions of real-world interactions 10. King, Keohane, and Verba (1994: 12) make the case for developing causal explanations of “even highly unusual events that do not fall into a large class of events.” 11. See also Elster 2000.
10
the games of july
that rely solely on information that was available to the actors at the time they made their decisions cannot be developed, as Bueno de Mesquita’s (1998) analysis of the end of the Cold War so well demonstrates. Morrow is also correct in asserting that some game model can always be molded to ‹t the facts of almost any individual event. But here too his objection misses a critical point. Closeness of ‹t is not the only criterion by which explanatory models are judged. Like any historical description, a game-theoretic explanation of an individual case can also be evaluated in terms of the plausibility of its assumptions about the motives of the players and other critical components of the model (i.e., the choices available to the players, the set of outcomes, and so on). For example, in an analysis of the Geneva Conference of 1954, I used a game-theoretic model of strategic deception to reconcile the well-known but unexplained discrepancy between the public and the private pronouncements of U.S. decision makers both before and during the negotiations that brought the Franco-Vietminh war to a close (Zagare 1979). Using the same model, Thakur (1982) developed six seemingly plausible alternative explanations of the diplomatic process at Geneva. But these competing descriptions of the conference were either empirically dubious or logically de‹cient and, hence, could be dismissed (Zagare 1982).
1.4. Blueprint This discussion brings up yet another criterion by which theories and explanations based on them can be judged: their competitive advantage or lack thereof. This criterion, which is generally associated with the philosopher of science Imre Lakatos (1970), is especially pertinent to any explanation of the July Crisis, simply because they are so plentiful. With so many to choose from, it is easy to go wrong. It is also easy to blend many of these explanations together to include every variable that can plausibly be associated with the onset of the Great War. Simpli‹cation is clearly in order. Inferior explanations must be cast aside. Stronger explanations, however, must be closely scrutinized and evaluated. Are they internally consistent? Are they empirically supported? Are they falsi‹able? And are they superior to their competitors? Chapter 2 reviews and evaluates extant explanations of major power wars in general, and of the Great War in particular. After quickly eliminating a number of candidate explanations, either because they are monocausal or because they are nonfalsi‹able, attention is turned to the leading contender: realist, or balance-of-power, theory and its corollary, classical deterrence theory.
Purpose and Method
11
The underlying assumptions, the theoretical characteristics, the major propositions, and the policy implications of the realist research program are examined and found wanting. Chapter 3 lays out the general contours and empirical expectations of perfect deterrence theory, which should be viewed as a theoretical alternative to the standard explanatory paradigm. Chapters 4–7 bring data to bear on theory. Component models of perfect deterrence theory are used to explain several critical determinants of the chain of events that led to the Great War. In chapter 4, the decision of Germany’s ‹rst chancellor, Otto von Bismarck, to enter into a defensive alliance with Austria in 1879 is examined in the context of the Tripartite Crisis Game. Understanding Bismarck’s motivation is essential for an appreciation of German foreign policy up to and including the July Crisis; it is also essential for comprehending the structural characteristics of the European state system just prior to the war. Chapter 5 explores both the Austro-Hungarian decision to seek Germany’s support, and the German decision unconditionally to provide that support, in the wake of the assassination at Sarajevo of Archduke Franz Ferdinand, the heir apparent to the Austro-Hungarian throne. The so-called blank check is found to have a natural explanation in the form of an equilibrium prediction of the Protégé-Defender Subgame of the Tripartite Crisis Game. Another model, the Asymmetric Escalation Game, is used in chapter 6 to explain the initial escalation of the July Crisis from a local war in the Balkans between Austria-Hungary and Serbia to a continental war that included all the great European powers with the singular exception of Great Britain. Because the ‹t between theory and fact is unusually close, this chapter exhibits many of the characteristics of what George and Bennett (2005) call process tracing. Since much of the action of the crisis took place before Great Britain entered the war, this chapter contains deeper description than any other. The description demonstrates a close association between the choices available to the players in 1914, their beliefs about each other’s intentions, and the intensi‹cation of the con›ict. Chapter 7 analyzes British foreign policy in the run-up to the Great War. Why did Britain’s foreign secretary fail to make clear a threat to support France in the event of a German attack, and precisely why did his policy fail? Again, the answer to this question is developed in the context of an equilibrium prediction of the Tripartite Crisis Game. The existence conditions associated with this prediction provide an answer to the ‹rst question; the theoretical characteristics of the equilibrium offer an answer to the second.
12
the games of july
Chapter 8 brings this analysis to a conclusion. A explanation for the outbreak of the Great War is summarized. The theoretical consequences of bringing data to bear on perfect deterrence theory are discussed. And the policy implications of this study, many of which are relevant to the world in which we currently live, are highlighted.
1.5. Scylla and Charybdis Under normal circumstances, the two tasks I have undertaken here—historical explanation and theory illustration and application—are fraught with dif‹culties. The dangers are magni‹ed when the goals are pursued simultaneously, for these twin objectives are sometimes at odds with one another, working at cross-purposes. When placed between a rock and a hard place, compromises are often required. These compromises, however, are likely to satisfy no one. Historical analysis, for example, is generally descriptively rich and long on details. But explanation demands simpli‹cation. Details that stand in the way of explanation frequently must be overlooked or ignored. In the analysis that follows, an abstract, formal, structure does most of the work of explanation. In consequence, deep description is frequently cast aside. Many historians will understandably be unhappy. But as the great American artist Georgia O’Keeffe once said in a completely different context, “Nothing is less real than realism. Details are confusing. It is only by selection, by elimination, by emphasis that we get at the real meaning of things.” It is also inevitable that some historians will be less than pleased with the connections I make between the events leading up to the Great War and the game models I use to analyze them. Among these connections, the assumptions I make about the preferences of the players have been the most problematic. And historians (and some political scientists) have offered little help here. For example, what goal was Germany pursuing in the run-up to the war? There is great controversy and a stunning lack of consensus among diplomatic historians about the answer to this question. In the end, judgments had to be made. And while I have made every attempt to justify these coding decisions, I appreciate that reasonable objections persist. Alas, there is no sure way to avoid the ambiguities in the historical record. I accept this reality as a certain limitation of my analysis. There may indeed be more than one plausible interpretation of the “facts.” Like historians, formal theoreticians are likely to ‹nd objectionable some
Purpose and Method
13
of what follows. The empirical application of a game-theoretic model to diplomatic history in general, and to the July Crisis in particular, is anything but straightforward. The primary problem involves bringing the theoretical concepts of an abstract model into an isomorphic relationship with a reality characterized by, among other things, nuance, ambiguity, equivocation, and duplicity. The real world is rarely as contained or as neat as any formal representation of it. Again, compromises must be made. Take, for instance, the issue of player aggregation. Purists (and I generally count myself among them) are likely to object not only to the fact that I treat states as players in a game, but also to the fact that in chapter 6 I consider two state-based coalitions as single players. After states have been admitted as players, there is a great temptation to reify them. I have tried to avoid that temptation by resting my conclusions about the policy objectives of Germany, Austria-Hungary, France, Russia, and Great Britain on appropriate primary and, when necessary, secondary sources rather than from some “objective” analysis of the national interest. In other words, rather than positing state goals, I have tried to uncover them from an examination of the contemporaneous diplomatic record. Still, Elster (2000: 693) argues that “rational choice explanations divorced from methodological individualism have a dubious value.” In principle, I agree. But strict adherence to principle all but eliminates the possibility of analysis, which is odd because Elster goes on to assert—wrongly, in my opinion—that “rational choice theory is the only theory in the social sciences capable of yielding sharp deductions and predictions.” The disaggregation of states, even to the level of a half dozen or so leading decision makers, destroys any semblance of tractability and with it any possibility of a satisfying explanation of all but the most trivial of circumstances. Readers must judge for themselves whether the violation of principle comes at too great a price. Some theorists, formal or otherwise, may also ‹nd fault with my decision to use generalized models to analyze certain events rather than to ‹ne-tune these models more precisely to adhere to the speci‹c circumstances I examine. I have felt uncomfortable with this approach, especially in chapter 6, where I specify the causal mechanism underlying an actual escalation spiral. I have nevertheless proceeded in order to maximize generality, but only when I have convinced myself that a more nuanced model would only yield the same conclusion. King, Keohane, and Verba (1998: 43) assert that “where possible, social science research should be both general and speci‹c: it should tell us something
14
the games of july
about classes of events as well as about speci‹c events at particular places. We want to be timeless and timebound at the same time.” It would be dif‹cult to state the objective of this book more succinctly.
1.6. Coda Although game theory has been a ‹xture of the security studies literature since the early 1950s, it has not been used until recently to study a complex of realworld events such as the one that unfolded during the middle of the summer of 1914. Accordingly, in this section, I provide an overview of those aspects of the mathematical theory of games that are relevant to an understanding of the explanation I develop, a brief (nontechnical) discussion of the intuition that lies beneath the explanation, and a sense of some of the limitations of using game-theoretic models to organize a case study. I begin with a few de‹nitions.12 The basic concept is that of a game itself. A game can be thought of as any situation in which the outcome depends on the choices of two or more actors—that is, when the choices of these actors (players) are interdependent. Players may be individuals or groups of individuals who act as coherent units. The players are assumed to be rational. Simply put, this means that they are purposeful, that is, that they have objectives and that they act to bring them about. It does not mean, however, that the players are necessarily intelligent in the sense that their objectives are wise, realistic, or even admirable. It also does not mean that a player will succeed. In game theory as in life, players are often misguided, shortsighted, imprudent, and unsuccessful. The lions of July were no exception. Games can be categorized in many ways. For my purposes here, the fundamental distinction concerns the rules that govern play. When there is no legal or other kind of constraint on the players’ ability to coordinate their actions, they are said to be playing a cooperative game. By contrast, a noncooperative game occurs whenever coordinated decision making is not possible. In the real world, coordinated action might be precluded for any number of reasons. For example, players might not be able to communicate with one another. Collusion might also be prohibited by statute, as would be the case in an economic market that was regulated by an antitrust law. Finally, a commitment problem 12. There are many excellent book-length expositions of the fundamentals of game theory. For the political scientist, there is none better than Morrow 1994. Condensed treatments can be found in Gibbons 1997; Zagare 2008.
Purpose and Method
15
will likely exist whenever “there be no Power erected, or not great enough for our security [such that] every man will and may lawfully rely on his own strength and art for caution against all other men” (Hobbes 1968 [1651]: 223–24). For what may be obvious reasons, noncooperative games are the rule and cooperative games the exception in international politics. Since the international system is a self-help system that lacks an overarching authority with the ability to force powerful and sovereign states to honor their commitments, any agreement that a state makes to cooperate with another will always be contingent on the state’s willingness to honor that agreement. Thus it should come as no surprise to learn that noncooperative game theory holds a particular attraction for theorists of international con›ict, as it does here. At the heart of the theory of noncooperative games is the concept of an equilibrium outcome (or strategy set). In game theory, an outcome is considered to be in equilibrium whenever no single player can induce a better outcome by selecting a different course of action, given that the choices of all the other players in the game do not change. At an equilibrium outcome, therefore, the strategy choice of each of the players is a best reply to the choices of all the other players in a game. An equilibrium outcome, however it is de‹ned, should be thought of a self-enforcing agreement among the players about how to play a game. The agreement is self-enforcing simply because it is in the interest of each player to carry it out. Whenever a noncooperative game is the only game in town, a selfenforcing agreement will clearly be the only meaningful type of agreement available to the players. Put in a slightly differently way, only equilibrium choices are consistent with (individually) rational behavior by all of the players in a game. When an outcome is not in equilibrium, at least one player can do better by switching to a different course of action. Thus, the assumption that the players in a game are rational leads naturally to the expectation that they will make choices that are associated with some equilibrium outcome. Game-theoretic explanations and predictions derive from this expectation. When players in a real-world game make choices that can plausibly be associated with an equilibrium outcome, an explanation is at hand. As Riker (1990: 175) has observed, “Equilibria are . . . identi‹ed consequences of decisions that are necessary and suf‹cient to bring them about. An explanation is . . . the assurance that an outcome must be the way it is because of antecedent conditions. This is precisely what an equilibrium provides.”
16
the games of july
Similarly, a game-theoretic prediction about future play presumes rational choice—that is, the assumption is that an equilibrium choice will be made by each of the players. Unfortunately, since a game will often have more than one equilibrium, this assumption does not always lead to a point prediction or to a fully satisfying explanation. The technical reason is that when multiple equilibria exist, the choices that lead to them may not always be interchangeable: even when all the players make choices associated with an equilibrium outcome, the resulting outcome may not be in equilibrium. This vexing problem is compounded by the fact that the equilibria of a noncooperative game may not always imply the same payoff to one or more of the players—that is, the equilibria may not be equivalent. Thus, in a game with multiple equilibria, it may be impossible to specify before the game is actually played out which course of action the players will take, or to explain after the game has concluded, why one rational strategic possibility rather than another came about. It is more than an understatement to suggest that the existence of multiple nonequivalent and noninterchangeable equilibria confounds both explanation and prediction in a game-theoretic model and, as Niou and Ordeshook (1999: 89) point out, in any theory of rational agency. For the most part, however, the games analyzed in this book almost always have equilibria that exist under unique (parameter) conditions. Thus, the explanations developed in this volume are, on their face, more compelling than explanations derived from models with multiple coexisting equilibria.13 Still, some problems remain. One dif‹culty is demonstrating that the conditions associated with the existence of a particular equilibrium, unique or otherwise, are satis‹ed empirically. This is not always an easy and straightforward task. Whenever this problem surfaces, a judgment call is necessary, and of course great care must be taken not to mold all-too-convenient facts. But this challenging problem is not con‹ned to game-theoretic studies. Indeed, it is endemic to the social science enterprise itself. Historians, for example, must deal with it all the time. Another theoretical predicament manifests itself whenever observed action choices are consistent with more than one equilibrium; when this happens, as it does in chapter 5, it is impossible to say for sure which equilibrium form actually came into play, although it may still be possible to speculate rig13. The problem is especially acute in a repeated game model. One of the so-called folk theorems of game theory states that in an in‹nitely repeated game, any combination of action choices that is individually rational can be part of some equilibrium. Thus, just about any behavioral pattern can be justi‹ed in a repeated game. There are no such models in this book.
Purpose and Method
17
orously. That an unambiguous conclusion may not always be possible is a fact of life I have come to accept. Even so, this limitation of game theory is tempered by the realization that even when there is more than one rational strategic possibility, some possibilities may in fact be ruled out—logically, empirically, or both. In the end, additional insights into real-world processes are obtained. The technical requirements (i.e., the de‹nition) of an equilibrium outcome in a noncooperative game turn largely on whether a game’s strategic structure is speci‹ed in strategic- or in extensive-form. In a strategic-form representation, the players are assumed to choose, simultaneously, a complete course of action (i.e., a strategy) before a game is played out.14 In the static environment of a strategic-form game, therefore, each player makes but a single choice. By contrast, in an extensive-form representation, the players are assumed to make choices, sequentially, at moves (or decision points) that are represented by nodes on a game tree.15 Thus, extensive-form games better capture a game’s ongoing dynamic. In a static (strategic-form) game, the standard equilibrium concept is due to Nash (1951). A strategy pair is a Nash equilibrium if no player could achieve a better outcome by switching, unilaterally, to another strategy. In a dynamic (extensive-form) game, where the players’ choices are sometimes contingent, the central equilibrium concept is subgame-perfect (Selten 1975). Nash equilibria exist in the dynamic context, but they may be based on incredible threats (i.e., on threats of irrational choice), whereas subgame-perfect equilibria require the players to plan to choose rationally at every node of the game tree. Nash and subgame-perfect equilibria are the accepted measures of rational behavior in games of complete information, in which each player is fully informed about the preferences of the other players. In games of incomplete information, in which, for example, at least one player is uncertain about the other’s preferences, rational choices are associated with Bayesian Nash equilibria (in static games) and with perfect Bayesian equilibria (in dynamic games).16 The ‹ne distinctions among the various equilibrium de‹nitions are less important, at least for the purposes of this book, than the intuition that lies beneath all of them. An equilibrium outcome—whether Nash, subgame-per14. Strategic-form games are sometimes referred to as normal- or matrix-form games. See ‹gure 2.1 for an example. 15. For this reason, the extensive-form is also termed the game-tree form of representation. See ‹gure 2.2 for an example. 16. The list is not exhaustive. For a precise de‹nition and a detailed discussion of the standard equilibrium forms mentioned in this paragraph, see Gibbons 1992.
18
the games of july
fect, Bayesian Nash, or perfect Bayesian—should be thought of as one of perhaps several rational strategic possibilities. In the context of any noncooperative game model, an explanation requires at minimum a plausible association of action choices with both an equilibrium and the conditions required to bring it about. All things being equal, the strength of the explanation is ampli‹ed as the plausibility of the association increases and as the number of rational strategic possibilities decreases. For this reason, I have gone to great lengths in this volume, but especially in chapters 4–7, to make as compelling a case as possible for the connection between the beliefs of the players, their action choices, and the existence conditions associated with (as it turns out) some perfect Bayesian equilibrium. Since my goals in this work mandate that these connections be established, I hope that an intuitive sense of the central role played by an equilibrium outcome in a noncooperative game model, along with the brief descriptions of some fundamental concepts found in the footnotes, will enable those readers either uninterested in technical details or with a low tolerance for formal de‹nitions to follow the argument. I also hope that game theorists and other formal modelers will understand the absence of proofs and symbolic argumentation. Of course, the mathematical particulars do exist, but since there is no apparent need to reproduce them here and I lack the space to do so, I simply note their source location whenever relevant.
If . . . an “equality of power . . . among the major powers” minimizes the likelihood of war, World War I should never have been fought. kenneth waltz
chapter two
Theories and Explanations
The earliest attempts to “explain” the Great War were virtually coterminous with its outbreak. In fact, some of these efforts preceded the hostilities, as many of the involved powers, but Germany in particular, tried to avoid being held responsible for the looming escalation of the crisis. Shortly after the ‹ghting started, each of the ‹ve major European powers released a colored book— white for Germany, red for Austria-Hungary, yellow for France, blue for Britain, and orange for Russia.1 The books were highly selective collections of of‹cial documents—some of them fraudulent, many of them intentionally misleading—designed to de›ect culpability for the war (Herwig 1987). The blame game had already begun. If one took these books seriously, World War I was truly a war that—to use David Lloyd George’s phrase—“nobody wanted.” When the war was over, the issue was settled, or so it seemed. Article 231 of the 1919 Treaty of Versailles, the so-called War Guilt Clause, held Germany and its allies responsible for “all the loss and damage to which the Allied and Associated Governments and their nationals have been subjected.” Under the threat of renewed hostilities, the German delegation at the Paris Peace Talks was forced to accept this provision. Nonetheless, the German government and its successors remained—quite justi‹ably—intent on repudiating it. A special bureau of the German Foreign Of‹ce (the Kriegsschuldreferat [War Guilt Section]) was even set up for this purpose. But the German government was not alone in its desire to revise this interpretation of the outbreak of the war. Professional historians, some of them underwritten by the German Foreign Of‹ce, soon got into the act. Revisionist 1. With the exception of the British Blue Book, all the colored books are available at the World War I Document Archive (http://www.lib.byu.edu/~rdh/wwi/). The Belgian Gray Book and the Serbian Blue Book are also available at this site.
19
20
the games of july
histories were written not only in Germany but also in Great Britain, France, Switzerland, and especially, the United States. These accounts sought a more balanced treatment than the judgment rendered by the victorious powers at Versailles. The American revisionist Sidney Bradshaw Fay (1966 [1928]) set the original standard for historical explanation in a two-volume work, The Origins of the War. Fay’s ‹rst volume examined the long-term or “underlying” causes of the war “before Sarajevo,” while the second volume discussed the shortterm or “immediate” causes of the war “after Sarajevo.” The long-term causes singled out by Fay included the system of alliances; the forces of militarism, nationalism, and economic imperialism; and the agitation of popular presses. Others (e.g., Joll 1984) have since added to this list. Nonetheless, it is clear that historical explanations that rely primarily on the identi‹cation of underlying causes will be found wanting. One drawback, as Fay himself (1966 [1928]: 1:32–33) explains, is that the underlying causes “are so complex and reach so far back into the past that any attempt to describe them adequately would involve nothing less than the writing of the whole diplomatic history of Europe since 1870, or rather from 1789; some questions go back to the age of Louis XIV, and even to that of Charlemagne.” The dif‹culty of establishing a causal link between this web of facts, events, and historical processes and the outbreak of the Great War should be apparent. But there are other problems as well. The innate time lag alone between a long-term cause and its purported effect renders satisfying explanation problematic (Kiser and Hechter 1991: 8). Worse still, those historians who have tried to explain the outbreak of the war in terms of its underlying causes have generally failed to specify the causal mechanism at work. Not surprisingly, post hoc, ergo propter hoc errors of inference are a common characteristic of this genre. That obvious de‹ciency is compounded by ›at-out assertions of a causal connection when no data exist to measure the presence or relative strength of pertinent variables. For example, some scholars have claimed that nationalist sentiments run amuck contributed to the outbreak of the war. But as Hamilton and Herwig (2003b: 22) point out, “With no serious public opinion surveys prior to the mid-twentieth century, we have no satisfactory indication of the prevalence, intensity, or import of any ‘mass’ attitude.” Given the inherent dif‹culty of establishing a clear causal linkage between any of the long-term factors and the outbreak of the Great War, it is no small wonder that many historical explanations focus on the con›ict’s immediate causes, but especially on issues of intentionality and responsibility. In this regard, the analytic method of choice has been the historical narrative.
Theories and Explanations
21
Historical narratives (including Fay’s second volume) have many advantages: they provide nuance, context, and detail; they also provide temporality and may even provide the basis for theory (Büthe 2002). But since historical narratives are unable to supply the element of necessity that is essential for establishing causality, they do not provide compelling explanations (D. Fischer 1970: 104). For that, a theory is required (Hanson 1958: 90; Trachtenberg 2006: 28). So, to construct a satisfying explanation of the Great War, we must start with a well-articulated theory. Ideally, the theory should be conceptually rich; it should be suggestive of multiple causal mechanisms, some of them distinct (Gerring 2005: 178–79). Most important, the theory that is used to construct an explanation of the events of July and early August 1914 should also be empirically plausible and logically consistent (King, Keohane, and Verba 1994: 105–6).2 Unfortunately, however, the theory of war initiation rooted in “the oldest and most prominent theoretical paradigm in international relations” (Legro and Moravcsik 1999: 5) fails to meet the latter two criteria.3
2.1. The Realist Paradigm and Classical Deterrence Theory The realist paradigm has long dominated the study of interstate con›ict and war. The paradigm’s provenance is quite impressive. Early realists include the Greek historian Thucydides, the Chinese military strategist Sun Tzu, and the Hindu statesman Kautilya. Realists also claim as one of their own some of the most prominent political philosophers in the Western tradition: Italian political tactician Niccolò Machiavelli, English philosopher Thomas Hobbes, Prussian military theorist Carl von Clausewitz, and even German idealist G. W. F. Hegel, inter alia, are often singled out for their in›uence on contemporary realist theorizing. With roots that can be traced back to ancient China, India, and Greece, it is no surprise that the realist paradigm has evolved over time to re›ect changing world realities. From the advent of the modern state system in 1648 to the conclusion of World War II, the paradigm was understandably closely associated with the operation of the multipolar European state system and balance2. Walt (1999a) downplays the signi‹cance of logical consistency as a criterion for evaluating theories, claiming that both empirical validity and originality are more important. A number of (formal) theorists have taken issue with Walt’s reasoning (Bueno de Mesquita and Morrow 1999; Martin 1999; Niou and Ordeshook 1999; Powell 1999; Zagare 1999). See Walt 1999b for the counterresponse. 3. Liberal explanations are considered, albeit brie›y, in chapter 3. As Levy (2002a: 358) notes, World War I “run[s] counter” to “the liberal economic theory of peace.”
22
the games of july
of-power politics (Gulick 1955). But with the dropping of atomic bombs, ‹rst on Hiroshima and then on Nagasaki, the realist research program was forced to adjust. Brodie (1946) was among the ‹rst to notice that the postwar international system differed markedly from the system that had existed before Poland was invaded in 1939. For one, it was now decidedly bipolar, dominated by two (super) powers from the periphery of the European state system. For another, it was well on its way to becoming nuclear. It was simply a matter of time. Sooner rather than later, the cost of all-out war in the postwar international system would be exponentially higher than in the prewar era. It was in this context that classical (or rational) deterrence theory was born. But the break with the past was not as sharp as is sometimes asserted. Classical deterrence theory retains almost all of the de‹ning assumptions of its parent paradigm. It is best interpreted, therefore, as a special case analysis of balanceof-power theory. Both the general theory and its application to the post-1945 international system attempt to explain the conditions associated with the onset of interstate war or the absence thereof. In other words, classical deterrence theory, like balance-of-power theory, is at heart a realist theory of the causes of war (Powell 1985: 76). Indeed, classical deterrence theory is arguably the most signi‹cant variant of realist theory; among many policymakers and academics, classical deterrence theory constitutes the conventional wisdom about interstate war in the modern era. Nonetheless—and this idea may surprise some readers—“to a quite extraordinary degree,” many of classical deterrence theory’s most “important conclusions” are drawn from “a speci‹c interpretation of a single historical episode: the coming of the First World War during the July Crisis in 1914” (Trachtenberg 1990/91: 120). The deep roots of classical deterrence theory in the string of events that led to the Great War provide further justi‹cation for a reexamination of those events from the vantage point of a theoretical competitor—perfect deterrence theory—in the second part of this book. To recapitulate brie›y, classical deterrence theory is nothing more and nothing less than a straightforward extension of a general equilibrium model of interstate politics, balance-of-power theory, to the ostensibly new political and technological realities that came about after World War II. To evaluate one manifestation of this paradigm, therefore, is to evaluate the other, and vice versa. The two theoretical variants share not only the same axiomatic base and the same dependent variable but also the same logical and empirical ›aws, as the following summary discussion of classical deterrence theory should make clear.
Theories and Explanations
23
2.1.1. structural deterrence theory Classical deterrence theory can conveniently be divided into two distinct, yet compatible, formulations: structural (or neorealist) and decision-theoretic deterrence theory. Even though these strands in the literature focus on different units of analysis, the assumptions they make, the conclusions they reach, and the policy prescriptions they offer are essentially the same.4 Their explanations for the outbreak of war in 1914 are also, in general, identical. For structural deterrence theorists, the international system constitutes the principal unit of analysis. The system itself is anarchic: there is no overarching authority to enforce treaties or other kinds of agreements. This “self-help” system is composed of undifferentiated units (i.e., states)5 that are rational and egotistical. The units are driven either by their nature to maximize power (Morgenthau 1973) or by their environment to maximize security (Waltz 1979).6 Structural deterrence theorists, by de‹nition, hold that the keys to international stability lie in the distribution of power in the international system and the absolute cost of war. More speci‹cally, structural deterrence theorists locate the root cause of interstate con›ict in asymmetric power distributions. By contrast, they contend that war is much less likely when the system is “in balance.” Because they contend that balances are easier to obtain when there are only two great powers rather than when there are three or more, the most prominent structural deterrence theorists hold that bipolar systems are, in general, more stable than multipolar systems (Gaddis 1986: 105–10; Mearsheimer 1990: 14; Waltz 1964: 882–86). Since the Great War was in fact waged under parity conditions, balance-of4. For a more complete description of the axioms, tenets, and de‹ciencies of classical deterrence theory, see Zagare 1996. See also Zagare 2004, which forms the basis of the remainder of this and the next chapter. 5. Legro and Moravcsik (1999: 13) argue that the assumption of undifferentiated actors with “‹xed and uniformly con›ictual” preferences distinguishes realism—and, by extension, classical deterrence theory—from other paradigms. Waltz (1979) clearly assumes that states are undifferentiated (or like) units. But as Wohlforth (2000: 183) points out, Waltz also asserts that “the aims of states may be endlessly varied.” Some ambiguity (if not a contradiction) surely exists here. Waltz (1979: 105) readily admits that states “differ vastly in their capabilities.” In what real sense, then, can the units be undifferentiated if their critical preferences are also assumed to vary? How can all other states be potential threats, as Mearsheimer (1990: 12) asserts, if only some states “think and act in terms of interests de‹ned as power” (Morgenthau 1973: 5)? And what are we to make of Waltz’s (1993: 47) comment that “our conviction that the United States was the status quo and the Soviet Union the interventionist power distorted our view of reality” if states are not uniformly motivated? Walt (1999a: 17) matter-of-factly observes that Waltz’s theory contains contradictions. All of which helps to explain why Christensen and Snyder (1990: 138) ‹nd that “any foreign policy and its opposite can sometimes be deduced from Waltz’s theory.” 6. Mearsheimer’s (2001: 22) “offensive realism” is a hybrid of these two views. He assumes that the structure of the international system motivates states to maximize power.
24
the games of july
power theory cannot explain the war’s occurrence. As Copeland (2000: 12) observes, the standard version of the theory “cannot explain how multipolar systems with tight alliances against the potential hegemon, such as that existed before 1914, can still fall into major war.” Or as Waltz (1993: 77) puts it, “If Mearsheimer [(1990)] is right in believing that an ‘equality of power . . . among the major powers’ minimizes the likelihood of war, World War I should never have been fought. The opposing alliances were roughly equal in military strength, and their principal members understood this.”7 Why, then, did the Great War break out? For Waltz and a number of other structural deterrence theorists, the answer lies in the absolute costs of warfare: in 1914, these costs were judged too low to deter aggression. In consequence, deterrence simply failed. By contrast, after 1945, not only was the interstate system in approximate balance, but the high costs of (nuclear) war were obvious to all. War, therefore, was now rendered unthinkable (i.e., irrational), which explains why some structural deterrence theorists contend that the probability of a premeditated (or rational) war between nuclear equals is virtually zero (e.g., Intriligator and Brito 1981: 256; Waltz 1990: 740). More generally, structural deterrence theorists see a strict monotonic relationship between the cost and the probability of war. The straightforward connection between the cost of warfare and the likelihood of deterrence success lies behind Waltz’s (1993: 77) characterization of Norman Angell’s (1910) wellknown argument that war was impossible in 1914 because its economic costs were prohibitively high as “not wrong, but merely premature.” Mearsheimer (1990: 19) is less subtle: “The more horrible the prospect of war, the less likely it is to occur.”8 If the high cost of war in the nuclear age all but rules it out, what, then, is the gravest threat to peace? For structural deterrence theorists, it is an accident or a mishap. In other words, rational leaders will almost certainly choose to avoid all-out war, although “irrational” leaders (of rogue states) may choose otherwise. In addition, a war may come about because of miscalculation, misperception, the loss of operational control, or some other kind of strategic mistake or human error. Given the connection many classical deterrence theorists draw between the 7. Mearsheimer (2001: 216) implicitly accepts this judgment when he observes that “German behavior before World War I points to an anomaly for offensive realism. . . . Germany had an excellent opportunity to gain hegemony in Europe in the summer of 1905 [during the ‹rst Moroccan Crisis]. Yet Germany did not seriously consider going to war in 1905 but instead waited until 1914, when Russia had recovered from its defeat and the United Kingdom had joined forces with France and Russia.” 8. See also Van Evera 1999: 30.
Theories and Explanations
25
cost of war and its probability, it is not dif‹cult to understand why they argue that quantitative arms races help to prevent war (additional weapons, they hold, increase the cost of war);9 why some contend that qualitative arms races, such as the dreadnought race between Great Britain and Germany just prior to the Great War, and some defensive weapons are destabilizing (because certain types of weapons will reduce war costs for one or both sides);10 and why others favor managed nuclear proliferation (again, because nuclear weapons make war more costly).11 Although structural deterrence theory is consistent with the absence of a superpower con›ict during certain periods of the Cold War, it is nonetheless inconsistent with other pertinent empirical realities. As Jervis (1985: 6) notes, “Many events present unexpected puzzles for standard deterrence theory.” It is dif‹cult to reconcile structural deterrence theory with the fact that most major-power wars, including the Great War, have been waged under parity conditions—unless, of course, one is willing to assert that the cost of con›ict was simply too low before 1945 to completely rule out “rational” wars. This claim is very dif‹cult to substantiate, though some scholars try. According to Levy (1983: 88–91), more than two million lives were lost during the Thirty Years’ War; well over one million people died as a consequence of the War of Spanish Succession; and almost two million people were killed during the Napoleonic Wars. The key decision makers surely must have recognized the potential for disaster in 1914. To claim that they did not, one must also argue, as does Van Evera (1999: 194), that somehow, someway, they also managed to overlook “the lessons of the American Civil War, the Russo-Turkish War of 1877–78, the Boer War, and the Russo-Japanese War, which had revealed the power of the new defensive technologies.” Structural deterrence theory is also inconsistent with the observation that power imbalances are poor predictors of interstate con›ict.12 More speci‹cally, unless they make ad hoc arguments that contravene the theory’s axiomatic base, structural deterrence theorists are hard put to explain the absence of war before the Soviet Union achieved “essential equivalence” with the United States during the 1970s. As Waltz (1993: 47; 2000: 13) suggests, to explain the absence of a U.S.-Soviet war up to the advent of nuclear parity by claiming that the United 9. See, e.g., Gray 1974; Intriligator and Brito 1984. 10. See, inter alia, Jervis 1978; Scoville 1981; Van Evera 1984. For a contrary view, see Quackenbush 2006a. 11. The classical deterrence theorists who argue for the selective proliferation of nuclear weapons include Mearsheimer (1990), Waltz (1981), and Van Evera (1990/91). 12. For the relevant citations, see Zagare and Kilgour 2000: 24–26.
26
the games of july
States was either a status quo power or a self-deterred democracy unwilling to violate moral precepts contradicts the assumption that egotistical, rational, and undifferentiated units populate the interstate system. Structural deterrence theorists also have dif‹culty explaining why the contentious relationship of the Soviet Union and China never escalated into an all-out con›ict, why the United States has not attempted to invade Canada since 1812, or why states generally fail to jump through “windows of opportunity” (Lebow 1984). In the speci‹c case of World War I, Waltz’s assertion that major power war was still seen as pro‹table in 1914 is dif‹cult to accept. To be sure, some military leaders anticipated that the war would be over by Christmas and that it would therefore be relatively inexpensive (Farrar 1973). But not those who mattered most. For example, Helmuth von Moltke (the younger), who was chief of the German General Staff when war broke out in 1914, was under no illusions about the prospect of a short war. In a secret letter to German chancellor Theobald von Bethmann Hollweg written on the eve of the con›ict, Moltke warned that a major power war “will destroy civilization in almost all of Europe for decades to come” (Förster 1999: 366).13 Moltke was not the only German general to conclude that the next war in Europe would be a drawn-out war of attrition and therefore excessively costly. According to Förster (1999), Moltke’s celebrated uncle, Helmuth von Moltke (the elder), who became chief of the Prussian General Staff in 1857, came to this realization shortly after his forces routed the French army at Sedan in 1870. Even the elder Moltke’s most famous successor, General Alfred von Schlieffen, recognized that nothing “like a rapid victory was possible in the vast eastern European theater” (Förster 1999: 356). In consequence, he devised his famous plan designed to avoid a lengthy war. But since Schlieffen also did not believe that a knockout blow of France was likely, his plan “made no sense” (361).14 In 1914, there were other reasons why it should have been obvious that the cost of the next war would be prohibitive. And it was. After all, all of the major 13. Still, Moltke pressed the political leaders for war. As Förster (1999: 367) notes, “It is hard to explain why.” (See also Herwig 2002.) It is not, however, hard to explain why Kaiser Wilhelm II and Bethmann Hollweg resisted Moltke’s advice. In addition to the anticipated costs of a long war, the members of the German political leadership group were “petri‹ed by the domestic consequences of even a victorious war” (Copeland 2000: 76). As it turned out, their fears were justi‹ed. 14. There is great controversy among historians about the proper interpretation of the Schlieffen Plan. Zuber (2002a: 46), for example, concludes that “there never was a Schlieffen Plan.” According to Zuber (2004: 3), the memorandum that Schlieffen wrote in 1905 was never intended to be implemented. Rather it was a bureaucratic document written solely to help make the case “for a massive increase in the German army,” which Schlieffen thought too small to take on the combined forces of France and Russia. (See also Zuber 1999, 2002b.) Zuber’s thesis has of course stirred debate. See, inter alia, Foley 2003, 2005a, b; Holmes 2001, 2002, 2009; Mombauer 2005 (which concludes that the debate is “of very little consequence to what we know about the origins of the First World War” [859]). For a succinct summary and additional citations, see Mombauer 2005; Lieber 2007.
Theories and Explanations
27
powers were either fully armed or in the process of becoming so. In the years leading up to the war, the British and the Germans were involved in a costly naval competition, the French and the Germans were engaged in a conscription race, and the Russians had embarked on an ambitious plan of military expansion. Thus, “there was a feeling in all the governments that the war would be big and that it was too risky an activity to engage in” (Kahn 1969: 358). In addition, during the crisis, many leading policymakers acted as if they understood that the consequences of a European con›ict would be dire. As the crisis reached its apex, British, German, and Russian leaders worked feverishly to avoid disaster. More than once, the British foreign secretary, Sir Edward Grey, offered great-power mediation as a mechanism for con›ict resolution; the German emperor, Wilhelm II, proposed a limited Austrian military operation against Serbia to appease Russia and cap the con›ict; Bethmann Hollweg forcefully, albeit belatedly, urged restraint on Austria to preclude escalation; and to the very end, the Russian czar, Nicholas II, stubbornly resisted full mobilization of his forces because he recognized that this act implied an all-out war. If war was so cheap, why were there so few takers? Other signs also indicate that key decision makers recognized that the stakes were inordinately high in 1914. According to Albertini (1952: 2:670), Austria’s foreign minister, Count Leopold Berchtold, resisted a full mobilization of the dual monarchy’s army because it would “cost millions.” Eventually, Berchtold would accede to pressure from the chief of Austria-Hungary’s General Staff, General Franz Conrad von Hötzendorf, but only after being told that “the Monarchy is at stake.” Finally, there is also ample evidence that the pressure of the crisis had a deleterious effect on decision makers in Berlin, Vienna, and elsewhere. For example, in the ‹nal days, the kaiser judged Bethmann Hollweg to be both physically and psychologically exhausted. Apparently so was Count Friedrich von Pourtalès, the German ambassador to Russia. Immediately after hand-delivering Germany’s declaration of war to Serge Sazonov, Russia’s minister of foreign affairs, Pourtalès broke down and burst into tears. Similarly, the Austro-Hungarian leadership group “was displaying all the signs of behavior under stress” (Jannen 1983: 60). Grey clearly was not the only key ‹gure who feared that the lamps were about to go off in Europe. Among those who mattered most in 1914, few seemed to believe that the war would be waged “on the cheap.”15 15. It is more than odd that Van Evera (1999: 32) offers an optimistic comment of Grey to the House of Commons in early August as evidence that war costs were seen to be low in 1914, but then, buried at the end of a lengthy footnote, points out that “Grey’s private view of the coming war was gloomier.” Indeed, it was much gloomier. Grey’s (1925: 1:303) conviction was that the looming war “would be a catastrophe for which previous wars afforded no precedent.”
28
the games of july
In this light, Waltz’s explanation of World War I rings hollow. But then, how can structural deterrence theory—or, for that matter, balance-of-power theory—be reconciled with the facts? It is clear that equilibrium theorists cannot point, as they often do, to a power imbalance, especially in light of the military stalemate that quickly evolved on the battle‹eld shortly after the ‹rst Battle of the Marne and of other objective indicators that a balance of power existed in Europe in 1914 (see, e.g., Kugler and Domke 1986; Organski and Kugler 1980). Within the con‹nes of structural deterrence theory, there is only one possible response. World War I was brought about because of a “serious miscalculation” (Mearsheimer 1990: 24). It was, in other words, just a big mistake (MacMillan 2003: 19). With this intellectual sleight of hand, structural realism morphs into the theory of “inadvertent war.”16 One version of this clearly ad hoc explanation is that World War I was caused by blundering decision makers (Brodie 1973: 11–28) who acted “irrationally,” either because they misperceived the stark realities of the international system (Van Evera 1999)—which were, in principle, knowable to all (Morgenthau 1973)—or because they suffered from some other kind of cognitive failure or psychological de‹ciency (Lebow 1981). Another version suggests that events had spiraled out of control, again irrationally, either because of the overly rigid system of alliances that existed in 1914 (Christensen and Snyder 1990) or because of the in›exibilities associated with the military plans of the great powers in general and of Germany in particular (Kissinger 1994: chap. 8). The theory of inadvertent war is really quite ingenious. It allows balanceof-power theorists, structural deterrence theorists, and, as shall be seen, decision-theoretic deterrence theorists to have their cake and eat it too. Periods of peace (e.g., the Cold War) that occur under parity are easily explained and viewed as con‹rming evidence. And those wars that break out when power is balanced are accounted for in a way that completely eliminates the possibility of paradigmatic falsi‹cation. This is of course a serious epistemological short16. Barbara Tuchman’s (1962) popular history of the ‹rst few months of the war, The Guns of August, is perhaps the most well-known rendition of the inadvertent war thesis. Tuchman argues forcefully, and to some readers convincingly, that at the time of the crisis, forces largely outside of human control, but especially Germany’s rigid war plans, caused the July Crisis to spiral out of control. In her view, then, the Great War broke out even though no one actually wanted it. One clear implication of her analysis is that individual and/or state preferences did not matter, that the war’s root cause cannot be ascribed to rational strategic decision making. A number of similar explanations exist for the outbreak of the war. Collectively, they constitute the inadvertent war school. (For further discussion and additional references, see Trachtenberg 1990/91: 120–21; 1991: 47–49.) The arguments of this school should be contrasted with those of historians such as Hamilton and Herwig (2003b: 40), who contend that “the leaders of the various governments in 1914 assessed their options, weighed the choices, and, fully aware of the most likely consequences, deliberately chose involvement” and also with the (rational choice) explanation developed in the remainder of this book.
Theories and Explanations
29
coming (Popper 1968), strongly suggesting that a theoretically sustainable explanation of the onset of major power war in early August 1914 will have to be developed outside of the structural realist research program.
2.1.2. decision-theoretic deterrence theory Decision-theoretic deterrence theory constitutes the second major variant of classical deterrence theory.17 Decision-theoretic deterrence theorists, who focus on the interplay of outcomes, preferences, and rational choices, begin where structural deterrence theorists leave off. In developing either formal or informal rational choice models based on the payoff structure of the game of Chicken (see ‹gure 2.1), early decision-theoretic deterrence theorists such as Schelling (1960, 1966), Ellsberg (1959), and Kahn (1962, 1969), or later theorists including Powell (1987) and Nalebuff (1986), fully embrace the central conclusion of structural deterrence theory: when its cost is exorbitant, major power war is “irrational.” In Chicken, war (or Con›ict) is the worst possible outcome for both players. In consequence, Con›ict can never be consistent with rational contingent decision making. This important strategic property helps to explain why most decision-theoretic deterrence theorists, like the majority of structural deterrence theorists, have come to conclude that World War I was inadvertent (or accidental).18 The reason why a mutually worst outcome is unquestionably irrational is that it can never be part of a (pure strategy)19 Nash equilibrium outcome in 17. O. Young (1975) calls this most in›uential approach to deterrence “manipulative bargaining theory.” Rapoport (1968) pejoratively refers to decision-theoretic deterrence theorists as “neo-Clausewitzians.” Danilovic (2002) labels the genre “commitment theory.” For expositional purposes, I here use the term decision theory as an umbrella category that includes expected utility, game-theoretic, and certain psychological choice models that share the modal assumptions outlined here. Schelling’s work, for example, draws on all three genres, as does Jervis’s. 18. In the strategic-form representation given by ‹gure 2.1, there are two players, States A and B, each with two broad strategic choices: either to cooperate (C) by supporting the Status Quo or to defect (D) from cooperation by trying to overturn it. These choices give rise to four possible outcomes. Each cell of the matrix also contains an ordered pair that indicates the preference ranking of that outcome to State A and State B, respectively. Each player’s highest-ranked outcome is indicated by a rank of 4, the next-most-preferred by 3, and so on. The outcome called Con›ict is the lowest-ranked outcome (i.e., 1) of both players. In the story that gives this game its name, two teenage drivers are on a collision course. In addition, when Chicken is played sequentially, the player with the ‹rst move will win. Thus, both players in this game have an incentive to seize the initiative by committing to a hard-line (i.e., D) strategy. For an extended game-theoretic analysis and the story, see Fink, Gates, and Humes 1998. For additional applications of Chicken to deterrence, see, inter alia, Brams 1975, 1985; Jervis 1979; Nicholson 1989; Russell 1959; G. Snyder 1971. 19. A pure strategy involves the certain selection of a particular course of action. A mixed strategy is composed of a probability distribution over the set of a player’s pure strategies. Con›ict is part of a mixed-strategy Nash equilibrium in Chicken. O’Neill (1992: 471–72), however, argues persuasively that this particular equilibrium fails as a normative device: it prescribes behavior that is “just the opposite of what one would expect.” Under the mixed-strategy equilibrium in Chicken, the worse the Con›ict outcome is for one player, the more likely it is that the other player will concede. See also n. 32.
30
the games of july
State B Cooperate (C)
Defect (D)
Status Quo
B Wins
(3,3)
(2,4)*
A Wins
Conflict
(4,2)*
(1,1)
Cooperate (C)
State A Defect (D)
Key: (x,y) = payoff to State A, payoff to State B 4 = best; 3 = next-best; 2 = next-worst; 1 = worst * = Nash equilibrium
Fig. 2.1. Chicken
any strategic-form game with strict preference rankings over outcomes.20 And, as discussed previously, only equilibrium outcomes are consistent with rational choices by all of the players in a game.21 Thus, by assuming that Con›ict is the worst outcome for both players, decision-theoretic deterrence theorists perforce presume war and the actors who engage in it to be irrational. In so doing, they take as virtually axiomatic a critical deduction of structural deterrence theory. By presuming that war is always irrational, decision-theoretic deterrence theorists presuppose the world envisioned by structural deterrence theorists. Hence, decision-theoretic deterrence theory can be interpreted as a micro- (or 20. Recall that a strategy pair is a Nash equilibrium if no player can achieve a better outcome by switching, unilaterally, to another strategy. Con›ict can never be an equilibrium outcome in Chicken because the strategies associated with it are not best replies to one another. In other words, both players can always do better given that the other defects (i.e., chooses D). 21. An equilibrium can also be thought of as a state of decision-making stasis: at an equilibrium, no single actor has an incentive to do anything other than what it is doing at the moment. Thus, once an equilibrium is settled on, it is likely to persist, at least until something underlying the game’s de‹ning characteristics changes. For example, when historian William J. Newman (1968) describes the bargaining process among Germany, France, and Great Britain that led to the Locarno Pact in 1925, he is in essence describing the creative process the end result of which was an equilibrium. Of course, as Newman also chronicles, that equilibrium did not survive the German chancellorship of Adolf Hitler and the other processes and events of the 1930s.
Theories and Explanations
31
unit-) level extension of structural deterrence theory, in effect mapping out what optimal strategic behavior would be like in the world envisioned by structural deterrence theorists. This means that the conclusions of these unitlevel theorists also have important implications for the empirical accuracy and the logical consistency of structural deterrence theory. The descriptions and prescriptions of decision-theoretic deterrence theory are well known and will not be rehearsed in detail here.22 Suf‹ce it to say that, inter alia, statesmen have been counseled to seize the initiative by making an “irrevocable commitment” to a hard-line strategy, to avoid defeat by “linking” one issue to another, to make an opponent’s concession more likely by making con›ict more costly,23 or even to feign “irrationality” to force an opponent to concede during a crisis. While provocative, these stratagems are of dubious empirical validity. In a recent review of the deterrence literature, Huth (1999: 74) ‹nds that “early arguments about the strategic advantages of the manipulation of risk and commitment strategies have not been fully supported by empirical research” (see also Danilovic 2001, 2002). Making the same point, Betts (1987: 30) observes that “the view that apparent recklessness and irrevocable commitment are more effective is usually more comfortable to pure strategists than to presidents.” Perhaps Jervis (1988: 80) puts it best, albeit in the context of a different historical period: “Although we often model superpower relations as a game of Chicken, in fact the United States and USSR have not behaved like reckless teenagers.” Like structural deterrence theorists, decision-theoretic (i.e., rational) deterrence theorists have great dif‹culty explaining the Great War without violating the paradigm’s “hard core” of protected assumptions (Lakatos 1970). Schelling (1966: 227), for example, features the Great War as his principal example of an “accidental or inadvertent war.” Sounding very much like Mearsheimer (1990), Kissinger (1994), and other equilibrium theorists, Schelling’s (1966: 221) explanation focuses not on calculation but on “the technology, the military organization, and the geography of Continental Europe in 1914.” There was, he writes, “no ‘‹nal’ decision.” In consequence, “no one can quite say just when the war started. There was a great starting of en22. G. Snyder (1972: 229–31) provides a comprehensive listing. 23. Both structural and decision-theoretic deterrence theorists, therefore, recommend policies that raise the cost of con›ict. Jervis (1972: 273) tacitly accepts this argument when he suggests that World War I would not have occurred had the key decision makers “known the costs of war.”
32
the games of july
gines, a clutching and gearing and releasing of brakes and gathering momentum until the machines were on a collision course” (emphasis added).24 In other words, the Great War was unintended. It was brought about by a simple mechanical failure: in this high-stakes game of Chicken, the brakes failed. And so did deterrence. Schelling is not the only decision-theoretic deterrence theorist to draw on the Chicken analogy to explain the Great War. Kahn (1969: 358), for example, who is notorious for using the dynamics of this game to gain insight into strategic matters,25 begins his explanation of World War I by noting the widely held view in Europe that the looming con›ict would be excessively costly. Of course, this perception led logically to the belief that “one side or the other would presumably withdraw before things got out of hand.” Why, then, did the war that “neither side really wanted” occur? According to Kahn (365), “Some crucial miscalculations were made.”26 Similarly, Jervis (1976: 94) argues that the incentive of each side to seize the initiative27 was an “immediate cause of the outbreak of war in 1914. Each of the continental powers believed that the side that struck ‹rst would gain a major military advantage. Since to wait for the other side to clarify its intentions could mean defeat, even a country that preferred the status quo to war would feel great pressures to attack.”28 To his credit, Jervis (1976: 95) recognized that his explanation of World War I was inconsistent with the expectations of classical deterrence theory. Neither the high costs of war nor what he considered to be credible threats to resist prevented the con›ict.29 Inexplicably, however, instead of rejecting the deterrence paradigm, he simply turned to another, the so-called spiral model, to explain the dynamics of this most critical case.30 Of course, Jervis was well aware 24. See also G. Snyder 1984: 482. 25. See, e.g., Rapoport 1965, which is sarcastically titled “Chicken à la Kahn.” 26. Kahn, however, is onto something here. As explained in detail in chapter 6, throughout the crisis, the Austrians believed that German support would deter Russia from intervening on Serbia’s behalf, and the Russians hoped that their July 29 partial mobilization would force the Austrians to back off. Both Austrian and Russian decision makers were mistaken. 27. When Chicken is analyzed informally, the incentive is implicit. Jervis (1979: 292, 301–2) claims that Chicken provides the basis of a “reasonable de‹nition of deterrence theory.” In fact, he observed that “only by deducing the bargaining tactics that should be effective in the game of Chicken was [he] able to see that an actor had to take care to avoid the danger of symmetry.” 28. Brodie (1946: 235), perhaps the seminal classical deterrence theorist, explains the war in almost exactly the same way: “It was largely Germany’s eagerness to seize the initiative which made [the war] ‘unavoidable’ in 1914.” See also Van Evera 1999: chap. 3. 29. Classical deterrence theorists who believe otherwise claim either that the costs were too low (e.g., Waltz 1993) or that the (British) threat to intervene lacked credibility (e.g., Kagan 1995: 211). 30. For the shortcomings associated with a spiral model explanation of the outbreak of war in 1914, see Copeland 2000: 59–60.
Theories and Explanations
33
that “spiral and [classical] deterrence theories . . . contradict each other at every point.” But he was reluctant to fully accept (or reject) either one since “neither theory is con‹rmed all the time” (1976: 84). For reasons that are dif‹cult to fathom, he did not seem to consider the possibility that both theories were inadequate.31 Almost to a theorist, therefore, decision-theoretic deterrence theorists see the Great War as at least in some sense inadvertent, accidental, unintentional, or irrational. But then, those working within the paradigm’s con‹nes are forced by its logic to explain it in this way: when war is so costly that no one wants it, it can only come about when serious blunders are made. Two problems exist with this explanation. First, it violates the paradigm’s core assumption that actors are instrumentally and procedurally rational (Fearon 1995). Speci‹cally, if war was in fact the worst outcome for both sides, why did one side or the other not sue for peace just as soon as war broke out? Why did Russia and France resist? Why did Germany and Austria-Hungary not back off when it became clear that their opponents would not? There is simply no satisfactory answer that is consistent with both the rationality assumption and with the premise that the war was so costly that it was the leastpreferred outcome for all the major players in 1914.32 Second, and equally important, the explanation is empirically dubious. As Trachtenberg (1991: 99) points out, “The most remarkable thing about all these claims that support the conclusion about events moving ‘out of control’ in 1914 is how little basis in fact they actually have.” In Russia, for example, the czar delayed full mobilization until he became convinced that it was the correct course of action. Similarly, in Germany, the kaiser issued instructions, on more than one occasion, that countermanded those of the chief of the German General Staff, and the kaiser’s orders were obeyed. In other words, the political leaders were in charge in St. Petersburg, in Berlin, and elsewhere (Williamson 31. As chapter 3 discusses, perfect deterrence theory explains not only deterrence success but also escalation spirals. It therefore subsumes, in a logically consistent way, both classical deterrence theory and the spiral model. 32. One possible explanation is that since Con›ict is part of a mixed-strategy equilibrium pair in Chicken, the Great War came about because this equilibrium had come into play. But, as O’Neill (1992: 473) explains, the mixed-strategy equilibrium in Chicken prescribes behavior that is “unreasonable, just the opposite of what one would expect.” It is not, therefore, a plausible descriptor of rational behavior in July and early August 1914, even if one believes that the complexities of the crisis can be captured by a simple 2 × 2 game. Another possible explanation is that the war resulted from a coordination problem. But a coordination problem exists in Chicken only when the players select their strategies simultaneously. It is dif‹cult, however, to sustain the argument that the war came about because both sides made a war choice simultaneously. For example, the Russians had a clear opportunity to back down after the Austrians declared war and started to bombard Belgrade on the night of July 28–29, and both the Russians and French could have stepped aside after Germany issued ultimatums to them on July 31. Much the same can be said of the Germans after the August 4 British ultimatum to cease and desist with the invasion of Belgium, and of Britain on August 5 after Germany failed to comply.
34
the games of july
and May 2007). Strategic choices were indeed made. Of course, these choices were not divorced from military and political constraints. But the only reason to characterize the players as (instrumentally) “irrational” is that the choices they made were inconsistent with both the assumptions and the expectations of decision-theoretic deterrence theory. In this light, it should come as no great surprise to learn that like structural deterrence theorists, decision-theoretic deterrence theorists are also hard put to explain, without logical contradiction, the “long peace” that followed World War II (Gaddis 1986). As a result, decision-theoretic deterrence theorists are unable to explain either the failure of deterrence in 1914 or its apparent success after 1945. It is important to understand why. To this end, consider now the Rudimentary Asymmetric Deterrence Game depicted in ‹gure 2.2 in extensive form. In this, perhaps the simplest deterrence situation that one can imagine, there are two players: State A (Challenger) and State B (Defender), and only three outcomes: Status Quo, (State) A Wins, and Con›ict.33 The Rudimentary Asymmetric Deterrence Game is a model of an asymmetric or one-sided deterrence situation: State B wishes to deter State A, but not the other way around. Thus, in the game depicted in ‹gure 2.2, State A begins play at decision node 1 by deciding whether to cooperate (C) and accept the status quo, or to defect (D) and demand its alteration. If State A chooses (C), the game ends and the outcome is the Status Quo. But if State A defects, State B must decide at node 2 whether to concede (C) the issue, in which case the outcome is A Wins, or deny (D) the demand and precipitate Con›ict. For the sake of argument, let us accept, for the moment at least, two core assumptions of classical deterrence theory: (1) that Con›ict is the worst possible outcome (i.e., the cost of war is exorbitant), and (2) that the players are instrumentally rational. Next, we ask what instrumentally rational players would do when presented with the choices in the Rudimentary Asymmetric Deterrence Game. To answer this question, the game tree of ‹gure 2.2 is examined using backward induction. To apply this procedure, one works backward up the tree and determines ‹rst what an instrumentally rational State B would do at node 2; then, using this information, the rational choice of State A at node 1 is 33. I use a simple convention to distinguish between the players and the outcomes in a game model and their real-world analogues. The names of the players (e.g., Defender) are capitalized; the labels of game outcomes (e.g., Status Quo) are italicized. The names of real-world actors and the labels of actual outcomes (e.g., defender and status quo) are not.
Theories and Explanations
State A
Cooperate
35
Node 1
Defect
Status Quo
Node 2
State B
Concede
A Wins
Deny
Conflict
Fig. 2.2. Rudimentary Asymmetric Deterrence Game
speci‹ed. The strategy set that survives the backward induction process in this extensive-form game with complete information will be part of a subgameperfect equilibrium.34 At node 2, State B is faced with a choice between conceding (i.e., choosing C), which brings about outcome A Wins, and denying State A’s demand (i.e., choosing D), which brings about Con›ict. But if Con›ict is assumed to be the worst possible outcome, an instrumentally rational State B will choose to concede since, by assumption, A Wins is the more preferred outcome. Given that State B will rationally concede at node 2, what should State A do at node 1? State A can choose either to cooperate, in which case the outcome will be the Status Quo, or to defect, in which case the outcome will be A Wins, because an instrumentally rational State B will concede at node 2. If State A prefers A Wins to the Status Quo—that is, if it has an incentive to upset the Status Quo and, therefore, needs to be deterred—it will rationally defect. Given the two core assumptions of classical deterrence theory, then, the Status Quo is unstable and deterrence rationally fails. This shows that even within the con‹nes of this elementary model, the theory’s de‹ning assumptions are logically incompatible with the possibility of deterrence success. Nonetheless, classical deterrence theorists contend that bilateral nuclear relationships are exceedingly stable. 34. As noted in section 1.6, a subgame-perfect equilibrium is the accepted measure of rational behavior in an extensive-form game of complete information. For a lucid discussion of backward induction and subgame perfection with easy-to-understand examples, see Morrow 1994: 121–35.
36
the games of july
The same conclusion can be drawn from a formal analysis of Chicken. Since the Status Quo is also not part of a (pure strategy) Nash equilibrium in Chicken, policies that unconditionally support the status quo in this game are also incompatible with rational choice. Thus, the assumptions that delineate decision-theoretic deterrence theory are inconsistent with the persistence of peace throughout the Cold War period.35 Moreover, since decision-theoretic deterrence theory is axiomatically consistent with structural deterrence theory, this de‹ciency of decision-theoretic deterrence theory casts doubt on the latter approach to deterrence as well. As Van Gelder (1989: 159) observes, the lack of congruence between the assumptions and the conclusions of classical deterrence theory “threatens the very foundations of nuclear deterrence as a rational strategy.” Some classical deterrence theorists do not seem particularly troubled by the logical contradiction. Casting consistency aside, they explain the stability of the Cold War period by allowing for the possibility of irrational threats being executed. This intellectual sleight of hand is quite common in the literature of deterrence, which is why Achen (1987: 92) notes that “far from leaning too heavily on rational choice postulates, ‘rational deterrence theory’ necessarily assumes that nations are not always self-interestedly rational.” Schelling (1966: 37), for example, argues that “it does not always help to be, or to be believed to be, fully rational, cool-headed, and in control of oneself or of one’s country.” Brodie (1959: 293) puts it this way: “For the sake of deterrence before hostilities, the enemy must expect us to be vindictive and irrational if he attacks us.”36 This inconsistent use of the rationality postulate is more than problematic. It is well established that any statement and its opposite can be derived from a logically inconsistent theory. For precisely this reason, contradictory theoretical frameworks cannot be falsi‹ed (King, Keohane, and Verba 1994: 105), which is why logical consistency is the sine qua non of good theory. Walt (1999a) notwithstanding, theoretical explanations that fail to satisfy this strict standard clearly must either be repaired or set aside. It is more than understandable, then, that some decision-theoretic deterrence theorists (e.g., Nalebuff 1986; Powell 1987, 2003) have attempted to avoid logical contradiction by introducing a deus ex machina, a third player called Nature, to carry out threats that are independent of the preferences of the players. For this group of theorists, successful deterrence is explained by the au35. The Status Quo might survive rational play under the mixed-strategy equilibrium, but only sometimes and not necessarily often. Thus, the existence of a mixed-strategy equilibrium in Chicken does not help to explain the absence of a superpower con›ict during the Cold War. 36. For other examples, see Gauthier 1984; Kahn 1962: 122; Rhodes 1988; G. Snyder 1966: 251.
Theories and Explanations
37
tonomous risk of an all-out con›ict, or by what Schelling (1960: chap. 8) calls the “threat that leaves something to chance.” But this theoretical pirouette also does not suf‹ce. As even Powell (1987: 725) admits, “Relying on Nature to impose the irrational sanctions does not really solve the credibility problem” and calls into question the logical foundations of decision-theoretic deterrence theory. This explanation is also dif‹cult to square with the fact that history provides no known example of an inadvertent war (Mueller 1995: 114–15; Trachtenberg 1990/91, 1991), or with the observation that during the Cold War, the superpowers worked mightily to reduce the probability of an accidental war.37
2.2. Coda The earliest attempts to explain the Great War were sponsored by the various belligerent governments and, therefore, are one-sided and patently unreliable. But soon after the war, diplomatic historians offered more dispassionate analyses. Following Fay, some of these accounts focused on underlying or long-term determinants; others explored the war’s immediate or short-term causes; and still others examined both. Diplomatic historians have made important and valuable contributions to our understanding of the origins of the Great War. But they fail to develop compelling causal explanations, although Trachtenberg (1991) comes very close. By their nature, it is almost impossible to establish a direct causal connection between variables that are temporally remote. And without an explicit theoretical component, even those explanations that call on contemporaneous factors to explain the war must necessarily fall short of the mark. It was thus natural to turn to the dominant paradigm in the ‹eld of international relations to look for the war’s cause(s). But the theory of war initiation rooted in the realist paradigm, here called classical (or rational) deterrence theory, was also found to be de‹cient, riddled as it is by empirical inaccuracies and logical inconsistencies.38 Even if its “originality” is stipulated, classical deterrence theory clearly fails to satisfy even the relaxed standards Walt (1999a) offers for evaluating theory. Jervis (1979: 292) notes that many of the theory’s policy prescriptions are “contrary to common sense.” And Rapoport (1992: 482) calls the theory’s nostrums “bizarre.” Classical deterrence theorists are unable to explain, in either a logically 37. For a discussion of some formal attempts to reconcile rationality with deterrence, see Zagare and Kilgour 2000: chap. 2. 38. Or, as some have put it, by puzzles, dilemmas, and paradoxes.
38
the games of july
consistent or an empirically supportable way, the outbreak of war under parity conditions in 1914. The argument that the war occurred because its costs were believed to be too low is dubious at best and circular at worst. And the argument that the war occurred because of diplomatic blunders, extraneous forces, and cognitive and psychological failures is inconsistent with classical deterrence theory’s hard core of protected assumptions of egotistical, self-interested, rational behavior. Similarly, the explanation that the “remarkable stability” of the Cold War period was a consequence of the “twin pillars” of deadly nuclear weapons and a carefully maintained strategic balance fails to square with the observation that a superpower war did not break out during periods of clear-cut U.S. superiority. Within the con‹nes of the theory’s axiomatic base, then, the only way to explain the long peace is to assume that the players are simultaneously rational (when they are being deterred) and irrational (when they are deterring their opponent). But since all things are possible when players can be either rational or irrational, classical deterrence theory, like realism in general, de‹es falsi‹cation.39 It is, therefore, a degenerative research program (Vasquez 1997). I do not mean to suggest that no chance events, no technological imperatives, no cognitive failures, no psychological de‹ciencies, and no idiosyncratic factors were at work during the July Crisis. Quite the contrary. It is well documented that all of these factors and several others probably came into play, in one way or another, in 1914. Many leading decision makers made their fateful choices under a great deal of psychological and physical stress; pressures also clearly existed on both sides to gain a tactical advantage by mobilizing more quickly than the opponent; the makers of foreign policy did not always operate with complete information; and signi‹cant misunderstandings occurred between friends and foes alike. In other words, the collection of choices made during the crisis were made by fallible human beings. They were not, however, the choices made by the actors who populate the realist paradigm: unitary, egotistical, and procedurally rational decision-making units that make cool, clearheaded, ends-means calculations that objectively maximize either a state’s power or its security.
39. For example, Van Evera (1999: 247–48) argues that nuclear weapons will deter those states that are “deterrable” but will not have a pacifying effect on those states that are “nondeterrable.” It is dif‹cult to disagree.
This time I shall not chicken out. wilhelm ii
chapter three
Perfect Deterrence Theory: An Overview
To this point, we have seen that extant explanations of the Great War fall short of the mark. Historical narratives fail to provide the element of necessity essential to a satisfying explanation, while studies that focus on long-term or underlying determinants are hard put to establish a causal connection between these background factors, some of which are near constants, and the speci‹c decisions that led to war in early August 1914. Similarly, on its face, the Great War would seem to refute so-called liberal explanations of interstate con›ict. The extensive network of economic relationships that existed in 1914 among the great powers of Europe did not prevent decision makers in Vienna, St. Petersburg, Berlin, Paris, and London from taking the ill-conceived steps that led to the catastrophe. Rather than explaining the con›ict, then, theories of economic deterrence are cast into doubt by the events that spoiled “Europe’s last summer.” Finally, explanations rooted in the realist paradigm were shown to be either logically or empirically de‹cient. Structural explanations for the Great War that claim that its perceived costs were insuf‹cient for deterrence to work are dif‹cult to sustain empirically. Recognizing this, some structuralists blame the decisions that led to war on incompetent policymakers who failed to heed the strategic imperatives of the international system. But since blundering actors are nowhere to be found in the paradigm’s set of core assumptions, the only thing that this argument demonstrates is the bankruptcy of the structuralist research program. Decision-theoretic deterrence theorists do not fare much better. Formal models that start with the structuralist conclusion that war was “irrational,” in both 1914 and during the Cold War, can explain neither—unless logical consistency is abandoned. In the latter case, these models explain anything and 39
40
the games of july
everything because they are nonfalsi‹able. Nonfalsi‹able theories are not very illuminating. To explain the Great War, then, another theory is needed. Accordingly, in this chapter I propose an alternative theoretical framework, perfect deterrence theory, that I use in the remaining chapters to explain why the interstate system failed in 1914. Here I simply sketch the theory’s broad contours and contrast its assumptions and theoretical characteristics, its major propositions, and its most important policy implications with those of classical deterrence theory. I do not attempt, now or later, to explain in minute detail any of the several models that constitute perfect deterrence theory. And in the empirical chapters that follow, only those theoretical details that are pertinent to an explanation of the July Crisis will be provided. Readers interested in a more pointed discussion and the technical particulars should consult Zagare and Kilgour (2000). This chapter is more detached from the events of July and August 1914 than any other in this volume. Perfect deterrence theory is a universal theory applicable to a wide range of interstate con›icts, and I present it as such to avoid obscuring its generality. As noted in chapter 1, one aim of this book is to construct a theoretical explanation of the outbreak of World War I. Another is to subject perfect deterrence theory to more intense empirical scrutiny. Both objectives would be undermined if perfect deterrence theory were presented selectively, solely in the context of the July Crisis.
3.1. Axioms and General Theoretical Characteristics Perfect deterrence theory was developed to overcome the empirical and the logical de‹ciencies of realist explanations for interstate con›ict initiation, limitation, escalation, and resolution. Since the standard realist explanation has evolved since the dawn of the nuclear age, it should be no surprise that many (but not all) of the most important distinctions between classical deterrence theory, the modern incarnation of the realist theory of war initiation and avoidance, and perfect deterrence theory involve the dynamics of the post–World War II international system in general and the strategic relationship of the superpowers in particular. The two theories offer divergent explanations not only of the Great War but also of the characteristics of the long peace that came about after 1945. As a consequence, much of what follows in this chapter has a more contemporary ›avor. But while this temporary shift in focus may strike some as empirically peculiar, it is important to remember that
Perfect Deterrence Theory
41
any theory that purports to explain why wars occur must also explain when and why they do not occur. A secondary purpose of this chapter, therefore, is to demonstrate that only perfect deterrence theory can offer a logically consistent and empirically plausible explanation for the absence of a superpower war since 1945. Unlike classical deterrence theory, perfect deterrence theory makes consistent use of the rationality postulate. A number of “plausibility probes” suggests that this commonsense approach to interstate con›ict is also consistent with the empirical record (Senese and Quackenbush 2003; Quackenbush 2010).1 Table 3.1 summarizes the most important differences between perfect deterrence theory and classical deterrence theory.2 As table 3.1 shows, some of the axioms of perfect deterrence theory and classical deterrence theory are the same. For example, both theoretical frameworks assume states (and their decision makers) to be instrumentally rational and egotistical. In perfect deterrence theory, however, they are not necessarily assumed to be undifferentiated. A number of other critical differences also exist between the two theories. First, in perfect deterrence theory, an outright challenge is not assumed to automatically culminate in war.3 In perfect deterrence theory, the players always have an opportunity not to resist or retaliate. Since response options are not necessarily executed, the possibility of a one-sided victory is not assumed away.4 Second, in perfect deterrence theory, only players can execute deterrent threats. Thus, in perfect deterrence theory, an opponent’s threat, not some impersonal force (i.e., Nature), is the principal source of the risks run by the players. Third, only rational (i.e., credible) threats can be carried out. This stricture ensures that the deductions of perfect deterrence theory remain consistent with the rationality postulate in general, and with Selten’s perfectness criterion
1. Empirical support for perfect deterrence theory is reviewed more systematically in Zagare 2007. See also section 8.2. 2. Not all of the items arrayed in table 3.1 are discussed in this chapter. For additional details, see Zagare 2004, the table’s source. See also Zagare 2007. 3. The assumption that all-out attacks are always reciprocated is common in the formal literature of deterrence. Powell 1987 is an instructive example. (For others, see Bueno de Mesquita, Morrow, and Zorick 1997: 17; Fearon 1994: 590; Kydd 1997: 379.) One direct consequence of this assumption is that ‹rst strikes can never be rational in Powell’s model—as Morrow (1994: 180–81) matter-of-factly observes—simply because they always lead to a player’s worst outcome, Con›ict. Hence, at the strategic level, Powell’s (and similar) models postulate rather than derive stability. But the same could also be said of the mainstream deterrence literature. As Powell (1985: 83) notes elsewhere, “In the theory [of deterrence that takes invulnerable second-strike forces as a given] the risk of an unrestricted nuclear attack is assumed away.” 4. See, e.g., the analysis of the Rudimentary Asymmetric Deterrence Game in chapter 2.
Table 3.1. Classical Deterrence Theory and Perfect Deterrence Theory Compared Classical Deterrence Theory
Perfect Deterrence Theory
Assumptions Actors States Procedural rationality Instrumental rationality All-out attacks Strategic stability Irrational threats Credibility Preferences/Interests
egotistical undifferentiated sometimes sometimes always reciprocated postulated may be executed constant objectively discovered
egotistical may be differentiated not assumed always may be reciprocated derived never executed not fixed subjectively defined
Theoretical Characteristics Logically consistent Empirical validity
no uncertain
yes consistent with extant empirical literature
unimportant/ignored robust/all but certain strictly positive and monotonic unstable
significant fragile/contingent nonmonotonic
very stable sufficient for deterrence success unexplained
potentially unstable necessary, but not sufficient, for deterrence success placed in theoretical context
supports opposes opposes supports coercive, based on increasing war costs and inflexible bargaining tactics
opposes supports supports opposes conditionally cooperative, based on reciprocity
Propositions Status quo Strategic deterrence Relationship between conflict costs and deterrence success Asymmetric power relationships Parity relationships Capability Limited conflicts and escalation spirals Policies Overkill capability Minimum deterrence “Significant” arms reductions Proliferation Negotiating stances
potentially very stable
Perfect Deterrence Theory
43
in particular—hence, the theory’s name.5 Finally, while credibility is ‹xed and constant in the most in›uential formal speci‹cations of classical deterrence theory, it is measured on a continuum in perfect deterrence theory. The last feature of perfect deterrence theory requires special comment. In classical deterrence theory, deterrent threats are, perforce, always presumed to be incredible because decision-theoretic deterrence theorists, by de‹nition, take con›ict to be the mutually worst outcome. If credibility is equated with instrumental rationality, as it is in both the formal and the wider strategic literature (see section 3.3.2), a threat that leads to a threatener’s worst outcome can never be credible because carrying out such a threat can never be rational (Zagare 1990). And threats that are always incredible can never vary.6 Once credibility is considered a constant, no logical relationship between it and stable deterrence can ever be established. But even if credibility were allowed to vary in classical deterrence theory, a fundamental problem would still exist: as long as incredible threats can be executed, the possibility of exploring the theoretical relationship between the credibility of threats and the operation of deterrence is precluded. By contrast, in perfect deterrence theory, credibility can indeed vary, and irrational threats cannot be carried out. Thus, unlike classical deterrence theory, perfect deterrence theory is well situated to explore the logical connection between threat credibility and the dynamics of dyadic interstate relationships. It is, therefore, a more general theory. At ‹rst blush, the axiomatic differences between classical deterrence theory and perfect deterrence theory might appear to be minor, perhaps even insigni‹cant. But as illustrated in the ‹lms Sliding Doors and Run Lola Run, small differences in initial assumptions can have important theoretical consequences and signi‹cant policy implications.7 The major conclusions of perfect deterrence theory are drawn from the application of its axiomatic base (see table 3.1) to four interrelated incomplete
5. Section 1.6 notes that that the perfectness criterion requires that players always make rational choices—that is, that they plan to choose the best course of action at every node of the game tree, including those that are never reached under rational play. 6. As Walt (1999: 123) correctly points out, some classical deterrence theorists (e.g., Schelling 1966) have speculated about mechanisms that enhance credibility or circumstances that make threats more or less credible. These discussions, however, should be kept separate from the theory that gives rise to them. Many of the prescriptions developed by decision-theoretic deterrence theorists require states to credibly (i.e., rationally) threaten war (or con›ict) to deter aggression. But in the brinkmanship models that underpin these prescriptions, it is always irrational to execute a deterrent threat. In other words, it is logically inconsistent to treat patently self-abnegating (i.e., absurd) threats as rational, as credible, or as variable (Trachtenberg 1991: 32). 7. For a formal demonstration, see Bueno de Mesquita 1985. In chaos theory, this is called the butter›y effect.
44
the games of july
information game models.8 Two of the models are of direct deterrence relationships in which at least one state is attempting to deter the other: the Generalized Mutual Deterrence Game (see ‹gure 3.1) is a generic model of those direct deterrence situations in which each state threatens the other. By contrast, in the Unilateral Deterrence Game (see ‹gure 3.2), one player (Defender) prefers to preserve the status quo while the other (Challenger) prefers to upset it. The third model, the Tripartite Crisis Game (see ‹gure 4.1), is a model of extended deterrence relationships wherein one state seeks to deter an attack against a third party. The fourth model, the Asymmetric Escalation Game (see ‹gure 6.1), is a hybrid that applies to either type of deterrence situation.9 Space and other considerations preclude a formal analysis of these models here. But formal demonstration is not the purpose of this chapter. Rather, it seeks to show that important theoretical differences, with signi‹cant policy implications, ›ow from ostensibly minor alterations of classical deterrence theory’s axiomatic base.
3.2. Explaining the Absence of War Classical deterrence theory and perfect deterrence theory offer divergent explanations not only of the outbreak of war but also of its absence. To highlight these important differences, I next offer an informal analysis of the strategic properties of the Generalized Mutual Deterrence Game under incomplete information.10 Since this model presumes undifferentiated actors (i.e., each player is dissatis‹ed with the status quo), it provides a particularly apt context in which to compare the deductions of classical deterrence theory with those of perfect deterrence theory.11 In the Generalized Mutual Deterrence Game, there are a number of condi8. Recall that in a game of incomplete information, at least one player is uncertain of another’s preferences. 9. The Rudimentary Asymmetric Deterrence Game is an elementary model, used primarily to de‹ne terms and illustrate concepts. 10. At the start of this game, both players simultaneously choose to cooperate (C) or to defect (D). If both choose either C or D, the game ends. But if one chooses C and the other chooses D, the player choosing C is provided with another opportunity to defect—that is, to retaliate. If at the end of the game, both players have chosen C, the Status Quo prevails; if both have chosen D, Con›ict (or war) results. If one player chooses C and the other D, the defecting player gains an advantage—either A Wins or B Wins. 11. The vertical line connecting State B’s moves at node 2 is called an information set. Information sets are simply graphical devices used to indicate what a player knows or does not know about prior choices at a particular decision point in a game. As drawn here, the information set joining nodes 2a and 2b indicates that when State B makes a choice at decision node 2, it does not know exactly where it is on the game tree. A decision node that is not joined with another in the same information set is called a singleton. (Nodes 1, 3a, and 3b are singletons.) A player whose present information set is a singleton can determine its place on the game tree and therefore can ‹gure out what choices had been made previously. There are two theoretically equivalent ways to interpret the sequence of the ‹rst two moves in the Gener-
Perfect Deterrence Theory
Nodes 2a and 2b
Node 11 Node
45
Nodes 3a and 3b
Status Quo (SQ) B Wins (CD)
Conflict (DD)
A Wins (DC)
Conflict (DD) Conflict (DD)
Key: C = Cooperate D = Defect
Fig. 3.1. Generalized Mutual Deterrence Game
tions under which the survival of the Status Quo is consistent with rational choice. Not all of these conditions, however, are equally probable. The Status Quo is most likely to endure when a Sure-Thing Deterrence Equilibrium exists.12 Under this equilibrium, neither player has an incentive to challenge the other. Hence, peace is at hand. Since many of perfect deterrence theory’s policy prescriptions ›ow from alized Mutual Deterrence Game: (1) State A makes the ‹rst move and State B, uninformed of that choice, makes the second; (2) both players make their choices simultaneously. In each case, neither player knows what the other has done (or will do) before making a choice. In a game of perfect information, all the players know where they are on the game tree whenever it is their turn to make a decision—that is, all decision nodes are singletons. The Generalized Mutual Deterrence Game, then, is not a game of perfect information. Perfect (and imperfect) information should be distinguished from complete (and incomplete) information. 12. Since it arises in a game of incomplete information, the Sure-Thing Deterrence Equilibrium is a perfect Bayesian equilibrium. (A detailed exposition of the technical requirements of a perfect Bayesian equilibrium appears in chapter 4. For a brief overview, see section 1.6.) Henceforth, unless quali‹ed, the term equilibrium should be taken to imply a perfect Bayesian equilibrium.
46
the games of july
Challenger
Not
Node 11 Node
Demand
Demand
Status Quo
Defender
Concede
Defender Concedes
Node Node 22
Resist
Challenger Back
Hold
Down
Firm
Challenger Defeated
Node 3
Conflict
Fig. 3.2. Unilateral Deterrence Game
the strategic properties of the Sure-Thing Deterrence Equilibrium (and analogous equilibria in related models),13 it is instructive brie›y to highlight some of its salient strategic characteristics. • For a Sure-Thing Deterrence Equilibrium to exist, both players must project highly credible threats. In the absence of this condition, the status quo is unlikely to survive rational play. Recall that decision-theoretic deterrence theorists implicitly presume that all deterrent threats lack credibility. As a result, they are, logically, unable to make this perhaps obvious but nonetheless incontrovertible connection between threat credibility and deterrence stability. • The likelihood that a Sure-Thing Deterrence Equilibrium will exist is increased, ceteris paribus, as the cost of con›ict is increased. Signi‹cantly, however, an increase in con›ict costs does not always increase 13. For example, the strategic properties of the Certain Deterrence Equilibrium in the Unilateral Deterrence Game are quite similar to those of the Sure-Thing Deterrence Equilibrium in the Generalized Mutual Deterrence Game (Zagare and Kilgour 2000: 149–50).
Perfect Deterrence Theory
47
the likelihood that a Sure-Thing Deterrence Equilibrium will exist, suggesting that there are distinct limits to the stabilizing impact of modern weapons systems and that an overkill capability, recommended by many classical deterrence theorists, is just that. • A Sure-Thing Deterrence Equilibrium is more likely to exist and, consequently, the status quo is more likely to survive when it is highly valued by the players. While this observation might appear to be self-evident, it is noteworthy that classical deterrence theorists, who tend to focus on threat capability, all but ignore the impact of satisfaction (or dissatisfaction) with the status quo on deterrence stability.14 In consequence, their policy prescriptions tend to slight the importance of diplomatic initiatives in preserving peace (e.g., Kagan 1995). In the Generalized Mutual Deterrence Game, there is a theoretical possibility, albeit remote, of the status quo persisting when the conditions taken as axiomatic by decision-theoretic deterrence theorists are approached. In the Generalized Mutual Deterrence Game, a Bluff Equilibrium exists whenever both players have threats that are all but incredible. Under a Bluff Equilibrium, both players generally, but not always, initiate con›ict. In consequence, there may be times in which no challenge to the status quo is made. It is unlikely in the extreme, however, for the status quo to survive rational play under a Bluff Equilibrium over the long haul. Within the con‹nes of the Generalized Mutual Deterrence Game, then, the only explanation of the long peace that is consistent with the canons of rationality and with the axioms of classical deterrence theory is that luck prevailed.15 Of course, since this dubious argument is inconsistent with the core conclusion that, under parity, the stability of nuclear and other high-cost relationships is extremely robust, few classical deterrence theorists make it. For this reason, these theorists have opted to explain the stability of the postwar period either by sacri‹cing logical consistency or by making special assumptions that require a disinterested actor to carry out all deterrent threats (see chapter 2). The recommendations that ›ow from the classical view of deterrence include policies that favor an overkill capability and promote proliferation, and that guard against arms reductions that are carried “too far” (Intrili14. As Powell (1985: 96) observes, classical deterrence theory holds “the fundamental con›ict of interest underlying a crisis as ‹xed.” 15. The same is true of Chicken. Under the mixed-strategy equilibrium, the status quo occurs sometimes, but not necessarily often.
48
the games of july
gator and Brito 1984). Moreover, during a crisis, statesmen are encouraged to seek an advantage by reducing ›exibility, by exercising implacability, or by behaving recklessly. In this context it should also be noted that it is possible for the status quo to survive rational play in the Unilateral Deterrence Game of ‹gure 3.2 even when all deterrent threats are minimally credible.16 One might argue, therefore, that this result squares classical deterrence theory with the restrictions of rationality. It would—except for the fact that it requires yet another assumption that lies outside the theory: differentiated actors, which explains why some classical deterrence theorists selectively adhere to the assumption that all states are similarly motivated. To be somewhat more speci‹c, in the Unilateral Deterrence Game, the players have distinct roles and distinct motivations: one player, the Defender, hopes to preserve the status quo, while the other, the Challenger, would prefer to overturn it. In classical deterrence theory, however, all states are considered alike.17 Thus, to use this simple game model to construct an explanation for the absence of a superpower con›ict, one must necessarily cast off yet another axiom of classical deterrence theory. All of which is simply another way of saying that any theory that posits egotistical, rational, and undifferentiated actors, all of whom lack a credible retaliatory threat, is inconsistent with the actual stability of the post–World War II international system. In contrast, an explanation of the stability of the post-1945 system arises quite naturally in perfect deterrence theory: all-out con›ict was avoided simply because each side’s retaliatory threat was suf‹ciently capable and credible to deter either superpower from attacking the other. While this explanation might appear unexceptional, it runs counter to the conventional wisdom. Perfect deterrence theory’s policy implications also stand in stark contrast to those of classical deterrence theory: states should, inter alia, develop a minimum deterrent capability, pursue arms control agreements, cap military spending, avoid proliferation policies, and when in crisis, seek compromise by 16. In this game, Challenger begins play by either accepting or demanding an adjustment of the status quo. If no demand is made, the game ends and the Status Quo prevails. But if Challenger demands a change, Defender must decide whether to concede or to resist. When Defender concedes, Challenger gains an advantage (i.e., the outcome is Defender Concedes). When Defender resists, Challenger either backs down or holds ‹rm. In the former case, Defender gains an advantage (i.e., the outcome is Challenger Defeated); in the latter case, Con›ict occurs. 17. The assumption of undifferentiated actors is explicit in structural deterrence theory. It is clearly implicit in the Chicken analogy and in most models developed by decision-theoretic deterrence theorists. Ellsberg’s (1959) model is an important exception.
Perfect Deterrence Theory
49
adopting ‹rm but ›exible negotiating stances and tit-for-tat military deployments. Walt (1999a: 25), then, is factually incorrect in asserting that perfect deterrence theory reinvents “the central elements of deterrence theory without improving on it.”18
3.3. Theoretical Propositions, Empirical Expectations, and Policy Implications What are perfect deterrence theory’s most important theoretical propositions? What empirical expectations arise from its axiomatic base? And what policy prescriptions follow from a strategic theory that respects logical consistency? This section answers each of these questions in the context of the theory’s principal variables, which include threat capability and credibility, the cost of con›ict, and satisfaction (or dissatisfaction) with the status quo. Among the more speci‹c queries raised are: What conditions are most conducive to deterrence success? What is the precise connection between threat credibility and the stability of the status quo? What negotiating style and crisis management techniques are most conducive to peace? Later chapters address a number of related questions in the speci‹c context of the Great War: How do the dynamics of direct deterrence relationships differ from those of extended deterrence relationships? Are limited con›icts possible, and if so, when? Why do some con›icts escalate while others do not?
3.3.1. capability In perfect deterrence theory, capable threats are threats that would hurt.19 Actions that hurt are those that leave a player worse off than if the action were not executed. Operationally, this means that one player’s threat is capable only if the other—the threatened player—prefers the status quo to the outcome that results when and if the threat is carried out. In other words, a threat will lack capability whenever the threatened player prefers to act even when a deterrent threat is acted on (Zagare 1987). When de‹ned in this way, a threat may lack capability for a number of reasons, the most important of which is that the threatened player calculates that the cost of con›ict is less than the cost of doing nothing. Belgium’s threat to re18. Table 3.1 highlights the divergent empirical expectations and policy prescriptions of classical deterrence theory and perfect deterrence theory. 19. This de‹nition is, in fact, Schelling’s (1966: 7).
50
the games of july
sist a German attack in 1914 is a good example.20 Germany’s plans did not turn on Belgium’s intentions—with or without Belgium’s resistance, Germany preferred to invade (Tuchman 1962: 40).21 Britain’s threat to intervene in a continental war may also have lacked capability. In 1914, the British Expeditionary Force was, by European standards, small. In consequence, some German leaders disparaged its import. Bismarck, for example, once joked that if the British army ever landed in Germany, he would simply have his constabulary force round it up and place it under arrest. Count Alfred von Schlieffen, the architect of Germany’s war plan, also thought that the British army would have no material impact should a land war break out on the continent (Kagan 1995: 212). Of course, the threat represented by the British ›eet was undoubtedly capable, which is one reason why Admiral Alfred von Tirpitz, Germany’s state secretary of the Imperial Navy, was not among those military leaders pushing for war in 1914. There is considerable opinion in the theoretical literature of international relations that threat capability constitutes a suf‹cient condition for deterrence success (Levy 1988: 489–90).22 Not so in perfect deterrence theory, where deterrence may fail even when threats are capable all around. On the other hand, in perfect deterrence theory, capability emerges as the only condition absolutely necessary for deterrence success; when one or both states in a mutual deterrence relationship lack capability, deterrence is bound to fail. Since weak states, almost by de‹nition, usually lack the ability to hurt larger, stronger opponents, it should come as no surprise that there is strong empirical support for the proposition that interstate con›ict initiators are generally stronger than their opponents (Bueno de Mesquita 1981: 155–56).23 These data are consistent with—indeed, they provide compelling systematic empirical support for—an important conclusion of perfect deterrence theory. In addition, Harvey’s 20. Another example is the facetious threat Germany’s chancellor, Otto von Bismarck, made to the Austrian foreign minister, Count Julius Andrássy, when negotiating the Austro-German Alliance in 1879. According to Langer (1950: 183), Bismarck, who very much wanted an agreement, moved very close to Andrássy and growled in a threatening tone, “Accept my proposal, else . . . else I shall have to accept yours.” This clearly was not a threat that hurt—or worked. As discussed in chapter 4, the terms of the Dual Alliance re›ected Austria’s preferences. 21. F. Fischer’s (1967; 1975) argument that Germany set out to precipitate a continental war in 1914 implicitly assumes that Russia’s and France’s retaliatory threats also lacked capability. 22. Quinlan (2000/2001: 142), for example, all but accepts the suf‹ciency of capability for stabilizing hostile bilateral relationships. He writes of the strategic relationship of the United States and the Soviet Union during the Cold War, “The prodigious size to which the two nuclear armouries grew imposed a massive caution almost irrespective of the precise credibility of doctrine for use.” Existential deterrence theorists such as Bundy (1983), who hold that the mere existence of nuclear weapons virtually assures strategic stability, also see a highly capable retaliatory threat as suf‹cient for avoiding crises and war. 23. Austria’s decision to attack Serbia in 1914 is the relevant case in point.
Perfect Deterrence Theory
51
(1998: 691) more recent empirical study “indirectly supports” perfect deterrence theory’s conclusions about the crucial role of capability in deterrence relationships.
3.3.2. credibility In the strategic literature, credible threats are frequently equated with threats that ought to be believed (e.g., Smoke 1987: 93); threats can be believed only when they are rational to carry out (Betts 1987: 12); thus, only rational threats can be credible (Lebow 1981: 15). In perfect deterrence theory, the formal de‹nition of credibility is consistent with the theoretical linkage between threats that are credible and threats that are both believable and rational: credible threats are precisely those that are consistent with Selten’s (1975) perfectness criterion—that is, with threats that the threatener prefers to execute. While it is perhaps not surprising to learn that perfect deterrence theory holds that a capable retaliatory threat is a necessary (but not suf‹cient) condition for deterrence success, it may in fact be surprising to learn that credible threats are neither necessary nor suf‹cient for deterrence to succeed. This means, inter alia, that deterrence may fail even when all retaliatory threats are capable, and deterrence may succeed even when all retaliatory threats are incredible.24 Nonetheless, given capable threats, deterrence is most likely to prevail, ceteris paribus, when all threats are highly credible, a straightforward and seemingly unexceptional result that simply cannot be derived from models that presume that all retaliatory threats are invariably incredible. The inability of classical deterrence theory to come to such an obvious conclusion without violating the canons of logic speaks to the inadequacies of its theoretical underpinnings. Although credible threats are not quite the “magic ingredient” of deterrence, as Freedman (1989: 96) asserts, they come close. Still, deterrence may succeed even when a defender’s threat is incredible. The key to this possibility, however, is not the characteristics of the defender’s threat but those of the challenger. In a number of deterrence situations, a challenger whose retaliatory threat is itself not credible cannot deter a defender from resisting. In consequence, the challenger is deterred and the status quo survives rational play. It is not inconceivable, then, for deterrence to succeed even when a defender prefers not to execute its endgame threat. Thus, in perfect deterrence theory, a credible threat is also not a necessary condition for successful deterrence. 24. For an informal demonstration, see Zagare 2004: 125.
52
the games of july
That deterrence might rationally work even when a defender’s threat is incredible is an important insight into the interactive nature of deterrence relationships. It is also an insight that is missed by those theorists who focus exclusively on the characteristics of a defender’s retaliatory threat.25 In consequence, they produce a misleading and perhaps even a distorted understanding of the dynamics of deterrence. Perfect deterrence theory also ‹nds that mutual deterrence can (but need not) fail when both players have capable and credible retaliatory threats. Even when deterrence is consistent with the strictures of rationality, other rational possibilities frequently exist, and some of them are associated with an all-out con›ict.26 This conclusion contrasts sharply with classical deterrence theory’s supposition that parity and high war costs virtually eliminate the possibility of a (rational) deterrence breakdown. In other words, in contrast to classical deterrence theory, perfect deterrence theory ‹nds that bilateral deterrence relationships are fragile and fraught with peril. That mutual deterrence is not necessarily robust has important implications for the wisdom of balance-ofpower policies in general, and proliferation policies in particular. To put this in a slightly different way, in the set of interrelated models that forms the basis of perfect deterrence theory, it is almost always the case that the conditions that make peace a real possibility are exactly the same as those associated with all-out con›ict. From the vantage point of perfect deterrence theory, then, wars do not arise as the inevitable consequence of impersonal forces that lie beyond human intervention or control.27 Rather, they result from choices made by fallible human beings acting rationally, though not necessarily wisely. The good news is that this implication means that skillful diplomacy and adroit statesmanship may sometimes save the day, suggesting that with different leaders or different policies, World War I could have been avoided. The bad news, however, is that peace can never be all but guaranteed, as some classical deterrence theorists suggest. World War I, unfortunately, illustrates the point. 25. For example, Lebow (1981: 85) writes that “four conditions emerge as crucial to successful deterrence. Nations must (1) de‹ne their commitment clearly, (2) communicate its existence to possible adversaries, (3) develop the means to defend it, or to punish adversaries who challenge it, and (4) demonstrate their resolve to carry out the actions this entails.” Of these four conditions, only the third, which can be interpreted as threat capability, emerges as a necessary condition in perfect deterrence theory. 26. More technically, multiple equilibria almost always exist. 27. According to Powell (1985: 84), the fact that “‘the participants are not fully in control of events’ is fundamental to much of strategic nuclear deterrence theory.” And Trachtenberg (1990/91: 120), who concurs, comments that the supposition that a major war could occur when statesmen lose control of events “is one of the most basic and most common notions in contemporary American strategic thought.”
Perfect Deterrence Theory
53
3.3.3. status quo evaluations In classical deterrence theory, states are generally thought of as undifferentiated actors. As such, they have identical interests and aspirations and equal motivations to overturn the status quo (Legro and Moravcsik 1999: 13).28 As Mearsheimer (1990: 12) writes, “All other states are potential threats.” There are no exceptions to this dictum. This means that there can be no variation in the utility (or disutility) states derive from the existing order. All states are presumed to be perpetually dissatis‹ed. (See also Mearsheimer 2001: 2.) By contrast, in perfect deterrence theory, the players are not necessarily assumed to be undifferentiated. Some, in theory, may be content with the prevailing status quo and, consequently, may lack the motivation to upset it. But even when both players are dissatis‹ed, the extent of their dissatisfaction may differ. In other words, in perfect deterrence theory, the value of the status quo is an important strategic variable: as satisfaction with the status quo increases, ceteris paribus, so does the likelihood of deterrence success.29 Again, because classical deterrence theorists tend to treat the value of the status quo as a constant, they are unable to derive logically this obvious conclusion. As a result, most classical deterrence theorists favor coercive policies that increase the cost of con›ict, in the process overlooking initiatives that may enhance the prospects for peace by eliminating a common (and empirically recognized)30 root cause of war: dissatisfaction with the status quo. As Van Gelder (1989: 163) observes, “It is too often forgotten that [successful deterrence] requires not only that the expected utility of acting be relatively low, but that the expected utility of refraining be acceptably high.” Again, Austria-Hungary’s decision to move against Serbia on July 28, 1914, illustrates the point.
3.3.4. the cost of conflict In both classical deterrence theory and perfect deterrence theory, the costs associated with con›ict have signi‹cant strategic implications. But important 28. Realism, whether classical or neo-, loses much of its explanatory power if only some states are taken to be power maximizers or if only some states are motivated by structural insecurity. Nonetheless, some decision-theoretic deterrence theorists (e.g., Ellsberg 1959) differentiate among actors. 29. Perfect deterrence theory is connected, theoretically, with power transition theory (Organski and Kugler 1980), which sees the international system as hierarchical rather than anarchistic. In a hierarchical system, the dominant state and its allies are generally content with the status quo. Thus the assumption of differentiated actors is not ad hoc in perfect deterrence theory, as it is in most manifestations of classical deterrence theory. For a discussion of the linkage between power transition theory and perfect deterrence theory, see Zagare 1996b, 2007. The connections are theoretically signi‹cant. After all, the Great War was a hegemonic war. As Copeland (2000: 117, 270) observes, in 1914 Germany tried to “take on the system.” 30. For chapter and verse, see Geller and Singer 1998: 64–65, 89–92.
54
the games of july
differences exist in the conclusions the two theoretical frameworks reach about the impact of increased costs on the likelihood of deterrence success. In classical deterrence theory, deterrence success becomes more and more likely as these costs increase. As already mentioned, the monotonic relationship between the costs of con›ict and the probability of deterrence success leads many classical deterrence theorists to recommend an overkill capability. One reason for this straightforward connection between cost and stability is that most classical deterrence theorists assume ‹xed preferences: players always prefer an advantage to the status quo, and they always prefer not to execute their deterrent threat (i.e., threats are always incredible). Thus, any increase in the cost of con›ict always has the same impact relative to other outcomes. In perfect deterrence theory, by contrast, the cost of con›ict is gauged against two other important strategic variables. The ‹rst is the value of the status quo. One consequence of this variable relationship is that in perfect deterrence theory, there is a minimum cost threshold below which deterrence cannot succeed. This is the point separating threats that are capable from those that are not. The second reference point is the value of concession. In perfect deterrence theory, the players may or may not prefer to concede rather than to execute a deterrent threat. As a result, there is also a maximum threshold beyond which further increases in the cost of con›ict do not contribute to the probability of deterrence success. Rather than an overkill capability, then, the logic of perfect deterrence theory is consistent with a policy of minimum deterrence, which rests on a threat that is costly enough to deter an opponent but that is not so costly as to render the threat itself incredible. An equally important difference is that there is no simple monotonic relationship in perfect deterrence theory between the cost of con›ict and the stability of the status quo, as there is in classical deterrence theory. In perfect deterrence theory there are circumstances under which an increase in con›ict costs will undermine a deterrence relationship. More speci‹cally, extended deterrence becomes more and more dif‹cult to maintain as con›ict costs rise, simply because defenders become more and more reluctant to respond to an indirect challenge, which is why, initially at least, decision makers in both Vienna and Berlin believed that Russia would stand aside as Austria-Hungary took vengeance on Serbia. In perfect deterrence theory, therefore, increased con›ict costs can be stabilizing under some circumstances, but may have the opposite consequence under others.
Perfect Deterrence Theory
55
3.3.5. negotiating styles and crisis management techniques Classical deterrence theory and perfect deterrence theory also differ about the best way to approach diplomacy. Recall that decision-theoretic deterrence theorists proffer coercive bargaining tactics that either reduce ›exibility or that increase an opponent’s con›ict costs to increase the probability of an opponent’s concession. By contrast, perfect deterrence theory recommends an approach rooted in reciprocity. Conditionally cooperative strategies such as titfor-tat that reciprocate both cooperation and noncooperation are associated, both theoretically and empirically, with successful compromise. In perfect deterrence theory, establishing reciprocity is tantamount to establishing credibility, which, in turn, makes deterrence more likely. Thus, it is reassuring that in Huth’s (1988) statistical analysis of extended deterrence relationships, ‹rm but ›exible negotiating styles and tit-for-tat deployments are highly correlated with extended deterrence success. Huth de‹nes a ‹rm but ›exible diplomatic stance as a signal that the defender is willing to compromise but not capitulate. And a tit-for-tat policy involves an actual response in kind during a crisis or mobilization. Thus, the essence of both a ‹rm but ›exible bargaining approach and a tit-for-tat response to an actual provocation is reciprocity, the norm that signals credibility when promised or threatened, and demonstrates it when practiced. There is a large empirical literature that is consistent not only with Huth’s ‹ndings but also with the theoretical expectations of perfect deterrence theory about the pervasiveness of reciprocal behavior in interstate interactions.31 This evidence attests to perfect deterrence theory’s explanatory and predictive power. By contrast, this widely observed norm is dif‹cult if not impossible for classical deterrence theorists to explain. In their models, which are based on preference structures derived from Chicken, mutual cooperation and mutual defection can never be part of a (pure strategy) equilibrium. Indeed, in Chicken, each player’s best (i.e., equilibrium) response is always the reverse of the other’s, which is why these models tend to speak to the question of which side can expect to win or lose in a crisis (see, e.g., Powell 1987). Ties, however, which involve reciprocity, are extremely rare events in brinkmanship models, making both war and peace unfathomable. Clearly, the pertinent theoretical puzzle for classical deterrence theory is explaining why peace is so often ob31. For a review, see Cashman 1993: chap. 6; Zagare and Kilgour 2000: chap. 10.
56
the games of july
served, why wars break out at all, and why crises (i.e., challenges) do not occur all the time. In this context, it is instructive to note that Trachtenberg’s analysis of what it would have taken to avert a con›ict in 1914 fully supports perfect deterrence theory’s principal diplomatic policy recommendation. According to Trachtenberg (1990/91: 143), had Germany taken a ‹rm but ›exible approach in 1914, World War I could have been avoided: Russia would not have mobilized, he argues, had German chancellor Theobald von Bethmann Hollweg made it clear to the Russian leadership “that war was not inevitable, that a political settlement was within reach, that Austria could be led to moderate her demands on Serbia, but that he needed a little time to bring her around.”
3.4. Coda Perfect deterrence theory provides a powerful perspective from which to view the dynamics of interstate con›ict avoidance, initiation, limitation, escalation, and resolution. In contrast to classical deterrence theory, it makes consistent use of the rationality postulate, is prima facie in accord with the empirical record, and makes commonsense policy prescriptions that are grounded in strict logic. Unlike classical deterrence theory, perfect deterrence theory makes no particular assumption about the costs of con›ict. It is, therefore, a more general theory, applicable to a much wider range of strategic relationships. In other words, perfect deterrence theory is not simply a divergent theory of nuclear war avoidance. Rather, it is a universal theory of con›ict initiation and resolution, applicable to both nuclear and to nonnuclear interactions. It is consequently well situated to provide a theoretically informed explanation not only of the long Cold War peace that characterized the international system after World War II, but also of the system’s breakdown in July 1914. Two explanations for the absence of war between the United States and the Soviet Union after 1945 are logically consistent with the strictures of perfect deterrence theory. One is that the system avoided a serious challenge because it was dominated by a satis‹ed superpower that was simply unwilling to wage a preventive war for marginal gains. The other is that the large Soviet land force that dominated the European continent throughout most of the postwar period implied war costs that were high enough so that both the Soviet Union and the strategically superior United States possessed retaliatory threats that
Perfect Deterrence Theory
57
were at once capable and credible. It remains an important empirical puzzle whether the long peace was an instance of unilateral or mutual deterrence. In the case of the Great War, it is not dif‹cult to derive an explanation of its outbreak in terms of perfect deterrence theory’s principal variables: a highly dissatis‹ed major power, Austria-Hungary, attacked a minor power, Serbia, not because the Serbian threat was devoid of credibility, but because it lacked capability. In the eyes of Austria’s leaders, nothing less than the monarchy was at stake. Whether it was the result of cold calculation, poor intelligence, false inference, or wishful thinking, whether it was politically driven, psychologically motivated, or the consequence of cognitive impairment, the Austrians’ belief that Russia would not intervene led them to conclude that the costs of doing nothing were intolerable. But this explanation, too, falls short of the mark, not because it is wrong but because it is incomplete. It leaves unanswered the most vexing questions about World War I, which concern not how it started but why what could have been a local war between Serbia and Austria-Hungary ‹rst escalated into a continental war that included Germany, Russia, and France and subsequently expanded into a wider world war when Great Britain’s efforts to deter a German attack on Belgium failed. The remainder of this book provides a logically consistent answer to these empirical and theoretical puzzles. Before attending to these questions in chapters 6 and 7, however, a number of subsidiary issues will be addressed. In chapter 4, I examine Germany’s 1879 decision to enter into a defensive alliance with Austria-Hungary, its former enemy and longtime rival. Bismarck’s original decision, in turn, provides the context for the explanation, offered in chapter 5, of the blank check his successors issued to Austria shortly after the assassination of Franz Ferdinand in June 1914. Bismarck’s 1879 decision to align Germany with Austria is theoretically perplexing. At the time it was negotiated, an alliance between the two continental competitors was considered unlikely. In fact, Bismarck had long resisted Austria’s previous requests to move into closer alignment. In addition, the terms of the alliance fully re›ected Austria’s rather than Germany’s terms. Why, then, did Bismarck suddenly reverse himself and enter into an agreement that, even in retrospect, is startling, an alliance that the kaiser opposed and that ostensibly offered Germany few defensive bene‹ts? Chapter 4 provides an explanation from within the theoretical con‹nes of perfect deterrence theory.
part two
Explaining the Great War
All politics reduce themselves to this formula: try to be à trios in a world governed by ‹ve powers. otto von bismarck
chapter four
Bismarck’s System
The European state system was in crisis for most of July 1914, but only on August 4, when German troops, heading toward the French border, invaded Belgium, did the system’s collapse become all but inevitable. For the ‹rst time since 1870, a war between the great powers of Europe was about to break out. Great Britain’s foreign secretary, Sir Edward Grey, was among those who recognized the extent of the looming tragedy. “The lamps are going out all over Europe,” he remarked. “We shall not see them lit again in our lifetime.” Just a few years earlier, however, the possibility of a general European war was considered remote. Some liberals, including Sir Norman Angell (1910), even believed that major power war was an anachronism, an outmoded vestige of an earlier and less “rational” era. Those who thought war unlikely (or irrational) had history on their side. Like the bipolar system that conditioned the relationship of the United States and the Soviet Union during the Cold War, the multipolar system that fell apart in 1914 had been “remarkably stable” (Waltz 1964). To be sure, there were some close calls. General war in Europe was a real possibility in 1875, 1878, 1885, 1887, 1898, 1906, 1908, 1911, and 1912 (Seaman 1963: 9). Yet no such war had occurred. Somehow, someway, the system always seemed to survive. That there had been no major-power con›ict in Europe from 1870 until 1914 was not entirely a matter of luck. Rather, the absence of a general European war during that time frame testi‹es to the political genius of one man, Otto von Bismarck, who ‹rst fashioned the system and later, manipulated it, all to the advantage of the king of Prussia. When Bismarck came to power in 1862 as its minister-president, Prussia was on the cusp of great-power status. Great Britain, of course, ruled the waves. France was the dominant land force on the continent. And Russia, with 61
62
the games of july
its massive population, constituted a latent threat to all European states. Even in Germany, it was not Prussia but Austria that was ascendant. The situation changed completely over the course of a few years. In 1864, in union with Austria, Prussia went to war with Denmark over control of Schleswig and Holstein. A quick victory over Denmark led, temporarily, to a joint Austro-Prussian condominium over the two duchies. Then, in June 1866, Bismarck used a disagreement with Austria over the political fate of the former Danish possessions as a pretext for a second war. Within seven weeks, Austria, too, was defeated. After the brief war with Austria, Prussia absorbed Schleswig, Holstein, and a number of small German states allied with Austria. The Austrian-dominated German Confederation was dissolved, with the North German Confederation, under Prussian leadership, taking its place. In the end, Austria was treated leniently, losing only Venice to Italy, Prussia’s ally of the moment. As he seemed to realize at the time, Bismarck had important strategic objectives that required a continuing relationship with his erstwhile ally. In 1870, Bismarck instigated yet another war, this time with France. Like Austria, France would fall quickly. But even before the ‹ghting was over, while Paris remained under siege, the Second German Empire was declared in Louis XIV’s palace at Versailles.1 The new empire united into a single political unit the military coalition of North and South German states that Bismarck had assembled to ‹ght the French. The confederation’s constitution stipulated that the king of Prussia, Wilhelm I, would become emperor (or kaiser), and that the minister-president of Prussia, Bismarck himself, would be Imperial Germany’s ‹rst chancellor. By 1871, then, the broad outlines of the Bismarckian system were in place. Five great powers lorded over the central state system: a weakened and embittered France, a humbled Austria-Hungary, an apprehensive but nonetheless detached Britain, a restless Russia, and a uni‹ed and, according to Bismarck, fully satiated Germany. With the formation of the empire, Bismarck had achieved his most pressing strategic objective. Henceforth, his foreign policy would be crafted primarily to consolidate this, his signal achievement. Under Bismarck, therefore, the newly forged German state began its life as a satis‹ed, status quo power. By contrast, France was at the top of the list of the discontented or dissatis‹ed nations. Unlike Austria—and against Bismarck’s counsel—France 1. The Holy Roman Empire was the ‹rst.
Bismarck’s System
63
was treated harshly after its defeat. Not only did France lose the provinces of Alsace and Lorraine, but it remained occupied until 1873, when it paid off a huge war indemnity. Correctly anticipating continuing French resentment, Bismarck knew that he would have to keep France isolated. His formula for doing so was exceedingly simple: in a system governed by ‹ve great powers, he observed, “try to be à trios.” Constructing a stable coalition of three major powers would not be easy, however. In the wake of France’s stunning defeat, any understanding with the country was simply out of the question. Similarly, alignment with Great Britain was unlikely; British policy was splendidly isolationist. This left only Austria and Russia among the major powers as potential allies. Of the pair, Russia was the key. Only Russia, especially were it to act in concert with France, posed a potential threat to the new German Empire. Bismarck’s ‹rst attempt at building a conservative coalition, the Three Emperors’ League (Dreikaiserbund), failed. This all-but-toothless 1873 security arrangement required only that Europe’s three major autocratic powers consult each other whenever the European status quo was threatened. In 1878, just such a dispute broke out between Austria and Russia over the disposition of territories that fell to Russia after its war with Turkey in 1877. The dispute was not so easily resolved. Bismarck refused to take sides, opting instead to play the role of an “honest broker” at the Congress of Berlin, where the “Eastern Question” was addressed. In consequence, the marginalized Three Emperors’ League collapsed. The Treaty of Berlin left Russia resentful. Russia’s signi‹cant gains at Turkey’s expense, as codi‹ed in the Treaty of San Stefano, were virtually nulli‹ed. Since Bismarck had presided over the congress, Russia blamed him, but only in part. From the Russian point of view, Austria had now become the principal nemesis (Langer 1950: 165). After all, the international conference in Berlin that humiliated Russia had been the idea of Austria’s foreign minister, Count Julius Andrássy. And Austria stood in the way of Russian ambitions in the Balkans. Russia’s lingering rivalry with Austria clearly threatened Bismarck’s plans. Bismarck thought the Balkans to be without strategic import, “not worth the bones of a single Pomeranian grenadier.” Yet he feared that Russia would press for concessions in the region. If Austria resisted, Germany might be drawn into a war with Russia. And if the pressures were not resisted, Austria might be pushed into an anti-German alliance with Russia or even with France (Langer 1950: 180). Neither outcome was in Germany’s interests. To help stabilize the
64
the games of july
status quo, then, Bismarck offered Austria an alliance—on terms that favored Austria. After a short period of negotiations in the fall of 1879, Austria accepted. But why was an alliance necessary? And why did Bismarck, who preferred a vague defensive arrangement, agree to an alliance directed against Russia only? This chapter addresses these and related questions in the context of a three-player crisis bargaining model, the Tripartite Crisis Game (Zagare and Kilgour 2003), that constitutes one of perfect deterrence theory’s component games.2
4.1. The Tripartite Crisis Game The Tripartite Crisis Game, given in extensive form in ‹gure 4.1, was speci‹cally developed to capture the strategic dynamic that conditioned the Austro-German discussions in 1879 (and subsequent renegotiations of the Dual Alliance until 1914).3 This fact, however, does not mean that the insights that can be drawn from an analysis of its properties are limited to any speci‹c historical event or relationship. Indeed, the Tripartite Crisis Game is a generic model of crisis bargaining and extended deterrence that can be used to generate theoretical knowledge about an important class of real-world events. Thus, while I use the Tripartite Crisis Game in this chapter to gain deeper insight into Bismarck’s motivation for negotiating a highly contingent defensive agreement with Austria on terms unfavorable to Germany, I use it for somewhat different purposes in subsequent chapters. Speci‹cally, in chapter 5, I rely on one of its proper subgames, the Protégé-Defender Subgame, to offer an explanation of the blank check given to Austria-Hungary by German leaders shortly after the assassination of Franz Ferdinand in Sarajevo. And in chapter 7, I use a special case analysis of the Tripartite Crisis Game to evaluate British foreign policy in the run-up to the Great War. Because the Tripartite Crisis Game model is central to the analysis of so many aspects of the run-up to the Great War, I examine its logical structure in more detail than is immediately required. I ‹rst use the Tripartite Crisis Game model to generate theoretical knowledge about certain types of extended deterrence relationships; I then use this knowledge to gain insight into speci‹c
2. For a related model, see Quackenbush 2006b. 3. See Zagare and Kilgour 2003, which forms the basis of the remainder of this chapter.
Bismarck’s System
Node 1
Challenger Not Demand (1 – x)
65
Demand (x)
Protégé
Status Quo
Concede (1 – y)
Node 2
Hold Firm (y)
Challenger Wins Not Support (1 – z)
Protégé
Node 4a
Realign
Protégé Realigns
Node 3
Defender Support (z)
Node 4b
Not Realign
Protégé Loses
Challenger
Back Down
Challenger Concedes
Press On
Conflict
Fig. 4.1. Tripartite Crisis Game
cases.4 I assume that particular knowledge can be generated most reliably in the context of more general (theoretical) categories. Only by knowing the general can we fully understand and explain the particular. The game has three players, Challenger, Defender, and Protégé, and the distinct roles and divergent strategic interests of these players correspond closely to those of Russia, Germany, and Austria, respectively, in 1879. Did they not correspond, the Tripartite Crisis Game would not be an appropriate model for analyzing their strategic relationship at that time. 4. But the empirical relevance of the Tripartite Crisis Game is not restricted to these cases. On its face, the model would also seem to apply, inter alia, to the extended deterrence relationships that existed in 1914 between Russia and Serbia (Joll 1992: 125), France and Russia, and Great Britain and France (see chapter 6) and to the presentday tripartite extended deterrence relationship of the United States, Taiwan, and China.
66
the games of july
The Tripartite Crisis Game begins with Challenger’s move at decision node 1. Challenger (Russia) initiates play by deciding whether to demand a concession from Protégé. If no demand is made, the game ends and the outcome called Status Quo obtains. But if a demand is made, a crisis occurs. In this contingency, Protégé (Austria) decides at decision node 2 whether to concede or to hold ‹rm. Concession leads to outcome Challenger Wins, holding ‹rm to a critical choice for Defender (Germany) at decision node 3, whether to support Protégé. Defender’s choice forces either Protégé or Challenger to make a subsequent decision. If Defender stands by Protégé, Challenger either presses on or backs down, leading to Con›ict or Challenger Concedes, respectively. If Defender abandons Protégé, Protégé is forced to reach an accommodation with Challenger.5 In the wake of its diplomatic defeat, however, Protégé may also reconsider its alignment status, choosing whether to maintain its now tenuous relationship with Defender (as Serbia did in 1909 after Russia withdrew its support during the Bosnian Crisis or as Austria did after it was forsaken by Germany in the 1913 Balkan Crisis) or to ‹nd a more reliable partner by realigning, perhaps even with Challenger. The outcome is Protégé Loses in the ‹rst instance and Protégé Realigns in the second. In either case, the concession to Challenger is the same. But when Protégé Realigns, Defender loses an important strategic partner. Fear of this outcome drove much of German foreign policy in 1879 and afterward.6 The Tripartite Crisis Game provides a rich theoretical framework for examining extended deterrence relationships. Nonetheless, the Tripartite Crisis Game model remains a simpli‹cation of the crisis bargaining process in some multipolar systems or subsystems. The most obvious simpli‹cation concerns the number of players: there are only three. Larger and more complex games are sometimes the rule, not the exception, in international politics. In Bismarck’s time, for example, ‹ve great powers dominated the European state system. Moreover, the choices the players make at the various decision nodes are dichotomized. The Tripartite Crisis Game could be formulated to include additional players, more nuanced options, or more differentiated outcomes. It is not immediately clear, however, that the political dynamic that is explored here and later in both chapters 5 and 7 would be made more transpar5. Implicit here is the assumption that Challenger is known to be stronger than Protégé, so that a confrontation would be unfavorable to an unprotected Protégé. 6. During the 1908–9 Bosnian Crisis, for example, both German chancellor Bernard Bülow and Friedrich von Holstein, a key foreign policy adviser in the German Foreign Of‹ce, feared that Austria would tilt toward England and France if it were not supported in its dispute with Russia. In consequence, “Bülow and Holstein felt they had no choice but to support Austria in its bid to annex Bosnia and Herzegovina” (Mercer 1996: 125; see also Kagan 1995: 162–63.)
Bismarck’s System
67
ent by extending the Tripartite Crisis Game to include additional players or more re‹ned choices. Rather than unduly complicating the model, then, the consequences of these and related re‹nements are assumed to be captured by the players’ utility assessments. For example, it is implicit in the subsequent analysis that each player’s utility for Con›ict takes into account the possibility that a war could expand. It is also assumed, inter alia, that the players’ utilities for Protégé Realigns re›ect their best assessment of this outcome’s wider strategic implications. The extended deterrence relationship that is modeled by the Tripartite Crisis Game, therefore, is deliberately distorted to gain both tractability and insight into its essential dynamic. While some readers may see this as a limitation of formal analysis, I see it as a strength: conceptual simpli‹cation allows important processes to be highlighted and less important aspects to be put to the side so that the most critical determinants of strategic behavior can be penetrated and understood. Of course, it is always a judgment call whether any simpli‹cation has achieved its purpose or whether theoretical parsimony has been replaced by gross oversimpli‹cation.
4.2. Outcomes and Preferences A game is de‹ned by both the rules that govern play and the players’ preferences over the set of possible outcomes. The game tree of ‹gure 4.1 succinctly captures almost all the rules associated with the Tripartite Crisis Game model, including the identi‹cation of the players, the choices available to them at every decision point, the sequence of play, and the set of outcomes that their choices can bring about. But since ‹gure 4.1 is silent on the question of player preferences, it does not yet fully constitute a game. Since there are three players and six outcomes in the Tripartite Crisis Game model, there are many different preference combinations, or distinct games, that can be associated with it. Of course, not all of these combinations are strategically or theoretically interesting; of those that are, some are empirically inconsistent with the circumstances that Bismarck confronted in 1879, while others are not compatible with standard assumptions about the importance of alignment patterns on crisis bargaining behavior.7 The preference assump-
7. In neorealist theory, for example, alignment is considered one of the two principal constraints on state behavior; system structure is the other. Alignments are important because they “supplement structure” by specifying the relationship of a system’s units (G. Snyder 1997: 22; see also Waltz 1979: 80). Alignments refer to the “expectations of states about whether they will be supported or opposed by other states in future interactions.” Alignment, then, is a belief or “a state of mind that in›uences, or may be in›uenced, by interaction” (G. Snyder 1997: 6, 22).
68
the games of july
TABLE 4.1. Initial Preference Assumptions for Tripartite Crisis Game Challenger
Protégé
Defender
Challenger Wins
Status Quo
Status Quo
Protégé Realigns
Challenger Concedes
Challenger Concedes
Protégé Loses
Conflict
Challenger Wins
Status Quo
Challenger Wins
Protégé Loses
Conflict or Challenger Concedes
Protégé Loses or Protégé Realigns
Conflict or Protégé Realigns
tions arrayed in table 4.1 are the exception. An analysis of the Tripartite Crisis Game in the context of these (posited) preference assumptions, therefore, provides a ‹rm base for understanding not only Bismarck’s motivation for aligning Imperial Germany with Austria-Hungary shortly after the Congress of Berlin, but also the general motivation of states to adjust their alignment policies in very speci‹c ways under the most challenging of strategic circumstances. In other words, I analyze the Tripartite Crisis Game with preference assumptions that encapsulate both an enduring strategic conundrum and a representative historical instance of it. In the end, both general theoretical knowledge and case-speci‹c insights are obtained. The columns of table 4.1 list the postulated ranking of the players’ preferences over the six outcomes from best to worst. For example, the assumption is that Challenger most prefers Challenger Wins, next-most prefers Protégé Realigns, and so on. No ‹xed preference assumption is made for outcomes contained in the same cell of the table. Thus, in what follows, Challenger could prefer Con›ict to Challenger Concedes or the reverse. The players’ relative preferences for these paired outcomes are the crucial explanatory variables of the model.
4.2.1. challenger’s preferences In the Tripartite Crisis Game, Challenger’s primary objective, as Bismarck saw Russia’s in 1879, is to obtain concessions from Protégé. Three outcomes of the Tripartite Crisis Game offer Challenger a clear improvement of the Status Quo. I take Challenger’s preference for these outcomes over the others as a given.8 Of these, I assume that 8. Crisis bargaining games played when this assumption is not satis‹ed are at once historically uninteresting and strategically trivial. In other words, deterrence is irrelevant when none of the parties involved is dissatis‹ed or is seen to be dissatis‹ed with the status quo.
Bismarck’s System
69
Challenger Wins ™Ch Protégé Realigns ™Ch Protégé Loses, where ™Ch means “is preferred to by Challenger.” Challenger Wins brings immediate grati‹cation and is relatively costless. Protégé Realigns humiliates Defender. Protégé Loses does not. Challenger’s postulated preference for Protégé Realigns over Protégé Loses means that Protégé’s alignment policy matters, not only to Defender, but also to Challenger.9 As demonstrated shortly, this seemingly innocuous assumption has important strategic implications. Of the remaining three outcomes, it is assumed that Challenger prefers the Status Quo to both Con›ict and Challenger Concedes. This assumption guarantees that Defender’s threat to support Protégé is capable.10 As discussed previously, threat capability is a necessary condition for deterrence success; without it, the Status Quo never survives in rational play, and (extended) deterrence always fails.11 Challenger’s preference between its two lowest-ranked outcomes, Challenger Concedes and Con›ict, determines its type. Challengers that prefer Con›ict to Challenger Concedes are called determined. Challengers with the opposite preference are termed hesitant.
4.2.2. protégé’s preferences Protégé’s presumed objective is to avoid acceding to Challenger. Among the three outcomes that do not involve concession, I assume that Status Quo ™Pro Challenger Concedes ™Pro Con›ict, where ™Pro means “is preferred to by Protégé.” Status Quo involves no loss or evident cost, whereas when Challenger Concedes, Protégé incurs an obligation to Defender. With Con›ict, on the other hand, there are additional costs as well as the risk of an unfavorable outcome. 9. The assumption that Challenger prefers Protégé Realigns to Protégé Loses is a simpli‹cation that does not hold all the time. For example, neither Britain nor France was interested in pursuing what G. Snyder (1997: 337–38) calls a “divide and rule” policy in the years leading up to World War I. Both Grey and President Raymond Poincaré of France believed that separating Austria-Hungary from Germany would likely provoke Germany. By contrast, in 1905, during the ‹rst Moroccan Crisis, Germany hoped to break up the Entente Cordiale by pressuring France for concessions in Morocco (Massie 1991: 363). 10. Recall that threat capability is de‹ned as the ability to hurt. Defender’s threat to support Protégé, then, is capable if Challenger prefers that the threat not be carried out. 11. The model, therefore, applies only when it is possible to deter Challenger, either because Challenger’s dissatisfaction with the status quo is relatively low or because the costs of con›ict are prohibitively high.
70
the games of july
Of the three outcomes that involve an outright loss, Protégé prefers Challenger Wins on the grounds that an immediate concession is less costly than a later concession made under duress. Protégé’s relative preference between the remaining two outcomes determines its type. A Protégé that least prefers to realign (i.e., prefers Protégé Loses to Protégé Realigns) is loyal; a Protégé with the opposite preference is disloyal.
4.2.3. defender’s preferences Defender is stuck between a rock and a hard place. Ceteris paribus, it would like to avoid Con›ict and preserve its bond with Protégé. The rub, of course, is that Defender might have to threaten to ‹ght to save the relationship. Structural deterrence theorists argue that this strategic conundrum is especially acute in multipolar systems, such as the system that existed in Europe in 1879, where multiple alignment options exist but where the costs associated with the loss of a strategic partner are also inordinately high.12 Of course, neither cost is incurred when the Status Quo prevails. The same is true when Challenger Concedes. But since deterrence is taken to be Defender’s primary motivation, I assume that Status Quo ™Def Challenger Concedes, where ™Def means “is preferred to by Defender.” This analysis of the Tripartite Crisis Game also takes as axiomatic that Challenger Concedes ™Def Challenger Wins ™Def Protégé Loses ™Def Protégé Realigns. A common negative shared by the latter three outcomes is indirect: the concession that Protégé is pressured to make to Challenger. At Challenger Wins, however, Defender’s per‹diousness is not transparent. At Protégé Loses, Defender’s type is apparent, but at least its relationship with Protégé, albeit weak-
12. Conventional neorealist wisdom holds that bipolar systems (i.e., systems with only two great powers) are characterized by rigid alignment patterns and ›exible policy options. Alignment patterns are rigid because realignment options are limited; policy is ›exible because disgruntled parties have nowhere to turn. By contrast, multipolar systems, with three or more great powers, are characterized by ›exible alignment patterns but constrained policy options. Alignment patterns are ›uid because state options are plentiful; still, policy is restricted because discontented partners can more easily realign (Waltz 1979: 169–70).
Bismarck’s System
71
ened, is maintained. At Protégé Realigns, Protégé defects, and Defender incurs a signi‹cant strategic cost. There are a number of reasonable possibilities for Defender’s evaluation of Con›ict relative to the other outcomes. Not all are interesting or theoretically important. To enhance generality, I focus on those that are. When Defender is always willing to support Protégé—that is, when Con›ict ™Def Protégé Loses—there is no tension between Defender’s goal of avoiding con›ict and retaining Protégé’s allegiance.13 To accentuate this tension, I assume that Protégé Loses ™Def Con›ict.14 But no ‹xed assumption is made about Defender’s relative ranking of Con›ict and Protégé Realigns. Defenders preferring Con›ict to Protégé Realigns are called staunch; those with the opposite preference are called per‹dious. To summarize, the version of the Tripartite Crisis Game model analyzed in this chapter is de‹ned by the game form of ‹gure 4.1, the preference restrictions listed in table 4.1, and the type designations given in table 4.2. The Tripartite Crisis Game is a true three-person game in that all three players face dynamically interactive choices. No player can rationally ignore another’s actions. To be sure, Protégé’s decision at node 4a and Challenger’s decision at node 4b can be made without reference to the preferences and the choices of other players. But Protégé’s best choice at node 2 depends on what it expects of Defender at node 3, which in turn depends on Defender’s estimates of both Protégé’s choice at node 4a and Challenger’s at node 4b. Similarly, Challenger’s TABLE 4.2. Player Type Designations Type/Player
Preference
Determined Challenger
Conflict ™Ch Challenger Concedes
Hesitant Challenger
Challenger Concedes ™Ch Conflict
Loyal Protégé
Protégé Loses ™Pro Protégé Realigns
Disloyal Protégé
Protégé Realigns ™Pro Protégé Loses
Staunch Defender
Conflict ™Def Protégé Realigns
Perfidious Defender
Protégé Realigns ™Def Conflict
13. When Defender prefers Con›ict to Protégé Loses, it always chooses to support Protégé at node 3. In consequence, Protégé never faces a choice at node 4a. 14. Defender’s preference for Challenger Wins and Protégé Loses over Con›ict implies that Defender is not heavily invested in the issues at stake. The assumption is consistent with Bismarck’s take on the strategic situation that existed in the Balkans in 1879. Extended deterrence relationships with intrinsically valuable stakes have a fundamentally different strategic dynamic than the one modeled herein. For an innovative investigation of the role of high stakes in extended deterrence relationships, see Danilovic 2002.
72
the games of july
best decision at node 1 and Defender’s at node 3 depend on each player’s expectations of both of the other players. Parenthetically, it should be mentioned that the Tripartite Crisis Game was designed to capture, in a single formal structure, elements of G. Snyder’s (1997: 37) “alliance” and “adversary” games.15 This theoretical integration is important because it permits a more thorough understanding of the “deterrence versus restraint dilemma” (G. Snyder and Diesing 1977: 438) that Imperial Germany faced, not only in 1879 but in 1914 as well. To deter Challenger, Defender is motivated to pledge strong support for Protégé.16 But stronger support also increases the probability that Protégé will hold ‹rm if a demand is received, risking a con›ict that Defender would prefer to avoid. Signi‹cantly, Protégé also faces a strategic dilemma in the Tripartite Crisis Game: the more credible Protégé’s threat to realign, the more likely, ceteris paribus, the status quo is to survive because of the greater probability that Defender will support Protégé at node 3 and that Protégé, expecting this support, will hold ‹rm against a demand. Of course, disloyalty is a double-edged sword, and the prospect of dislodging Protégé from Defender’s orbit may, under certain conditions, prompt Challenger to instigate a crisis. These conditions are discussed in the analysis that follows.
4.3. Analysis I now turn to a general discussion and summary evaluation of the strategic properties of the Tripartite Crisis Game with incomplete information. In the informal analysis that follows, all relevant information is assumed to be common knowledge, except that the players may be uncertain about each others’ type.17 Speci‹cally, the players are assumed to be fully informed about the game de‹ned by the rules of play, as re›ected in the game tree of ‹gure 4.1, and the preference orderings insofar as they are given in table 4.1. Defender and Protégé are also assumed to possess private information about their type. As far as all the players know, Defender may be either staunch or per‹dious, and Protégé may be either loyal or disloyal. But to keep things 15. For a more detailed discussion of these games, see section 7.1. 16. Left open is precisely how Defender may signal its support. Depending on circumstances, Defender could, inter alia, back Protégé at an international conference, as Britain did with France during the 1906 Algeciras Conference that marked the end of the ‹rst Moroccan Crisis; mobilize its army or put its navy on alert in a way that Challenger would be forced to choose between war and peace; or make a visible and very public commitment to defend Protégé. For an innovative listing of “commitment tactics,” see G. Snyder 1972. 17. For a formal analysis, see Zagare and Kilgour 2003.
Bismarck’s System
73
simple and to make things interesting, I further assume that Challenger is known to be determined—that is, that all the players know that Challenger prefers Con›ict to Challenger Concedes. To assume that Challenger’s type is unknown to the other players would unduly complicate the analysis. And to assume that Challenger is known to be hesitant would render the Tripartite Crisis Game theoretically trivial. Regardless of the information structure of the game, the Status Quo is the only outcome that can be supported at any form of strategic equilibrium when Challenger’s preference for Challenger Concedes over Con›ict is common knowledge.18 Table 4.3 shows this to be the case if and when information is complete. Table 4.3 also shows that at least under complete information, Protégé must be disloyal for the Status Quo to survive rational play against a determined Challenger. In other words, Protégé’s preference for Protégé Realigns over Protégé Loses is a necessary condition for extended deterrence success whenever all the players know each other’s preferences. As will be shown below, however, when information is incomplete, a high likelihood that Protégé is disloyal is neither necessary nor suf‹cient for the survival of the Status Quo. In fact, under certain conditions, a higher probability that Protégé is disloyal undermines the stability of the Status Quo. To specify these conditions, however, the set of perfect Bayesian equilibria in the Tripartite Crisis Game (with incomplete information) must be identi‹ed ‹rst. (As discussed in chapter 1, the concept of an equilibrium strategy combination [outcome] is the basis of all solution concepts in noncooperative game theory, and a perfect Bayesian equilibrium is the accepted standard TABLE 4.3. Outcomes Associated with Subgame Perfect Equilibria in Tripartite Crisis Game
Protégé: Determined Challenger: Hesitant
Defender: —Staunch— —Perfidious— Loyal Disloyal Loyal Disloyal Challenger Wins
Status Quo
Status Quo
Status Quo
Challenger Challenger Wins Wins Status Quo
Status Quo
18. In other words, it is not dif‹cult at all to explain deterrence success when Challenger is known to be hesitant. This assumption about Challenger’s type will be relaxed in chapter 7, where the goal is to explain the failure of British policy to deter a German attack on Belgium and France in August 1914.
74
the games of july
of rational behavior in dynamic [i.e., extensive-form] games with incomplete information.) A perfect Bayesian equilibrium speci‹es an action choice for every type of every player at every decision node or information set belonging to the player;19 it must also indicate how each player updates its beliefs about other players’ types in light of new information obtained as the game is played out.20 The set of perfect Bayesian equilibria constitutes the empirical expectations (i.e., the predictions) of the Tripartite Crisis Game model with incomplete information. When the behavior of real-world counterparts of the players in the model is found to be consistent not only with the action choices associated with one of the equilibria but also with the set of conditions that give rise to it, an explanation is derived (Riker 1990). The explanation will be both straightforward and natural: given the players’ preferences, however wise or foolish they may have been, and their perceptions (i.e., beliefs) about one another, however accurate or inaccurate they proved to be, the players’ (actual) behavior was an instrumentally rational response to the strategic problem they (collectively) encountered. A correct prediction is also a strong indication that the underlying game model is both theoretically relevant and empirically viable. Even a nontechnical description of the perfect Bayesian equilibria of the Tripartite Crisis Game requires a few additional assumptions and a bit of new notation. To model the consequences of the players’ uncertainty about Defender’s and Protégé’s types, assume now that Defender is staunch with probability pDef (where 0 < pDef < 1), that Protégé is disloyal with probability pPro (where 0 < pPro < 1), and that the values of pDef and pPro are known to all the players.21 Of course, each player is assumed to be fully aware of its own type (or preferences). The simplifying assumption that all the players know Challenger’s type leads to an important observation: the outcome of the Tripartite Crisis Game will always be Con›ict whenever Challenger is determined and the play of the 19. For further discussion and instructive examples, see Morrow 1994: chaps. 6–8. 20. In an extensive-form game of incomplete information, the players’ initial (or a priori) beliefs are taken as givens. The de‹nition of a perfect Bayesian equilibrium, however, places no restriction on the players’ updated (or a posteriori) beliefs “off the equilibrium path”—that is, on beliefs at nodes that are never reached under rational play. A perfect Bayesian equilibrium is sometimes supported by a posteriori beliefs that are inconsistent with a player’s a priori beliefs. Perfect Bayesian equilibria based on internally inconsistent beliefs are implausible. In consequence, they will not be considered as rational strategic possibilities in this and subsequent chapters. 21. The two belief variables, pDef and pPro, can be thought of as measures of the credibility of Defender’s and Protégé’s threats, respectively.
Bismarck’s System
Challenger
Not Demand
75
Node 1
Demand
Protégé
Status Quo
Concede (1 – y)
Node 2
Hold Firm (y)
Defender
Challenger Wins
Node 3
(q)
ProtégéDefender Subgame
Not Support (1 – z)
Node 4a
Conflict
Protégé
Realign
Protégé Realigns
Support (z)
Not Realign
Protégé Loses
Fig. 4.2. Tripartite Crisis Game when Challenger is determined
game reaches node 4b.22 This information enables us to simplify the game form. Figure 4.2 illustrates the reduced game and highlights a proper subgame,23 the Protégé-Defender Subgame. To solve the Tripartite Crisis Game, this proper subgame must ‹rst be solved. Then Challenger’s node 1 choice is considered in the context of the subgame solutions. To be sure the Protégé-Defender Subgame is important and interesting in its own right. Indeed, since it models a situation of extended immediate deter22. Determined Challengers strictly prefer Con›ict to Challenger Concede and, hence, never rationally back down at this decision point. 23. A proper subgame is any part of an extensive-form game that can be considered a game unto itself.
76
the games of july
rence in which the status quo has already partly broken down, I use it in chapter 5 to gain insight into the Austro-Serbian stage of the July Crisis. Nonetheless, I do not dwell on its strategic dynamic here. Rather, for now I focus strictly on the Tripartite Crisis Game, which should be thought of as a model of extended general deterrence.24 I use this more elaborate model to uncover Bismarck’s motivation for entering into what was considered an unlikely an alliance with Austria-Hungary in 1879.
4.3.1. the protégé-defender subgame To begin the analysis, observe that Protégé’s choice at node 4a is strictly determined by its type. A disloyal Protégé always realigns; a loyal Protégé never does. Note as well that before Defender can make a choice at node 3, it will observe Protégé’s prior action choice at node 2. Defender can use this new information to reassess Protégé’s type (and likely choice) at node 4a. Since Defender’s choice at node 3 depends, in part, on its assessment of Protégé’s type, the additional information it obtains is useful. By contrast, any observations that Protégé makes before choosing at node 4a will be beside the point. As noted, Protégé’s choice at this decision node is strictly determined by its type. Given these considerations, it follows that a perfect Bayesian equilibrium of the Protégé-Defender Subgame will consist of a ‹ve-tuple of probabilities [yD, yL; zS , zP , q] where: yD = the probability that a disloyal Protégé will choose to hold ‹rm at node 2 yL = the probability that a loyal Protégé will choose to hold ‹rm at node 2 zS = the probability that a staunch Defender will choose to support Protégé at node 3 zP = the probability that a per‹dious Defender will choose to support Protégé at node 3 q = Defender’s updated probability that Protégé is disloyal, given that Protégé holds ‹rm at node 2. The ‹rst four probabilities are strategic variables describing Protégé’s and Defender’s choices, contingent on their type. The ‹fth probability is the a posteriori probability, updated by Defender after Protégé’s choice to hold ‹rm at node 2 has been observed.25 24. For the distinction, see Huth 1988: chap. 1. For related models of extended deterrence, see Smith 1998; Werner 2000. 25. Equilibrium values for the belief variable, q, and certain other technical characteristics of the perfect Bayesian equilibria will be discussed only when relevant.
Bismarck’s System
77
There are four nontransitional perfect Bayesian equilibria in the ProtégéDefender Subgame: Settlement, Separating, Hold Firm, and Bluff.26 Table 4.4 summarizes their technical properties.27 As previously mentioned, the behavioral patterns associated with these equilibria are crucial determinants of Challenger’s node 1 choice and, consequently, of the overall dynamic of the Tripartite Crisis Game. Of the four plausible perfect Bayesian equilibria, one is labeled Settlement.28 At any Settlement Equilibrium, Protégé always concedes at node 2 (i.e., yD = yL = 0). In consequence, when a Settlement Equilibrium is in play, the outcome is always Challenger Wins. Anticipating abandonment by Defender, Protégé simply accepts defeat—as Serbia did in 1908 when it was forsaken by Russia. The Separating perfect Bayesian equilibrium separates the players by type: a disloyal Protégé always holds ‹rm at node 2; a loyal Protégé always concedes (i.e., yD = 1, yL = 0). Similarly, a staunch Defender always supports Protégé at node 3; a per‹dious Defender never does (i.e., zS = 1, zP = 0). The outcome of the Protégé-Defender Subgame under a Separating Equilibrium, then, is Challenger Wins when Protégé is loyal; Con›ict when Protégé is disloyal and Defender is staunch; and Protégé Realigns when Protégé is disloyal and Defender is per‹dious. TABLE 4.4. Plausible Perfect Bayesian Equilibria and Existence Conditions of the Protégé-Defender Subgame with Incomplete Information Strategic and Belief Variables Protégé
Defender
Existence Conditions
Equilibrium
yD
yL
zS
zP
q
Settlement
0
0
1
0
>d1
Separating
1
0
1
0
1
Hold Firm
1
1
1
0
pPro
pDef > e1 pPro > d1
Bluff
1
•
—
0
d1
pDef > e1 pPro < d1
pDef < e2 e2 < pDef < e1
Note: • = fixed value between 0 and 1; — = value not fixed although some restrictions apply.
26. One transitional equilibrium is ignored. An equilibrium is transitional if it exists only when the parameters of a model satisfy a speci‹c functional relationship (i.e., an equation). The justi‹cation for ignoring transitional equilibria is that however the parameter values are obtained, they are very unlikely to satisfy any speci‹c equation. 27. Not all of these properties are pertinent. Those that are not will be disregarded. For a full discussion and additional details, see Zagare and Kilgour 2003. 28. There are two forms of the Settlement equilibrium, but since one form is based on implausible beliefs, it is ignored. For the particulars, see Zagare and Kilgour 2003: 612.
78
the games of july
Under the Hold Firm perfect Bayesian equilibrium, Protégé always holds ‹rm at node 2 (i.e., yD = yL = 1); hence its name. Staunch Defenders always support Protégé; per‹dious Defenders always withhold support (i.e., zS = 1, zP = 0). Thus, the outcome under a Hold Firm Equilibrium is Con›ict if Defender is staunch. When Defender is per‹dious, the outcome depends on Protégé’s type: Protégé Loses when Protégé is loyal; Protégé Realigns when Protégé is disloyal. The ‹nal and perhaps most interesting perfect Bayesian equilibrium of the Protégé-Defender Subgame is the Bluff Equilibrium. At a Bluff Equilibrium, disloyal Protégés always hold ‹rm; loyal Protégés sometimes do the same (i.e., yD = 1; 0 < yL < 1). This means that a loyal Protégé’s node 2 choice is a bluff. Hoping to elicit Defender’s support by sending a (false) signal, it holds ‹rm even though it has no intention of realigning. When Defender is staunch, the bluff works, sometimes, though a per‹dious Defender never supports Protégé. This result is not surprising. Per‹dious Defenders never support Protégé under any perfect Bayesian equilibrium of the Protégé-Defender Subgame (i.e., zP always equals 0). The action choices of a staunch Defender under a Bluff Equilibrium are dif‹cult to categorize. A staunch Defender sometimes, but not always, supports Protégé (i.e., 0 < zS < 1). This behavioral tendency is in fact surprising. It runs counter to Defender’s equilibrium behavior under all the other perfect Bayesian equilibria of the Protégé-Defender Subgame. Protégé would probably view a staunch Defender’s willingness to withhold support as a veiled threat. Challenger might interpret it as a promise—of sorts. For Defender, it is a way to resolve the deterrence versus restraint dilemma. By hedging its bet, it achieves an optimal mix of both of these important security commodities. Defender’s action choices under a Bluff Equilibrium provide insight into an empirical puzzle raised by Fearon (1997). In Fearon’s game model of costly foreign policy signaling, a player has two strategies to communicate its interests: it can signal that its “hands are tied” or that its “costs are sunk.” Players never rationally bluff with either signal, leading Fearon to wonder “why we sometimes observe halfhearted signals when convincing ones are possible” (71). But this behavior arises naturally in the Tripartite Crisis Game. Under the Bluff Equilibrium in the Protégé-Defender Subgame, a staunch Defender’s strategy corresponds to a signal that is strong enough to deter all but the most determined Challengers, yet not so strong that loyal Protégés become intransigent and provoke crises. Defender’s rational objective is balance: too strong a
Bismarck’s System
79
commitment in›ames Protégé, whereas too weak a commitment incites Challenger. The signal is fuzzy, then, because it has two different audiences. By deterring Challenger, the signal minimizes the risk of con›ict and helps to stabilize the status quo; by restraining Protégé, it reduces the risk of chain-ganging and protects Defender’s alignment relationship with Protégé. This mixed message is surprising because it is delivered by a staunch Defender that would prefer to ‹ght to save its relationship with Protégé. None of the plausible perfect Bayesian equilibria of the Protégé-Defender Subgame coexist. To see this, consider now ‹gure 4.3. Along the horizontal and vertical axes of this ‹gure are graphed, respectively, the belief variables, pDef and pPro. Several constants, such as d1 and e2, are also indicated along the axes. These constants, whose technical characteristics will be ignored here, are convenient thresholds for categorizing the perfect Bayesian equilibria of the Protégé-Defender Subgame.29 The two belief variables can be interpreted, easily and quite naturally, as measures of Protégé’s and Defender’s threat credibility: the higher pPro , the more likely/credible Protégé’s threat to realign, and the higher pDef , the more likely/credible Defender’s threat/promise to support Protégé.30 Furthermore, since the belief variables indicate Defender’s and Protégé’s expectations of support or nonsupport as the crisis unfolds, the unrealized values of these variables capture the particular alignment pattern that both constrains and determines behavior in the Tripartite Crisis Game. As ‹gure 4.3 shows and as one might very well expect, a Settlement Equilibrium exists precisely when Defender is very likely per‹dious (i.e., when pDef < e2). Intuitively, when Defender is unlikely to be staunch, Protégé will (rationally) choose to settle with Challenger if and when it is faced with a decision at node 2. When Defender’s credibility is low, Protégé’s threat to realign is of no strategic consequence. In other words, the conditions associated with the existence of a Settlement Equilibrium do not depend on Protégé’s type. The reason is straightforward: given Protégé’s preferences, node 4a is never reached in the play of the game. Thus, Protégé’s threat to realign never factors into either Challenger’s or Defender’s strategy choice. Protégé’s credibility determines which of two perfect Bayesian equilibria 29. For a detailed discussion, see Zagare and Kilgour 2003. 30. This operational de‹nition of credibility re›ects an important axiomatic difference between classical deterrence theory and perfect deterrence theory. As chapter 2 shows, in classical deterrence theory, credibility is treated as a constant; by contrast, in perfect deterrence theory, credibility is not ‹xed.
80
the games of july
pPro 1
d1
0
Settlement
e2
Separating
e1
Hold Firm
1
pDef
Bluff
Fig. 4.3. Location of perfect Bayesian equilibria in Protégé-Defender Subgame
will exist when Defender’s credibility is high (i.e., when pDef > e1). When Protégé’s threat to realign is highly credible (i.e., when pPro > d1), a Hold Firm Equilibrium exists. But when it is low (i.e., when pPro < d1), a Bluff Equilibrium comes into play. Again, this result is consistent with intuition. A staunch Defender should be more willing to support Protégé as the credibility of Protégé’s realignment threat increases. When Protégé’s threat to realign is relatively credible (as it is under a Hold Firm Equilibrium), a staunch Defender will always choose to support Protégé. After all, a staunch Defender, by de‹nition, prefers Con›ict to Protégé Realigns. But as the credibility of Protégé’s threat
Bismarck’s System
81
erodes (as it does under a Bluff Equilibrium), Defender becomes progressively more likely to withhold support. A Separating Equilibrium occurs between the two extremes of Defender’s credibility (i.e., when e2 < pDef < e1). As indicated, under a Separating Equilibrium, play in the Protégé-Defender Subgame is determined strictly by the players’ types. Only disloyal Protégés hold ‹rm, and only staunch Defenders offer support. As with play under a Settlement Equilibrium, Protégé’s initial credibility (pPro) is of little moment because Protégé signals its type by its node 2 choice.31 As a result, when Defender makes its choice at node 3, it will know for sure what choice Protégé will make at node 4a should Defender choose not to support Protégé. In consequence, Defender’s node 2 choice is dictated by its type (preferences).
4.3.2. the tripartite crisis game with incomplete information Having laid out the contours of the Protégé-Defender Subgame, I now return to an analysis of the Tripartite Crisis Game with incomplete information, focusing on Challenger’s node 1 choice. Recall that at node 1 Challenger either makes no demand (in which case the outcome is Status Quo) or makes a demand (in which case the outcome is that of the Protégé-Defender Subgame). The outcome of the subgame, in turn, depends on which perfect Bayesian equilibrium will come into play—that is, on the two belief (credibility) variables pPro and pDef . Clearly, Challenger will choose to make a demand if and only if its utility for the outcome of the Protégé-Defender Subgame exceeds its utility for the Status Quo. As it turns out, the key to Challenger’s decision is the relationship of the threshold probabilities that separate the various equilibrium regions of the Protégé-Defender Subgame to another threshold probability, c1. This threshold is the level at which Challenger’s utility for a lottery between Protégé Realigns and Con›ict equals its utility for the Status Quo. If the probability of Con›ict exceeds c1, Challenger prefers the Status Quo. If it is less, Challenger prefers the lottery. Since the likelihood of a crisis increases as c1 increases, this quantity, which I refer to as Challenger’s critical risk,32 can be thought of as a measure of Challenger’s propensity to foment a crisis.33 31. Recall that disloyal Protégés always hold ‹rm; loyal Protégés always concede. More technically, under a Separating Equilibrium, q (Defender’s updated probability that Protégé is disloyal given that Protégé holds ‹rm at node 2) equals 1. 32. The term was coined by Ellsberg (1959) to refer to a roughly analogous threshold probability in his in›uential crisis bargaining model. 33. The threshold probability c1 = (cPR – cSQ) / (cPR – cC+), where cPR = Challenger’s utility for Protégé Realigns, cSQ = Challenger’s utility for the Status Quo, and cC+ = a determined Challenger’s utility for Con›ict.
82
the games of july
In the Tripartite Crisis Game with incomplete information, a determined Challenger’s critical risk increases as its utility for Protégé Realigns increases, as its utility for Con›ict increases, or as its utility for the Status Quo decreases. This means, ceteris paribus, that a Challenger that places a relatively high value on breaking up the relationship between Protégé and Defender is more likely to precipitate a crisis than one that does not. It also means that a Challenger averse to Con›ict is less likely to issue a demand than one that is not. And ‹nally, it means that a highly dissatis‹ed Challenger is more likely to upset the Status Quo than one that is relatively satis‹ed. For expository purposes, it is useful to think of Challenger’s critical risk as being at one of three levels: low, medium, or high. Low values are those below the threshold probability e2 that serves as the boundary between the regions of Settlement and Separating Equilibria (see ‹gure 4.3). High values are those above the threshold probability e1 that sets the region of Separating Equilibria apart from the regions of Hold Firm and Bluff Equilibria. Medium values are in between (i.e., e2 < c1 < e1), falling squarely in the region of Separating Equilibria.34 Figure 4.4 provides a graphical summary of Challenger’s node 1 decision for each theoretically meaningful level of Challenger’s critical risk: low (Case A), medium (Case B), and high (Cases C1 and C2).35 In each case, an area of gray shading is superimposed on the equilibrium structure of the Protégé-Defender Subgame as given in ‹gure 4.3. The shading delimits the conditions under which Challenger demands an adjustment of the Status Quo at node 1. (No shading means that Challenger prefers to accept the Status Quo). Table 4.5 provides a verbal summary. As ‹gure 4.4 shows, Challenger becomes progressively more likely to issue a demand as c1 increases. The shaded area, where a demand is issued, grows larger and larger as one moves rightward and downward from Case A through Case B to Cases C1 and C2. Consider now Challenger’s node 1 choice given that Defender’s credibility is low (i.e., when pDef < e2). This contingency is just about Protégé’s worst nightmare. Recall that when Defender is very likely per‹dious, a Settlement Equilibrium uniquely exists. Under any Settlement Equilibrium, Protégé always concedes at node 2, and the outcome is always Challenger Wins. Since the assumption is that a determined Challenger prefers Challenger Wins to Status Quo, Challenger will always issue a demand at node 1 in order to secure a more 34. For the speci‹cs, see Zagare and Kilgour 2003. 35. The technical distinction between subcases C1 and C2 has no bearing on the discussion here.
Bismarck’s System
Case A
Case B
Case C1
Case C2
83
Fig. 4.4. Possible Geometries of No-Demand Region of Tripartite Crisis Game
preferred outcome, even when its critical risk is low. Thus, in each case depicted in ‹gure 4.3, the region indicating the existence of a Settlement Equilibrium is shaded. What happens at intermediate levels of Defender’s resolve (i.e., where e2 < pDef < e1), where a Separating Equilibrium governs play of the Protégé-Defender Subgame? As ‹gure 4.4 suggests, Challenger’s critical risk interacts with
84
the games of july
the credibility of both Protégé’s and Defender’s threats to determine Challenger’s node 1 choice. When Challenger’s critical risk is low (Case A), Challenger rationally accepts the Status Quo at node 1, but only when the credibility of Protégé’s and Defender’s threats (i.e., the values of pPro and pDef) is relatively high. Much the same can be said when Challenger is moderately motivated (Case B), except that proportionately higher levels of Protégé and Defender credibility are required to deter Challenger. Finally, when the magnitude of Challenger’s critical risk is high (i.e., exceeds the threshold probability e1), Challenger always issues a demand and is never deterred (Cases C1 and C2). Thus, under a Separating Equilibrium, Challenger’s behavior is fully consistent with our initial conclusion: ceteris paribus, as its critical risk increases, so does Challenger’s propensity to foment a crisis. The same general pattern holds when Defender is likely staunch (i.e., when pDef > e1) and either a Bluff or Hold Firm Equilibrium is in play—as long as Challenger’s critical risk is low (Case A) or moderate (Case B). But a very interesting anomaly arises when Challenger is highly motivated (Cases C1 and C2). Before discussing this irregularity, however, notice that even when Challenger’s motives are intense (i.e., c1 > e1), there is always a region under which no demand is made and Challenger is deterred. One might expect as much in
TABLE 4.5. Challenger’s Node 1 Choice at the Perfect Bayesian Equilibrium of the Tripartite Crisis Game Challenger’s Critical Risk c1
Settlement Equilibrium pDef < e2
Low c1 < e2
Case A
Medium e2 < c1 < e1
Case B
Challenger always demands
Case C1
Outcome is always Challenger Wins
High c1 > e1 Case C2
Separating Equilibrium e2 < pDef < e1
Hold Firm Equilibrium pDef > e1 pPro > d1
Bluff Equilibrium pDef > e1 pPro < d1
Challenger sometimes demands
Challenger never demands
Challenger sometimes demands
Challenger sometimes demands
Challenger never demands
Challenger sometimes demands
Challenger always demands
Challenger sometimes demands
Challenger sometimes demands Challenger always demands
Bismarck’s System
85
these cases, given that Defender is most likely staunch. But the high likelihood that Defender is staunch and will stand by Protégé is a necessary but not suf‹cient condition for deterrence to hold against a highly determined Challenger. In general, for the Status Quo to survive, Protégé must also likely be disloyal. Since a disloyal Protégé always holds ‹rm (under either equilibrium form), and a staunch Defender always supports a disloyal Protégé (under a Hold Firm Equilibrium), Challenger can always be dissuaded from issuing a demand at node 1. A reputation for disloyalty, then, can serve Protégé well. In the context of this conclusion it is easy to appreciate the strategic rationale of Lord Palmerston’s well-known maxim about the impermanence of Great Britain’s commitments: “We have no eternal allies and we have no perpetual enemies. Our interests are eternal and perpetual, and these interests it is our duty to follow.” Equally pertinent, though, is Metternich’s less diplomatic aphorism: “One must keep an eye on his allies, no less than on his enemies.” Observe, though, that in Cases C1 and C2, the left-hand boundary of the (unshaded) region where no challenge takes place slopes upward and to the right. This fact means that, against a determined Challenger, there are conditions under which Protégé’s threat to realign (directed at Defender) is in fact counterproductive. What is particularly striking about this observation is that exactly the opposite pattern occurs under either a Settlement or a Separating Equilibrium, where the slope of the left-hand boundary of the no-challenge region is negative. Under most conditions, then, Protégé’s threat to realign increasingly induces Defender’s support (at node 3), which, in turn, tends to discourage Challenger from initiating a crisis. But when Challenger is highly motivated and its critical risk high, it is possible that a heightened likelihood that Protégé is disloyal will actually make matters worse and provoke a challenge. Not only does the Tripartite Crisis Game model reveal this unexpected behavioral consequence, but it also identi‹es, precisely, the conditions under which it comes into play. One could attempt to explain this anomaly by referring to Challenger’s preferences. Since the assumption is that Challenger prefers Protégé Realigns to Protégé Loses, one might conjecture that, ceteris paribus, Challenger would be more motivated to issue a demand as the likelihood of realignment increases. But this explanation does not generally hold. Indeed, when Challenger’s motivation is low or moderate, the argument runs the other way. The opportunity to break up Defender’s and Protégé’s relationship changes Challenger’s behavior only when Challenger’s critical risk is high, beyond the threshold e1. Under certain conditions, therefore, a determined Challenger might
86
the games of july
choose to foment a crisis in the hope of disrupting a fragile strategic relationship. The prospect of separating Defender and Protégé, however, is a long shot; it is also quite risky since this motivation comes into play only under a Hold Firm Equilibrium. Although Protégé is likely disloyal under a Hold Firm Equilibrium (pPro > d1), it is also the case that Defender is likely staunch (pDef > e1). Since a staunch Defender always supports Protégé under a Hold Firm Equilibrium, crises played out in this strategic environment tend to escalate to Con›ict. Although the next three chapters develop a more nuanced explanation of German foreign policy in 1914, one could plausibly explain both Germany’s strong support of Austria-Hungary in early July and the subsequent escalation of the crisis in terms of the behavioral expectations derived from an analysis of the Tripartite Crisis Game and the incentive it sometimes offers to a highly motivated Challenger to contest the status quo. In 1914, the German chancellor, Theobald von Bethmann Hollweg, who initially expected that the crisis would terminate in a localized war between Austria-Hungary and Serbia, both hoped and believed that a third Balkan War would shatter Russia’s alliance with France (Levy 1990/91: 160). Hence, the blank check. Of course, he was mistaken. But, given the equilibrium structure of the Tripartite Crisis Game with incomplete information, the tragic consequences of Bethmann Hollweg’s erroneous beliefs, which were shared by decision makers in Vienna, are not the least bit surprising. What, then, should a threatened Protégé do? The answer in general is not clear-cut. Against a weakly or even a moderately dissatis‹ed Challenger, an ambiguous alignment policy can help by strengthening Defender’s incentive to support Protégé, thereby deterring Challenger. But against a highly motivated Challenger, a shaky relationship could prove fatal by encouraging Challenger to initiate a crisis. To bolster deterrence in this case, Protégé must tilt, ever so gingerly, toward Defender.36 Of course, this stratagem may cost Protégé some autonomy as well as some leverage in its relationship with Defender (Morrow 1991). Moreover, other potential challengers may see Defender as less likely to stand by Protégé in a crisis (because Defender will have less to lose), and Protégé may thus face additional challenges. Protégé, in other words, is strategically con›icted. Sometimes a reputation for disloyalty makes a challenge less likely—but not always. Paradoxically, Protégé may attain its greatest bargaining leverage prior to negotiating its rela36. Subtlety is required here, as too much loyalty could also precipitate a crisis.
Bismarck’s System
87
tionship with Defender—aligning too closely with Defender may increase Protégé’s risk, and it may consequently demand greater compensation for doing so. Provided that Defender’s (and Protégé’s) credibilities are past certain thresholds, Defender’s best bet would be to demonstrate a stronger relationship with Protégé. A general alliance or a mutual defense pact would serve this purpose well. Both increasing the likelihood of Defender’s intervention and reducing the likelihood of Protégé’s realignment strengthen deterrence by dampening a highly motivated Challenger’s incentive to contest the status quo. And, of course, the absence of a crisis makes realignment even less likely.
4.4. Explaining the Austro-German Alliance of 1879 Having laid out the broad contours of the Tripartite Crisis Game model with incomplete information, I turn next to the question of Bismarck’s motivation for negotiating the Dual Alliance with Austria-Hungary. It is clear that con›ict avoidance was no small part of Bismarck’s decision-making process in 1879, as both Kissinger (1994: 158–59) and Massie (1991: 79) attest. But this explanation for the union of the two German-speaking states is incomplete, for this was no ordinary alliance. To begin, the Austro-German alliance was “the ‹rst permanent arrangement in peace-time between two Great Powers since the end of the ancien régime” (Taylor 1954: 264). In addition, the particulars of the agreement fully re›ected the standpoint of Austria, the weaker and more dependent partner. Afraid of offending France and England, Austria sought a defensive pact aimed solely against Russia. Although Bismarck preferred a more general arrangement, the negotiated treaty embraced the Austrian position. Moreover, Germany received few defensive bene‹ts from the alliance: Germany did not need Austria if Russia attacked, unless France joined in. In addition, since German interests almost dictated that Germany would have to defend Austria in the event of a Russian challenge, a formal alliance could have only a minimal impact on the credibility of Germany’s (extended) deterrent threat. In other words, pDef was already quite high. Note that this is precisely the condition under which the high likelihood that Protégé is disloyal can increase the probability of a crisis. Still, Bismarck pushed hard to ‹nalize this treaty. When the kaiser resisted Bismarck’s policy recommendation, the chancellor even threatened to resign (Taylor 1955: 189). Why, then, would Bismarck place so much impor-
88
the games of july
tance on a one-sided arrangement that did not fully re›ect his own preferences and seemingly offered Germany limited strategic advantages? The answer lies, in large part, in the alliance’s impact on Austria’s likely behavior and, by extension, the alignment status of the European system. As G. Snyder (1997: 90) explains, Bismarck’s primary motive in negotiating an alliance with Austria was preclusive: to prevent Austria from allying elsewhere. . . . Bismarck reasoned . . . that if Austria were unsupported by Germany in a severe crisis with Russia, the internal political balance might shift and Austria might be driven by its weakness to make a deal with Russia about the Balkans, leading to an anti-German alliance with Russia and France.
If Snyder is correct, Bismarck must have thought that in the event of a crisis with Russia, there was some good chance that German support for Austria could fail to materialize.37 Still, it is unlikely that Bismarck sought an alliance with Austria solely to tie Germany’s (i.e., the kaiser’s) hands, since actual German intentions (and likely contingent behavior) were somewhat beside the point, of little immediate strategic consequence. What really mattered, both in theory and in practice, was Russia’s perception of Germany’s intentions (i.e., pDef). The German extended deterrent threat, however, was naturally high.38 In consequence, it did not need to be shored up by means of a formal alliance. By contrast, the risk that a forsaken Austria might drift outside the German orbit weighed heavily on Bismarck’s mind (Langer 1950: 180). Until mid1879, Bismarck had resisted repeated Austrian overtures for a formal alliance. The explanation is simple: there had previously been no good reason to worry about Austrian loyalty or, by extension, a Russian challenge. What changed suddenly was news that Austria’s pro-German foreign minister, Count Julius Andrássy, planned to resign. Anticipating an Austrian foreign of‹ce far less committed to Germany, Bismarck moved quickly to conclude an alliance before Andrássy left of‹ce. 37. In the absence of this belief, Snyder’s argument is dubious. Langer (1950: 175–76) asserts that as early as 1876, Bismarck believed that “Germany could not afford to see Austria completely defeated and deprived of her position as a great power” (emphasis added), suggesting that there were thresholds that would trigger German support (or nonsupport). Langer concludes that “it does not follow from [Bismarck’s pursuit of an alliance with Austria] that the chancellor had de‹nitely decided to back Austria against Russia.” 38. Nevertheless, some doubt existed about German intentions. British prime minister Benjamin Disraeli “was not at all certain that Germany would come to the aid of Austria in the event of a Russian attack” (Langer 1950: 187). And if some doubt existed in Britain, it is likely that some uncertainty also existed in Russia about the extent to which Germany would stand by Austria in a crisis.
Bismarck’s System
89
The Tripartite Crisis Game model explains why. Clearly, Bismarck’s policy push was driven by a desire to alter Russia’s expectation of Austrian behavior in the event of a challenge (Taylor 1954: 263). It is more than telling that Bismarck felt compelled to exaggerate the extent of the Russian threat to convince the kaiser of the necessity of a formal agreement (Taylor 1954: 259–60). After all, the incentive for Defender to align itself more closely with Protégé exists only when a determined and highly motivated Challenger is likely willing to run the risk of war to adjust the prevailing status quo. Put differently, Bismarck hoped to forestall an immediate crisis with Russia by negotiating an agreement that reduced the probability that Austria might be tempted to join with Russia in an anti-German coalition. The shortterm respite that the defensive alliance with Austria allowed would also buy him the time necessary to revive the Three Emperors’ League. In Bismarck’s mind, this conservative association among Germany, Austria, and Russia would reinforce France’s isolation and help to solidify Germany’s gains from the wars of 1866 and 1870. Just two years later, the league was reestablished.
4.5. Coda The intricate alliance system that Bismarck devised helped keep Europe at peace until 1914. But as Massie (1991: 81) notes, “When Bismarck was gone, new patterns would form, new games would be played.” It is likely no coincidence that when war ‹nally came, it came in the context of a confrontation between Austria and a determined Russia in the Balkans that escalated to war only after Germany, fearful that the Dual Alliance would otherwise collapse, issued its longtime partner a blank check. In the next three chapters, I explain why the blank check was issued and why the con›ict ‹rst escalated and then expanded to include all the major powers in the European state system.
The great European war would come out of some damned foolish thing in the Balkans. otto von bismarck
chapter five
Vienna, Berlin, and the Blank Check
Archduke Franz Ferdinand, the heir apparent to the Austro-Hungarian throne, was assassinated in Sarajevo on June 28, 1914. At 11:00 PM on August 4, Great Britain declared war on Germany. What transpired in between has come to be known as the July Crisis. It is something of a misnomer. The July Crisis should not be thought of as a singular event. During this interval, a number of distinct decisions were made—in Vienna and Berlin, in St. Petersburg and Paris, and ‹nally in London—and a series of important games were played out over several weeks. Taken together, these decisions and games brought about World War I, an allout con›ict that George Kennan (1979: 3) has described as “the great seminal catastrophe” of the twentieth century. From almost every vantage point, the Great War was an unmitigated disaster. In the years that followed the armistice that ended it, the participants, from the kaiser on down, sought to justify their behavior and to explain their actions. A tragedy of this magnitude, however, required much more than an explanation. A scapegoat was needed. Initially, and to some extent understandably, German leaders were ‹ngered. Eventually, however, as geopolitical considerations overtook sober analysis, all the participants were exonerated (Mombauer 2002: 123). The consensus that eventually emerged was that World War I was an accident for which no one in particular was responsible. As Britain’s prime minister David Lloyd George put it in his War Memoirs, the participants had simply “slithered” into the con›ict. Today, despite challenges from some historians, most notably F. Fischer (1967) and his disciples, the view that World War I had in some sense been accidental continues to hold sway, especially among strategic analysts and those political scientists who study interstate con›ict. After all, in 1914 the two lead90
Vienna, Berlin, and the Blank Check
91
ing coalitions were in rough balance, and if the conventional wisdom was correct, World War I had to be an aberration (Waltz 1993: 77). Rather than conceding the obvious, however, many realists blamed fate, dumb luck, or blundering decision makers for the failure of balance-of-power theory to correctly anticipate the onset of a major-power con›ict in 1914. Others traced the con›ict to cognitive, perceptual, or psychological de‹ciencies that presumably precluded rational decision making. Either way, the war was seen as one big mistake. There is, however, very little empirical support for the accidental war thesis (Trachtenberg 1991). Indeed, an overwhelming amount of evidence suggests that World War I came about as a consequence of a series of “conscious and calculated” decisions that surely had disastrous consequences but that were instrumentally rational decisions nonetheless (Tunstall 2003: 131; Williamson 1991: 213). Accordingly, in this and the next two chapters, I examine the interactive choices in three critical games that, collectively, capture the essentials of the crisis: (1) Germany and Austria-Hungary’s joint decision, reached early in July, to resolve the latter’s problems with Serbia; (2) the sequence of choices that brought about a general European war when Germany declared war on Russia and France in early August; and (3) the outbreak of a world war after the German invasion of Belgium and Great Britain’s entry into the war. Three important questions guide the analysis. Speci‹cally, why on July 5 did the kaiser give Austria-Hungary carte blanche to pressure Serbia? Why did the con›ict escalate to a European war shortly after the Austrian ultimatum to Serbia expired? And ‹nally, why did the British attempt at deterrence fail so spectacularly? To answer these questions, I begin with some historic context.
5.1. Background As chapter 4 shows, when German chancellor Otto von Bismarck ‹rst negotiated the Dual Alliance with Austria-Hungary in 1879, he had a long-term strategic purpose in mind. Bismarck’s goal was not simply to draw the Hapsburg monarchy closer to Imperial Germany to foreclose the possibility of a war between Austria-Hungary and Russia. Rather, the Dual Alliance was also viewed as a necessary ‹rst step toward the creation of a larger (blocking) coalition that would enable Bismarck to solidify the gains made in the wars of 1866 and 1870. Since by then Bismarck had announced that Germany was a “satiated” power, his diplomacy was to be guided by his well-known aphorism: “All
92
the games of july
politics reduce themselves to this formula: try to be à trios in a world governed by ‹ve powers.” To help preserve the status quo, therefore, Bismarck resurrected the Three Emperors’ League (Dreikaiserbund). The Three Emperors’ League was a defensive alliance: Germany, Austria-Hungary, and Russia pledged, inter alia, a benevolent neutrality should any member of the league ‹nd itself at war with a fourth power—that is, with France. In Bismarck’s mind, however, the agreement would not only help to isolate France but also allow him to moderate the competition between Austria and Russia in the Balkans, which, if left unmanaged, could easily escalate into a war that would threaten German interests. In addition to reestablishing the Three Emperors’ League in 1881, Bismarck entered into a number of other strategic arrangements and secret treaties. For example, in 1882, the Dual Alliance morphed into a Triple Alliance when Italy sought German protection from France. Similarly, in 1883, both Germany and Austria entered into a mutual defense pact with Rumania. The most signi‹cant of these arrangements, however, was the secret treaty that Bismarck negotiated with Russia in 1887. Known as the Reinsurance Treaty, this three-year pact obligated German neutrality in the event that Austria-Hungary attacked Russia, and Russian neutrality in the event that France attacked Germany. The Reinsurance Treaty was the capstone of “the Bismarkian system, . . . a network of interlocking alliances, carefully balanced and kept in order by the master diplomat in Berlin” (Massie 1991: 82). So sound was this system that it survived for almost a quarter of a century after Bismarck left of‹ce, despite several acute crises and a number of minor-power wars. In the end, however, the system did collapse due, in no small part, to the ineptitude of German diplomacy. The problems began immediately. Almost as soon as Bismarck was driven from of‹ce in 1890 by the new German kaiser, Wilhelm II, the chancellor’s successors in the Wilhelmstrasse allowed the Reinsurance Treaty to lapse. Historians place great signi‹cance on this sudden policy shift which signaled a “new course” in German foreign policy. As Craig (1978: 230) has suggested, “One could argue plausibly that [the decision not to renew the Reinsurance Treaty] was the most crucial of all those made between 1890 and the outbreak of the First World War and that it set in train the whole chain of calamity that led toward the catastrophe.” Although Bismarck’s successors undoubtedly had bureaucratic and idiosyncratic reasons for shifting the direction of German foreign policy, they had strategic reasons as well. The kaiser’s key foreign policy advisers, the most
Vienna, Berlin, and the Blank Check
93
in›uential of whom was Friedrich von Holstein, the ‹rst counselor of the Foreign Ministry’s Political Department, hoped to forge a new alliance in which Russia was replaced by Great Britain. At the same time, by adopting a more aggressive approach to world affairs (Weltpolitik), the German foreign policy establishment hoped to capture its “place in the sun.” The policy clearly failed. Cut adrift, Russia had little choice but to align more closely with France. By 1894, Russia and France were in a formal alliance and committed to each other’s defense should either be attacked by Germany. As Kagan (1995: 126) succinctly writes, “Bismarck’s worst nightmare had become a reality.” But it actually got worse than this. Germany’s new course—but especially its aggressive program of building state-of-the-art battleships (dreadnoughts)—both challenged and alienated the British. The German leadership hoped and expected that Britain, when faced by a surging Germany, would seek accommodation. Instead, Britain gradually moved closer to France. In 1898, the British and French almost went to war over control of the Upper Nile. But shortly after the Fashoda Crisis, they began to reconcile. By 1904, the two erstwhile enemies reached an agreement, the Entente Cordiale, that settled a number of colonial disputes and, more generally, brought the two countries’ foreign policies into closer alignment.1 It was only a matter of time until all the dots were connected: in 1907, Great Britain and Russia reached an accord designed to reduce the intensity and scope of their competition in Asia. In no sense could Germany now be considered part of a strategic triad.2 Today we might say that Germany was contained, or “in the box.” In the years leading up to the Great War, however, the Germans used the term Einkreisung to describe their encirclement by hostile forces. The partnership between Germany and Austria-Hungary was not inconsequential, however. As chapter 2 discusses, the European political system was in rough balance in July 1914.3 In realist theory, a war was very unlikely—unless there were extenuating circumstances. Of course, there always are. In this case, the problem was with Austria-Hungary, Germany’s junior partner. The Hapsburg monarchy was a multiethnic monstrosity that had long teetered on the edge of disintegration. Ruled by two minority nationalities, the 1. The most important understanding concerned Morocco and Egypt. In exchange for recognizing British interests in Egypt, France received a free hand in Morocco. According to B. Schmitt (1934: 67), when former British foreign secretary Lord Roseberry found out about this arrangement, he declared, “It means war with Germany in the end.” 2. Italy was not thought to be a reliable partner. Indeed, in July 1914, when push came to shove, it defected from the Triple Alliance. 3. See also Kennedy 1987: chap. 5; Kugler and Domke 1986.
94
the games of july
Germans in Austria and the Magyars in Hungary, Austria-Hungary was generally regarded, after the Ottoman Empire, as the “next sick man of Europe.” If the comparison was apt—and it was—Austria-Hungary would be the next to implode. Ethnic defections to neighboring states represented the gravest threat to the dual monarchy’s survival. Of these states, Serbia was considered the most determined to foment unrest, especially since it had emerged as the big winner after the Second Balkan War in 1913. So when Franz Ferdinand was assassinated on June 28, the oligarchs in Vienna quickly decided to eliminate Serbia, either by absorption or by dismemberment (Jannen 1983: 55). It is more than likely, however, that the assassination was but a pretext. Indeed, the policy Austria-Hungary intended to implement had been formulated before Sarajevo. Just how far in advance, however, is open to debate. Some historians point to the so-called Matscheko Memorandum as proof that Austria-Hungary was intent on eliminating Serbia even before Franz Ferdinand was murdered. The memorandum, which advocated an aggressive Balkan policy, was completed on June 24 and was intended to be presented to the members of Austria’s foreign policy establishment, including the archduke (Tunstall 2003: 130). Others, however, trace the plan further back: Abel (1941: 857), for example, argues that “the Austrian power group . . . made up their minds to have war with Serbia several years before the outbreak of hostilities and were only awaiting a favorable opportunity.” Abel’s claim is hard to sustain. Most historians now believe that the de‹nitive decision to eliminate Serbia was made at a meeting of Austria’s Common Ministerial Council on July 7. Even still, the ultimatum was not issued until July 23; the Austrian declaration of war against Serbia ‹nally came on July 28. The delay between decision and implementation is not dif‹cult to explain. Two factors stand out. First, the mobilization plan that was favored by the chief of Austria-Hungary’s General Staff, General Franz Conrad von Hötzendorf, required at least two weeks to implement. Known as Plan B (Balkans), this strategy involved partial mobilization against Russia and major offensive action against Serbia.4 Compounding this operational problem was the fact that a signi‹cant number of Austrian troops were on “harvest leave.” As Williamson (1983: 28) notes, “If these troops were suddenly recalled to their units, their re-
4. An alternative all-or-nothing plan, Plan R (Russia), called for total mobilization and an offensive directed against Russian forces in Galicia. A more limited option, Plan M (Montenegro), was never seriously considered. For additional details, see Tunstall 1993; Williamson 1983.
Vienna, Berlin, and the Blank Check
95
turn would immediately be detected. This in turn would deprive the Hapsburg decisionmakers of any chance for even modest surprise, alarm the other powers, probably disrupt the mobilization process, and—not without grave consequences—leave the crops unharvested.” Given these considerations, no military action could be taken until the leaves expired on July 25. A second source of delay was the desire of the Austrian foreign minister, Count Leopold Berchtold, to put off issuing an ultimatum until a scheduled visit of France’s president, Raymond Poincaré, and prime minister, René Viviani, to Russia was concluded on July 23. Taken together, these constraints “meant that ‘settling accounts’ with Serbia could not begin until the last week of July” (Tunstall 2003: 139). These implementation considerations aside, however, another, more basic, reason explains why closure was not reached in Vienna immediately after the assassination. This aspect of the delay is best understood in the context of the Tripartite Crisis Game, which is reproduced here for convenience as ‹gure 5.1. The Tripartite Crisis Game provides a rich theoretical framework for analyzing extended deterrence relationships in a multipolar system. In chapter 4, it is used to explain what was, at the time, Bismarck’s out-of-character decision to enter into an alliance with Austria-Hungary. And in chapter 7, an important variant of the model is used to explain Sir Edward Grey’s still-controversial policy not to commit Britain unequivocally to the defense of France until after Belgium was invaded. In this chapter, however, its main purpose is to shed some theoretical light on a policy choice that stands in stark contrast to Grey’s, the so-called blank check. One might think that the German decision to pledge unquali‹ed support to its most signi‹cant ally is easy to explain. But in the past, German support of Austria-Hungary had been anything but automatic. Indeed, at a meeting that took place at Konopischt, Franz Ferdinand’s country estate in Bohemia, just weeks before he was assassinated, the kaiser turned down his friend’s request to back Vienna’s plan to implement a more aggressive Balkan policy (Fromkin 2004: 98). In early July, however, the mercurial kaiser completely reversed himself, telling the Austrian ambassador, Count László SzögyényMarich, that he could “rely on Germany’s full support” for what was essentially the same hard-line policy initiative he had previously rejected (Geiss 1967: 76). In what follows, the Tripartite Crisis Game model is recalibrated, ever so slightly, to develop a theoretically informed explanation of Wilhelm’s fateful decision to stand unconditionally by Austria-Hungary at the beginning of July 1914. After that, the causal signi‹cance of the blank check is evaluated.
96
the games of july
Challenger Not Demand (1 – x)
Node 1
Demand (x)
Protégé
Status Quo
Concede (1 – y)
Node 2
Hold Firm (y)
Defender
Challenger Wins
Support (z)
Not Support (1 – z)
Node 4a
Node 4b
Protégé
Realign
Protégé Realigns
Node 3
Not Realign
Protégé Loses
Challenger
Back Down
Challenger Concedes
Press On
Conflict
Fig. 5.1. Tripartite Crisis Game
5.2. The Tripartite Crisis Game Redux Recall from the last chapter that the Tripartite Crisis Game model has three players, Challenger, Defender and Protégé. Challenger begins play by deciding to make a demand. If a demand is made, Protégé must decide whether to concede or hold ‹rm. If Protégé holds ‹rm, Defender chooses either to support or withhold support from Protégé. If Defender supports Protégé, Challenger either backs down or presses on; if Defender withholds support, Protégé chooses whether or not to realign. The outcomes associated with these choices are summarized in ‹gure 5.1. In discussing this game here, all previous assumptions about preferences are retained. Table 5.1 restates the particulars.5 5. See also Zagare 2009b, which forms the basis of the remainder of this chapter.
Vienna, Berlin, and the Blank Check
97
In using the Tripartite Crisis Game—or any game model, for that matter— to analyze a real-world event, one must caution against too literal an interpretation of the nomenclature used to describe the players, their choices, and the outcomes. Models rarely conform exactly to the messy world of politics, interstate or otherwise, where choices are made by fallible human beings suffering from a variety of perceptual, cognitive, psychological, and related limitations and in a wide variety of strategic contexts. Thus, a few emendations must be made to the Tripartite Crisis Game before it can be used to explore the decision-making process in Vienna immediately after the assassination in Sarajevo. I begin with the speci‹cation of the players. Assuming that the assassination of Franz Ferdinand represented a clear challenge to the integrity of the Austro-Hungarian empire, which is how this event was viewed in Vienna and, to a lesser extent, in Berlin, the association of two of the players in the model world with their real-world analogues is straightforward. After Sarajevo, Austria-Hungary (Protégé) had to decide whether to hold ‹rm by taking steps, whatever they might be, to eliminate the threat posed by Serbia, or to concede the issue and attempt to preserve the monarchy by reforming internally. Of course, if Austria-Hungary sought reprisal against Serbia, Germany (Defender) would have to decide whether to support its partner. From Berlin’s point of view, support clearly brought with it the risk of war (with Russia), while nonsupport risked the rupture of the Dual Alliance (Farrar 1972). The stark choice that Bismarck sought to de›ect in 1879 and thereafter would simply not go away. While no great conceptual leap is required to connect, plausibly, the roles played by Austria-Hungary and Germany with two of the players in the Tripartite Crisis Game, the association of a real-world actor with the player named Challenger is more problematic. Since Challenger makes two choices in the Tripartite Crisis Game, it is important to treat this question seriously (Riker and Ordeshook 1973: 239). If a model is to tell us anything at all about a TABLE 5.1. Preference Assumptions for Tripartite Crisis Game Challenger
Protégé
Defender
Challenger Wins
Status Quo
Status Quo
Protégé Realigns
Challenger Concedes
Challenger Concedes
Protégé Loses
Conflict
Challenger Wins
Status Quo
Challenger Wins
Protégé Loses
Conflict or Challenger Concedes
Protégé Loses or Protégé Realigns
Conflict or Protégé Realigns
98
the games of july
real-world interaction, it must bear a reasonably close resemblance to the event in question. Player assignments should not be made willy-nilly. With this in mind, consider ‹rst the initial move made by Challenger. From Vienna’s vantage point, Serbia was clearly viewed as the challenger. As Kagan (1995: 188) points out, “The Austrians did not doubt Serbian complicity” in the archduke’s murder. However, little hard evidence supports this judgment. The assassin, Gavrilo Princip, was indeed a Bosnian Serb, but his action was neither initiated nor sanctioned by the Belgrade government, although Serbian prime minister Nicola Pasic likely had foreknowledge of the plot (Fromkin 2004: chaps. 19, 24). Thus, it strains credulity to consider Serbia the node 1 decision maker. In fact, it would seriously distort reality to consider any real-world government to have made the initial choice in the game. On the other hand, the assassination of the heir apparent (Thronfolger) to the Austro-Hungarian throne was real and could not be ignored, irrespective of who devised the plan or how it was implemented. How, then, can the event that is generally considered to have brought on the Great War be handled? One way—indeed, the most defensible way—is to consider the assassination as a random event or as an act of nature. Given this interpretation, the choice at node 1 can be treated as a fait accompli that simply precipitated the proper subgame that commences with Protégé’s (i.e., Austria’s) choice at node 2. Of course, Challenger also makes a choice (at node 4b) in the Protégé-Defender Subgame, so we are not quite off the hook yet. Challenger’s choice arises in the context of Protégé’s node 2 decision to hold ‹rm and Defender’s node 3 decision to support Protégé. Given these prior decisions, Challenger at node 4b must decide whether to accept the challenge (press on) or concede (back down). Pressing on implies con›ict. Backing down implies concession and defeat. But who, if anyone, faced this decision in July 1914? Since, for the moment, we are primarily interested in explaining Austria’s node 2 choice, we must answer this question as seen by Austria-Hungary. The usual suspects are Russia and Serbia. Neither, however, will do. Consider ‹rst Russia. From Vienna’s point of view, Russia was simply not a player in early July. As Williamson (1983: 27) points out, “Possibly the most striking feature of the Hapsburg decisionmaking process in July 1914 was its failure to think seriously anew about Russia’s position and its possible intervention. The policymakers acted as if Russia did not exist.”6 Below I explain why the likely Russian reaction was ignored by 6. See also Jannen 1983: 55; Tunstall 2003: 142.
Vienna, Berlin, and the Blank Check
Challenger
Not Demand
99
Node 1
Demand
Protégé
Status Quo
Concede (1 – y)
Node 2
Hold Firm (y)
Defender
Challenger Wins
Node 3
(q)
ProtégéDefender Subgame
Not Support (1 – z)
Node 4a
Conflict
Protégé
Realign
Protégé Realigns
Support (z)
Not Realign
Protégé Loses
Fig. 5.2. Tripartite Crisis Game when Challenger is determined
Austrian foreign minister Berchtold and other key decision makers at the Ballhausplatz, the home of Austria-Hungary’s Foreign Ministry. Consider now Serbia. Like Russia, Serbia’s response was not seriously considered by decision makers in Vienna. In fact, the ultimatum it delivered on July 23 was carefully crafted to leave Belgrade with little choice but to reject it (Kautsky 1924: no. 49). It was, as British foreign minister Edward Grey characterized it, “the most formidable document [he] had ever seen addressed by one State to another that was independent” (Gooch and Temperley 1926, 11: no. 91). Of course, rejection implied war, which was Vienna’s preferred outcome. Assuming (1) that no real-world actor made an actual choice at node 1 and
100
the games of july
(2) that the Austrian ultimatum, issued with Germany’s backing, implied Con›ict, the Tripartite Crisis Game reduces to the game highlighted in ‹gure 5.2 (previously referred to as the Protégé-Defender Subgame). So, in the interest of parsimony, the Tripartite Crisis Game will be cast aside, at least temporarily, and instead the Protégé-Defender Subgame will be called on to do the heavy explanatory lifting in this chapter. Recall, however, that previously the assumption was made that Protégé preferred Challenger Concedes to Con›ict. Now, however, the assumption about Austria-Hungary’s preferences is precisely the opposite. One might well ask, then, whether this lack of congruence between our initial and our present assumptions calls into question the model’s ‹t with the facts on the ground. The short answer is that the discrepancy between initial and present assumptions is of little theoretical moment—in this instance. The reason is straightforward. The analysis that follows rests on the further assumption that Con›ict will follow if Protégé holds ‹rm and is supported by Defender at node 3. Therefore, Protégé’s relative preference between Challenger Concedes and Con›ict can have no bearing on its node 2 choice. All of which suggests that the Protégé-Defender Subgame offers a reasonably close approximation of the alliance game played between Austria-Hungary and Germany in early July 1914.
5.3. Analysis Given the prima facie correspondence between the choices faced by the leadership groups in Vienna and Berlin in early July and those of the players in the Protégé-Defender Subgame, it is entirely reasonable to expect that one could glean an understanding of the event in question by exploring the subgame’s equilibrium structure (see table 5.2). Indeed, Austria’s behavior in the aftermath of Sarajevo is entirely consistent with the expectations of the Tripartite Crisis Game model and is not dif‹cult to explain. At the very ‹rst available opportunity (June 30), Austria’s foreign minister, Berchtold, met with the emperor, Franz Joseph, to chart a course of action. What followed was “a series of steps, all conscious and calculated, all designed to bring about war with Serbia” (Tunstall 2003: 131). That this is so is not very surprising. Berchtold, who previously had not been among those at the Ballhausplatz advocating a hard-line Serbian policy, now came to believe that the time had come for a “‹nal and fundamental reckoning” with Serbia (Kautsky 1924: no. 7). Even still, he was among the more temperate members of the empire’s ruling clique (Evans 1988: 34–36). For ex-
Vienna, Berlin, and the Blank Check
101
ample, on June 29, the day immediately after the assassination, General Conrad urged Berchtold to take “immediate measures,” by which he meant full mobilization, against Serbia (Geiss 1967: 64).7 The lone holdout among members of the Common Ministerial Council, the dual monarchy’s governing agency, was Hungary’s minister-president, Count István Tisza, who had a veto over any war decision. Nonetheless, the center of opinion among the makers of Vienna’s foreign policy strongly favored a hard-line policy. The overwhelming consensus seemingly ruled out the possibility, initially at least, of play taking place under a Settlement Equilibrium, leaving three other logical strategic combinations, each of which was in fact consistent with the confrontational approach favored by Berchtold. Table 5.2 shows, however, that an unquali‹ed (i.e., yD or yL = 1) hard-line policy by Protégé (i.e., Austria) is not supported in equilibrium unless there is also ‹rm and unequivocal (i.e., zS = 1) support from a staunch Defender, except when play takes place under a Bluff Equilibrium. There is little reason to believe, however, that a Bluff equilibrium was anything other than a theoretical possibility in July 1914. Recall from the discussion in Chapter 4 that a Bluff Equilibrium can exist only when it is more likely than not that Protégé is loyal, which was clearly not the perception in Berlin after Sarajevo. Indeed, Germany’s chancellor, Theobald von Bethmann Hollweg, believed that AustriaTABLE 5.2. Plausible Perfect Bayesian Equilibria and Existence Conditions of Protégé-Defender Subgame with Incomplete Information Strategic and Belief Variables Protégé Equilibrium
Defender
Existence Conditions
yD
yL
zS
zP
q
Settlement
0
0
1
0
>d1
Separating
1
0
1
0
1
Hold Firm
1
1
1
0
pPro
pDef > e1 pPro > d1
Bluff
1
•
—
0
d1
pDef > e1 pPro < d1
pDef < e2 e2 < pDef < e1
Note: • = fixed value between 0 and 1; — = value not fixed although some restrictions apply.
7. The general’s attitude was not new and therefore cannot be attributed to the assassination. According to Strachan (2001: 69), “Conrad ‹rst proposed preventive war against Serbia in 1906, and he did so again in 1908–9, in 1912–13, in October 1913, and May 1914: between 1 January 1913 and 1 January 1914 he proposed a Serbian war twenty-‹ve times.”
102
the games of july
Hungary might be de facto disloyal (see below). Such a belief is inconsistent with the existence of a Bluff Equilibrium and the probabilistic support by a staunch Defender associated with it.8 If the Bluff Equilibrium is eliminated on empirical grounds, only two other perfect Bayesian equilibria, Separating and Hold Firm, remain as possible descriptors of the game played between Austria-Hungary and Germany during the Austro-Serbian stage of the crisis, which ended, according to Farrar (1972: 9), on July 22, the day before Austria delivered its ultimatum. Unlike the Bluff Equilibrium, neither can be excluded empirically, since either equilibrium form is possible as long as Protégé is likely disloyal. For either the Hold Firm Equilibrium or the Separating Equilibrium to exist, however, the probability that Defender is staunch must not be inconsequential. Given this requirement, it should come as no surprise to learn that almost immediately after the assassination, Count Alexander Hoyos, Berchtold’s chef de cabinet, was sent to Berlin to represent the Austrian position and to sound out the likely German response. Vienna’s decision makers had several good reasons to be uncertain about their ally’s position. On at least two previous occasions, in 1912 and again in 1913, Kaiser Wilhelm had failed to stand with Austria in a crisis. In consequence,“the alliance with Germany was shaky” (Williamson 1983: 21). In addition, Berlin’s ambassador to Vienna, Baron Heinrich von Tschirschky, was urging caution (Kautsky 1924: no. 7). Hoyos arrived in Berlin on July 5, bringing with him a hastily rewritten version of the Matscheko Memorandum that prescribed a military solution to the Serb “problem” (Kautsky 1924: no. 14). By sharing this document with the Wilhelmstrasse, Berchtold hoped to signal Vienna’s strong policy preference. Hoyos also brought a personal, handwritten letter from Franz Joseph that asked what Germany would do if Austria moved against Serbia (Kautsky 1924: no. 13). The obvious implication was that Austria would implement its hardline policy if Germany would support it—that is, if the German government was staunch. According to Tunstall (2003: 135), Hoyos’s “secret mission had only one purpose: war. There was no talk of deterrence, much less of peace.” In the end, Vienna got what it wanted. After a slight hesitation, the kaiser, in what has since become known as the blank check, told Austria’s ambassador to Berlin, Count László Szögyény-Marich, that Austria could “rely on Germany’s full 8. Austria’s failure to strike quickly, however, eventually raised some doubts in the German Foreign Of‹ce about the seriousness of Vienna’s intentions (Fromkin 2004: 176).
Vienna, Berlin, and the Blank Check
103
support” even in the unlikely event of war with Russia (Geiss 1967: 76–77).9 The next day, Bethmann Hollweg seconded the kaiser’s sentiments and, like Wilhelm, urged “immediate action” (Geiss 1967: 79). Clearly, Hoyos’s mission had been accomplished. Why did Germany, in effect, offer Austria carte blanche support? A number of compelling reasons can be offered. But one stands out precisely because it is anticipated by the equilibrium structure of the Protégé-Defender Subgame of the Tripartite Crisis Game model. As Kagan (1995: 191) explains, Bethmann Hollweg realized that “a failure to act could lead to the defection of Austria from the German alliance or its collapse.” In other words, there was a very high probability that Austria would either choose to realign at node 4a or, for Germany what amounted to the same thing, cease to operate as a great power. If it were the latter, Austria was de facto disloyal in 1914. What is striking about the level of support that Austria received is that it, too, is entirely consistent with the expectations of the Tripartite Crisis Game model. For reasons explained above, it is unlikely that the conditions in July 1914 were such that a Bluff Equilibrium could exist. In addition, the Austrians were disinclined to play under a Settlement Equilibrium. This leaves, as theoretical possibilities, either a Separating or a Hold Firm Equilibrium as a potential descriptor of actual behavior. Under either equilibrium form, however, Defender either supports Protégé completely or not at all—that is, if Defender is staunch, zS = 1, and if Defender is per‹dious, zP = 0. When viewed in this context, then, the blank check is not very dif‹cult to understand. It is tempting to speculate counterfactually about what Austria would have done if the kaiser had not offered his unconditional support. It is of course possible that Austria would have moved anyway. After all, Berchtold was being pushed in that direction by Conrad and all the members of the Common Ministerial Council except Tisza. And Berchtold himself was leaning toward a confrontation. To take decisive action against Serbia in the absence of a German guarantee, however, Austria would have had to have missed a very strong signal about Defender’s type. Notice that in the Protégé-Defender Subgame, only per‹dious Defenders withhold support completely at node 3. There is only one perfect Bayesian equilibrium that can exist when Defender is very likely per‹dious—Settlement. Under a Settlement Equilibrium, Protégé, regardless of type, concedes at node 2. Thus, there are compelling the9. The kaiser’s perception was that Russia would stand aside. As chapter 6 shows, this initial belief helps to explain Germany’s reaction to the partial mobilization of Russia’s army on July 29.
104
the games of july
oretical reasons to believe that Austrian decision makers would have reconsidered their policy options and behaved in a much more temperate way had German leaders reacted differently. Several empirical reasons exist as well. For one, Austria had previously abandoned plans to push Serbia hard when Germany demurred. Moreover, Franz Joseph favored a peaceful resolution of the crisis. And ‹nally, while Berchtold seemed to be inclined toward a hard-line policy, he did not seem unconditionally committed to acting alone—that is, without Germany’s backing. An interesting question that should be addressed at this point is why, in formulating their policy, Austria’s policymakers seemed to ignore the possible reaction of Russia, why they “acted as if Russia did not exist.” Of course, no de‹nitive answer can be given. But Jannen (1983: 74) speculates that Berchtold and others in the inner circle simply succumbed to psychological stress: It has been argued here that Austro-Hungarian decision-makers were responding to real problems and real threats, but that they were responding to them unrealistically. They had been subject to accumulating stress and fears from a wide range of sources long before the assassination and were seeking to reduce stress through a variety of psychological mechanisms. After the assassination, particularly given the symbolic and emotional importance of such an event, they could not tolerate the further stress entailed by uncertain negotiations over uncertain solutions. The assassination therefore acted as an immensely powerful catalyst that both raised their fears and anxieties to levels that burst the restraints that had hitherto contained them, and presented an external enemy, Serbia, upon whom such fears and their resultant aggression could be discharged. In the face of the psychological needs thus generated, war with Russia did not matter.
It is dif‹cult to take issue with Jannen’s assessment that decision makers in both Vienna and Berlin operated in a highly charged psychological environment, as Holsti, North, and Brody (1968) established long ago in their classic study of the July Crisis. But it does not follow that the mere existence of stress can explain the apparent oversight in Austrian preparation. For one, Jannen’s argument is seemingly at odds with the detailed planning at the Ballhausplatz that took account not only of the impact of mobilization on the dual monarchy’s harvest but also of the whereabouts of the French president and prime minister. In addition, Jannen’s explanation ignores the purpose of Hoyos’s
Vienna, Berlin, and the Blank Check
105
mission—to determine Germany’s likely response to a Russian attack, either actual or threatened. Still, the documentary record does not indicate that any prolonged discussion of St. Petersburg’s attitude took place in Vienna. One reason may be that Berchtold and those around him simply did not care what Russia did given that Germany had their backs. Levy’s (1990/91: 162) argument that Austria preferred either a localized war with Serbia or a continental war that included Germany, Russia, and France to a brokered peace is consistent with this possibility. Of course, this explanation merely begs the question of why no detailed consideration took place of what Great Britain would do if a continental war occurred. Another explanation that is consistent with both the Tripartite Crisis Game model and the short shrift given to Britain’s attitude is that Austria simply assumed that Germany’s support would deter Russia. After all, in 1909, when Germany threatened war, Russia stood aside and recognized the Austrian annexation of Bosnia-Herzegovina. There was no good reason for Vienna to believe that Russian policymakers would act differently this time around. On July 5 at least, the kaiser opined that Russia “was in no way prepared for war, and would think twice before it appealed to arms” (Geiss 1967: 77). After the war, Bethmann Hollweg (1920: 126) admitted that he did not think that Russia would intervene. It should not be very surprising, therefore, to ‹nd that this sentiment was shared by the Viennese foreign policy elite. It was certainly Conrad’s view. His preferred plan of attack (Plan B), which was directed primarily against Serbia, clearly rested on this premise (Fromkin 2004: 204). In any event, the course of Austrian policy was set at a July 7 meeting of the Common Ministerial Council. The of‹cial protocol of that meeting reveals that Tisza still had reservations; the Austrian foreign minister, however, did not consider them to be material. According to Berchtold, “An agreement has been arrived at, since the propositions of the Hungarian Premier would in all probability lead to a war with Serbia” (Geiss 1967: 87). To restate all this in a slightly different way, an equilibrium had been settled on. For its part, Austria-Hungary decided to hold ‹rm—with certainty. After the meeting, steps were taken to draft an ultimatum that no sovereign state could accept. Austria’s ‹rm stand was matched with unquali‹ed German support, support which Berlin was more than willing to provide. Of course, the blank check was predicated on the assumption that Vienna would move
106
the games of july
‹rmly and quickly. Immediate action, those at the Wilhelmstrasse thought, would enable Austria both to subjugate Serbia and, even more important, to increase greatly the probability that Austria’s war with Serbia would remain localized. Given the expectation of rational behavior drawn from an examination of the Protégé-Defender Subgame, none of this should come as a shock. As William Riker once put it, albeit in an entirely different context, “Many institutional forces, personal idiosyncrasies, and so on were doubtless involved in bringing about this outcome. . . . Nevertheless it also appears rationally best to those who have the power to bring it about, and from this perspective of history, therefore, it appears natural” (Riker and Ordeshook 1973: 200–201).
5.4. Coda The blank check plays a central role in most histories of the July Crisis. But the causal signi‹cance of the event that led to it, the double murder in Sarajevo, remains exceedingly contentious. At one extreme are those who see the war as highly contingent and hence altogether avoidable. For example, Lebow (2000–2001: 607), employing what he describes as a “minimal rewrite” counterfactual thought experiment, argues that “without the assassinations there would have been no war in the summer of 1914.” Thus, for Lebow and others (e.g., B. Schmitt 1944: 173) who argue that the war was not inevitable, the archduke’s death constitutes both a necessary and suf‹cient causal incident. At the other extreme are those who hold that the war was overdetermined and hence all but unavoidable. In this view, Sarajevo was but a catalyst for the war. Since catalysts, like streetcars, come along all the time, the Thronfolger’s murder cannot be considered a necessary condition for war in 1914. But it was an all-but-suf‹cient condition, given that the necessary structural conditions were already in place (Schroeder 2007; Thompson 2003: 462). An informal analysis of the Tripartite Crisis Game model provides one way to penetrate this controversy. Assuming that the model represents a rough approximation of the triangular strategic relationship among Germany, AustriaHungary, and Russia from Bismarck’s era forward, one can ask whether, over time, the model’s theoretical expectations have been satis‹ed, and under what conditions. Next, one can compare the conditions that existed in July 1914 to those that brought about other behavioral patterns. In the end, theoretically contingent insights about the necessity and suf‹ciency of the murder in Sarajevo can be inferred. We begin with what we know, starting in 1879. At that time, Bismarck was
Vienna, Berlin, and the Blank Check
107
worried that Russia might attempt to achieve a realignment by aggressively challenging Austrian interests in the Balkans. If so, there were no good outcomes for Germany. One possibility was that Germany would be drawn into an unwanted war with Russia. The other was an Austrian and Russian rapprochement that would clearly pose a strategic threat to Germany. Thus, to help stabilize the status quo, Bismarck entered into a defensive alliance with Austria. Long term, the alliance achieved its objectives. The competition between Austria and Russia in the Balkans was successfully managed until 1908, when behavioral patterns changed suddenly, along with the underlying conditions associated with them. In what has since come to be known as the Bosnian Crisis, the latent Russian threat to Austria was realized, shifting actual play away from the status quo into the Protégé-Defender Subgame. As would also be the case in June 1914, however, the challenge was anything but premeditated. The botched implementation of a deal between the Austrian and Russian foreign ministers led to the formal annexation of Bosnia and Herzegovina by the Austrian government in October 1908. The Russian foreign minister, Alexander Izvolsky, had secretly agreed to support the annexation in return for Austrian support of Russia’s right of free passage through the Bosporus Straits and the Dardanelles. But the Russian government repudiated both the deal (when it became public) and Izvolsky. An unintended crisis emerged, therefore, when subsequently Russia refused to recognize the Austrian fait accompli. It was at this point that the dynamics of the Protégé-Defender Subgame took over. As in 1879 and again in 1914, “there was some fear in Germany that Austria might defect to France and Russia if not supported” (G. Snyder and Diesing 1977: 540). Thus, it is not at all surprising that Germany backed Austria to the hilt. Forcing the issue in March 1909, Germany issued what amounted to an ultimatum. Unlike 1914, however, war was averted. Without ‹rm support from France, Russia had little choice but to back down. Still other behavioral patterns emerged in the years immediately preceding the Great War. For example, when in 1912 Serbian troops occupied the Albanian port of Durazzo during the First Balkan War, some members of the Austrian leadership group pressed for a military response. For this, though, they received little encouragement in Berlin (Clark 2000: 186–93). The reason, according to Wilhelm, was that there was “absolutely no risk for Austria’s existence or even prestige in a Serbian port on the Adriatic Sea” (Cowles 1963: 301–2). In October 1913, however, the German leadership group strongly backed an Austrian ultimatum demanding that Serbia’s troops vacate disputed
108
the games of july
territory in Albania. And the Serbs complied. At the time, according to Williamson (1991: 153), Wilhelm “was at his bellicose worst.” Nonetheless, in mid-June 1914, the kaiser again refused an Austrian request for unconditional support for an aggressive Balkan policy (Fromkin 2004: 98–99). Thus, throughout the prewar period, German support of Austria blew hot and cold, depending on how German policymakers assessed the threat to Austrian integrity and, not incidentally, the likely involvement of Russia (Clark 2000: 187). It is in the context of these other plays of the Tripartite Crisis Game that we can consider the causal signi‹cance of the double murders in Sarajevo. First, can the assassination be considered a necessary condition for the outbreak of World War I? The answer depends on whether the assassination is thought of as a perfect storm or as the last straw. Lebow (2000–2001) sees it as the former, Thompson (2003) and Schroeder (2007) as the latter. For Lebow, the assassination represented a clear challenge to the survival of the Austrian Empire, in›amed opinion in both Berlin and Vienna, and eliminated the most potent advocate of peace within the Hapsburg oligarchy. In this regard, Lebow is undoubtedly correct. But the question remains: Had the assassination not occurred, could these same or equivalent conditions have come to pass at some later time? To deny that they could have, as Lebow does, one must overlook the reality that similar conditions had in fact existed in the past, that the actual play of the Tripartite Crisis Game had indeed taken place within the theoretical con‹nes of the Protégé-Defender Subgame, and with a similar outcome. If German support of Austria was necessary for a great-power war in 1914, there is no good reason to believe that it could not have materialized again, as it had on several prior occasions. And if an Austrian determination to confront Serbia and/or Russia was a necessary condition for a European con›ict, there is also no reason to believe that it would not have surfaced once more, as it had previously. Thus, it is dif‹cult to agree with Lebow that no other provocation could have brought about the war. The assassination of Franz Ferdinand was simply not a necessary condition for a World War at the beginning of the twentieth century. But was it suf‹cient? Here it is dif‹cult to agree with Thompson (2003) and others who see World War I as inevitable or at least highly likely. For Thompson, the necessary structural conditions for war clearly existed in 1914. Great power rivalry density was at a maximum, alliance polarity was very high, and a power transition was in the works. In his view, then, Europe in 1914 was
Vienna, Berlin, and the Blank Check
109
a powder keg and Sarajevo was but one of many potential sparks, an all-butsuf‹cient but entirely unnecessary condition for major power war. As anyone who has struggled to light a ‹re can attest, however, not all sparks land. Many fail to ignite, even in an actual tinderbox. An equilibrium analysis of the Protégé-Defender Subgame suggests that several rational strategic possibilities existed in 1914 and that not all of them were associated with unconditional German support of Austria. Many of these other possibilities are re›ected in the historical record. To argue that the existence of necessary structural conditions made war highly likely if not inevitable after Sarajevo is to ignore these theoretical and actualized possibilities. Maybe it was an amazing stroke of luck that a European war did not occur in, say, 1913, when Austria, with strong German support, was prepared to evict Serbia from Albania. But the requisite structural conditions were no different immediately after the Second Balkan War than in July 1914 (Thompson 2003: 469). Why didn’t the powder keg ignite then? To say that luck prevailed is to offer no explanation at all. All of which is to say that the spark was not necessary (other sparks could have come along, as they had in the past) and the powder keg was not suf‹cient to bring about the Great War (the presence of a spark did not always cause this particular powder keg to combust). It is a mistake, therefore, to imbue the assassination of Franz Ferdinand with either too much or too little causal importance. Perhaps it is best to think of the archduke’s murder as but one of several important steps on the road to war (Senese and Vasquez 2008). For “the great seminal catastrophe” to have occurred, other steps had to be taken as well. To be sure, it helped (if that is the right word) that the blank check was issued—and cashed. Still, the local war in the Balkans also had to escalate to the level of a continental con›ict and then expand to the level of a worldwide con›agration. Any explanation of the Great War that excludes both the escalatory endgame brought about by the Russian partial mobilization decision and Britain’s concomitant failure to deter a German invasion of Belgium, therefore, is anything but complete. True, the road to a world war in 1914 started in Sarajevo, but it need not have begun there. Nor did the journey have to end the way it did. Lebow (2000–2001: 592) is both half right and half wrong. But so is Thompson (2003). While the war was most certainly highly contingent, it was also not overdetermined. As the next two chapters show, there were so many points at which events could have easily gone the other way. Of course, they did not. In what follows, I explain why.
Now or never. kaiser wilhelm ii The game is up. william shakespeare
chapter six
La Guerre Européenne To the casual observer, not much of note seemed to be happening in Europe immediately after the meeting of the Austro-Hungarian Common Ministerial Council on July 7, 1914. The kaiser, intent on not tipping Germany’s hand, had the day before departed on a planned cruise aboard his private yacht, the Hohenzollern. And other leaders, in both Berlin and Vienna, acted as if nothing was afoot. Behind the scenes, however, Austrian of‹cials were being encouraged, strongly and incessantly, by German policymakers “not to neglect the present moment, but to treat Serbia with full energy, so as to clear out the conspirators’ nest once for all” (Geiss 1967: 110). Nonetheless, Austria was slow to move. On July 14, Hungary’s minister-president, István Tisza, dropped his few remaining objections to the ultimatum. But the Ballhausplatz was still working on the document’s wording. Vienna obviously saw no need to rush. After all, the troops were still on leave and the French president and premier were not scheduled to leave St. Petersburg until July 23 (113–14).1 Toward the end of July, however, rumors began to swirl throughout Europe that Austria intended to seek restitution from Serbia. There were other indications as well (Rich 2003: 218). Thus, when the ultimatum was ‹nally delivered to Belgrade on July 23, soon after Raymond Poincaré and René Viviani had departed, Russian decision makers were hardly surprised. In fact, since Russia’s foreign minister, Serge Sazonov, feared that it was coming, he had already warned against it (Kautsky 1924: no. 120). The details of the ultimatum were formally conveyed to Sazonov by the Austro-Hungarian ambassador, Count Friedrich Szápáry von Szápár, early the 1. Vienna initially believed that the French delegation would leave on July 25.
110
La Guerre Européenne
111
next day. According to Szápáry, Sazonov “was more saddened and depressed than excited; the tactics he followed were not to commit himself with regard to Russia’s future attitude” (Geiss 1967: 175). Privately, however, Sazonov was more than agitated. As he exclaimed to his aide, Baron M. F. Schilling (1925: 28–29), “C’est la guerre européenne.” Szápáry’s démarche precipitated a ›urry of diplomatic activity in St. Petersburg. Sazonov had lunch at the French embassy with the French and British ambassadors, later joined by the ambassador from Rumania. He had already decided to take a hard-line stand against Austria and was looking for support from France and Britain. Sazonov got exactly what he wanted from the French. As George Buchanan, the British ambassador, reported to the British foreign secretary, Sir Edward Grey, “France would not only give Russia strong diplomatic support, but would, if necessary, ful‹ll all the obligations imposed on her by the alliance” (Gooch and Temperley 1926: 11: no. 101).2 The explanation for this coordinated stand can also be drawn from an analysis of the Protégé-Defender Subgame discussed in detail in chapter 5. The model applies here as well for two speci‹c reasons. It would only be played out if (1) Challenger (i.e., Austria) invaded Serbia (which it eventually did) and (2) if a con›ict was implied if both Defender (France) and Protégé (Russia) held ‹rm, which was also the case. Apparently, Russia and France had settled on an equilibrium (i.e., a joint strategy pair) similar in spirit, if not in detail, to the equilibrium negotiated during the Hoyos mission to Berlin: if pressed too far, Russia would hold ‹rm against Austria (with certainty) and France would support the Russian decision (also with certainty). For reasons that are fully explained in chapter 7, however, the British ambassador was noncommittal at the luncheon at the French embassy. As Buchanan reported to Grey, he had held “out no hope that [the British government] would make any declaration of solidarity that would entail engagement to support France and Russia by force of arms.” The best he could do was to suggest that Grey might be able to issue a vague warning to the Austrians and Germans that “it would be dif‹cult for England to remain neutral” should a European war break out. Finally, Buchanan speculated that “the moment had passed when it might have been possible to enlist French support in an effort to hold back Russia” (Gooch and Temperley 1926: 11: no. 101).3 2. Some historians (e.g., B. Schmitt 1930: 2:89) question whether the unequivocal support expressed by French ambassador Maurice Paléologue was consistent with his charge. (For a discussion, see Jannen 1996: chap. 25; Kiesling 2003: 246–47). The dispute has little bearing on this analysis. 3. The rest of this chapter is based on Zagare 2009a.
112
the games of july
After lunch, Sazonov convened the Russian Council of Ministers. As Spring (1988: 57) notes, this meeting was “the critical point for Russia in the July crisis.” The decisions reached that afternoon established the type of player Russia would be in the days that followed: Sazonov’s inclination toward a hard-line policy was supported by the rest of the government and on the following day was rati‹ed by the czar. The agreed upon strategy was multifaceted, covering a number of contingencies and revealing a clear hierarchy of objectives. More than anything, the Russians wanted to defuse the crisis before it further intensi‹ed. Accordingly, the Council of Ministers approved Sazonov’s proposal to ask Austria for an extension of the ultimatum’s deadline and to encourage Serbia to accede to as many of Vienna’s terms as possible. Of course, these measures could always fail, in which case Serbia was to be urged not to resist an Austrian invasion. As one might expect, the latter suggestion was not well-received by the Serbs (Stokes 1976: 70). The most important decision made on July 24, however, concerned what Russia would do if Austrian troops marched against Belgrade. According to an internal Foreign Ministry memorandum, “it was decided in principle” to implement a partial mobilization of the Russian army and navy “and to take other military measures should the circumstances so require” (Geiss 1967: 190). The next day, the czar formally endorsed this recommendation and agreed “to enforce throughout the entire Empire the order for the period preparatory to war” (207). As Trachtenberg (1991: 76) notes, with this decision, “the crisis had moved into its military phase.” Depending on circumstances, Russia was now prepared for either a limited response (i.e., a partial mobilization against Austria) or an escalatory response (i.e., a full mobilization against both Austria and Germany) should the need occur. Even still, as late as July 27, no European government had made any irrevocable choices. True, Austria-Hungary and Germany had decided on a joint course of action, and Austria had issued a demanding ultimatum, but no signi‹cant military steps had been taken. Similarly, Russia and France had decided to stand together, and Russia had developed a strategy that took various contingencies into account, but no overt military plan had been implemented. In other words, neither side had as yet mobilized, either fully or partially, for war. But the status quo would not long endure. After all, the Austrian ultimatum had already expired; Serbia had rejected its most humiliating conditions; the German Foreign Of‹ce was still urging immediate action; and Vienna remained intent on crushing the Serbian “viper.”
La Guerre Européenne
113
The Austrian intent was realized at noon on July 28 when a telegram declaring war was sent to the Serbian government. The next morning, AustroHungarian gunboats opened ‹re on Belgrade, the ‹rst of four critical moves in a game that would, in short order, lead to a European war that pitted the Dual Alliance of Austria-Hungary and Germany against Russia and France. (The expansion of this con›ict into a world war that also involved Great Britain is discussed in chapter 7.)
6.1. Asymmetric Escalation Game To understand the dynamic process that eventually, albeit brie›y, involved four of the European great powers in a continental con›ict, consider now the Asymmetric Escalation Game depicted in ‹gure 6.1, which is analyzed below with incomplete information. This model, which is another component game form of perfect deterrence theory, was speci‹cally designed to analyze intense interstate disputes that involve at least two distinct levels of con›ict, one limited and the other unlimited.4 The empirical ‹t between this model and the events that led to the European phase of the war is especially close. It therefore constitutes a powerful tool for understanding the escalatory process that led, eventually, to the initiation of armed con›ict by Germany in early August 1914. The Asymmetric Escalation Game is a two-player noncooperative game. In what follows, I associate Austria-Hungary and Germany with the player called Challenger,5 and Russia and France with the player called Defender. The assignment of these roles to the two players is not dif‹cult to justify. Clearly, the Austrian declaration of war against Serbia was seen, at least in St. Petersburg, as a direct challenge to Russia’s standing as a great power. Moreover, the ‹rst signi‹cant Russian decision was doubtless a defensive reaction to the Austrian declaration of war and the bombardment of Belgrade. More dif‹cult to justify, however, is the assumption that (1) Austria-Hungary and Germany constituted a single decision-making unit and (2) that Russia and France made concerted choices from July 28 and afterward. Indeed, neither assumption is entirely consistent with the known facts. Decision mak4. For the application of the model to NATO’s 1999 war with Serbia over Kosovo, see Quackenbush and Zagare 2006. 5. Austria-Hungary is listed as the ‹rst (principal) member of the coalition. This is no accident. For all intents and purposes, the German government played a subordinate role in actual decision making prior to the Belgian invasion on August 4. And by that time, the die had already been cast. As Williamson (1991: 196) notes, “The steps that pushed Europe toward war were taken in Vienna. The support given by Berlin simply con‹rmed and assured that the Hapsburg decision to settle accounts would this time be a military solution rather than a diplomatic one.”
114
the games of july Challenger
C (1 – x)
Node 1
D (x)
Defender
Status Quo (cSQ, dSQ) C (1 – y – z)
E (z)
D (y)
Challenger
Defender Concedes (cDC, dDC)
D (1 – w)
Defender’s first-level threat
Node 2
Nodes 3a, 3b
E (w)
Limited Conflict (cDD, dDD)
D
Defender Escalates (Wins) (cDE, dDE)
Defender
D
Challenger Escalates (Wins) (cED, dED)
Challenger
E
Challenger’s escalatory threat
E
All-Out Conflict (cEE, dEE)
Node 4
Defender’s second-level threat
All-Out Conflict (cEE, dEE)
___________ Key: Challenger: C = Cooperate D = Demand E = Escalate Defender:
x = probability Challenger initiates at node 1 w = probability Challenger escalates at node 3a y = probability Defender responds in kind at node 2 z = probability Defender escalates at node 2
C = Concede D = Defy E = Escalate
Fig. 6.1. Asymmetric Escalation Game
ers in Vienna and Berlin, for example, had slightly different objectives,6 sometimes possessed private information that they did not share with each other, and did not always operate under the same set of constraints.7 Similarly, in the 6. Berlin sought to preserve its alliance with Austria; Vienna sought to preserve its standing as a great power. Nonetheless, in the game they played with Russia and France, their preferences converged. 7. Equally problematic is the assumption that the German government acted as a single unit throughout the crisis. But since the divergent tendencies and underlying preferences of Kaiser Wilhelm II, Chancellor Theobald von Bethmann Hollweg, foreign secretary Gottlieb von Jagow, chief of the General Staff Helmuth von Moltke (the younger), and others surfaced when decision making was still centered in Vienna, the lack of coordination among German leaders is not particularly material to this discussion.
La Guerre Européenne
115
actual play of the game, all of Defender’s critical choices were made in St. Petersburg, not in Paris. In other words, France, which Remak (1971: 354) judges as the nation “least responsible for the outbreak of the war,” was a relatively minor player during the most critical stages of the crisis. Nonetheless, a number of good reasons exist for treating the two coalitions as a single player. One important justi‹cation is that members of both alliances had agreed on a joint strategy well before Austria’s declaration of war. As chapter 5 shows, given German support, Austria-Hungary had decided to take a hard-line approach toward Serbia. Similarly, with France’s backing, Russia was determined to stand ‹rm and resist an Austrian challenge to Serbia’s political integrity.8 In addition, treating the two alliances as a unit considerably simpli‹es the analysis. Ample precedent exists for this particular player assignment. In their classic study of interstate crises, for example, G. Snyder and Diesing (1977: 94) view each coalition as a uni‹ed actor. And in most empirical studies of interstate disputes, Germany and Austria-Hungary are grouped together as a single entity during the July Crisis.9 On the other hand, there will be times when we must stand back from this simplifying assumption—just a bit—to get a clearer picture of how and why the game terminated as it did. Doing so will become especially necessary after July 29, when decision makers in Vienna and Berlin began to interpret the world differently and, in consequence, to work at cross-purposes.10 It was at this point in the game that Germany lost control of its ally, that critical decision-making was thoroughly monopolized by the Hapsburg monarchy. These reservations aside, the Asymmetric Escalation Game is particularly well suited for examining situations, like the one that existed at the end of July 1914, where more than one type of con›ict outcome is possible. It is, therefore, a much more re‹ned model than either the Generalized Mutual Deterrence Game or the Unilateral Deterrence Game discussed in detail in chapter 3 or most other game-theoretic models of the escalation process. In the Asymmetric Escalation Game with incomplete information, Challenger begins play at node 1 by deciding whether to initiate a crisis. If Challenger turns away from a confrontation (by choosing C), the Status Quo (SQ) 8. This reason and convention are the primary reasons why France is considered a participant in the game. For all intents and purposes, this particular assumption is innocuous. 9. See, e.g., Danilovic 2002; Huth 1988. 10. This is not to say that their preferences diverged. Levy is correct (1990/91: 162) when he concludes that Austria-Hungary and Germany had identical (ordinal) utility functions. Despite a commonality of purpose, however, German leaders clearly placed a much higher probability on the likelihood that Russia would intervene than did Austria-Hungary after Russia ordered partial mobilization on July 29. More technically, the (updated) beliefs of Germany’s leaders about Russia-France’s type differed from those held by the Austrian foreign minister, Count Leopold Berchtold, and other key decision makers in Vienna.
116
the games of july
obtains. But if Challenger initiates con›ict (by choosing D), Defender decides (at node 2) whether to capitulate (by choosing C) or to respond; if the latter, Defender also decides whether to respond in kind (by choosing D) or to escalate (by choosing E). Capitulation ends the game at Defender Concedes (DC). If Defender responds, Challenger can choose whether to escalate at nodes 3a and 3b. If Challenger is the ‹rst to escalate (at node 3a), Defender has an opportunity at node 4 to counterescalate. Limited Con›ict (DD) occurs if Defender responds in kind and Challenger chooses not to escalate at node 3a. Challenger Escalates/Wins (ED) if, at node 4, Defender chooses not to counterescalate. Similarly, the outcome is Defender Escalates/Wins (DE) if Challenger chooses not to counterescalate at node 3b. All-Out Con›ict (EE) results whenever both players escalate. As ‹gure 6.1 indicates, there are two distinct paths to All-Out Con›ict in the Asymmetric Escalation Game. The connection between the initial choice facing Austria-Hungary and Germany in late July 1914 and Challenger in the Asymmetric Escalation Game should be obvious. After Sarajevo, the Dual Alliance could have either accepted a humiliating status quo by doing nothing (i.e., by choosing C) or sought to modify it by choosing D and demanding its alteration. Austria’s intent to contest the status quo was clearly signaled on July 23, when it issued its ultimatum to Serbia, and realized on July 28, when it ‹nally declared war. When Austria began to shell Belgrade on July 29, decision makers in St. Petersburg faced a set of options that closely parallel the choices available to Defender at node 2 of the Asymmetric Escalation Game: Russia could have stood aside and accepted (C) the Austrian attempt at a fait accompli, measured its response (D) with a partial mobilization directed only against Austria, or signi‹cantly escalated (E) the con›ict with a full mobilization of its army directed against both Austria and Germany.11 It was generally understood in St. Petersburg that an escalatory choice (i.e., full mobilization) was likely to lead to war with Germany (Trachtenberg 1990/91: 126), whether that escalatory choice was made initially at node 2 or subsequently at node 4. Unlike a full mobilization, however, a partial mobilization did not necessarily imply a war between Russia and Germany. On July 27, Germany’s foreign secretary, Gottlieb von Jagow, told both the French ambassador, Jules Cambon, 11. Jannen (1996: 279) characterizes the Russian general mobilization as “a major escalation of the crisis.” Langdon (1991: 60) and Trachtenberg (1990/91: 146) offer similar descriptions.
La Guerre Européenne
117
and the British ambassador, Sir Edward Goschen, that Germany would not mobilize against Russia if the Russian mobilization was directed only against Austria-Hungary, but that Germany would countermobilize if Russia subsequently mobilized against Germany as well (Geiss 1967: 245, 253). In other words, the implications of a partial mobilization were unclear. What is clear, however, is that in the event of a partial mobilization, the Austrian response would be critical, which helps to explain why the locus of decision making shifted from Berlin to Vienna after the czar backed away from full mobilization on Wednesday, July 29. We know that, empirically, the partial mobilization decision was implemented shortly after news reached the Russian capital that Austria-Hungary had declared war, and that a cause-and-effect relationship existed between these two events (Geiss 1967: 262). Indeed, as indicated above, the decision to react in this way had been made in advance, at the meeting of the Russian Council of Ministers on July 24. At this point, the choice facing Austria-Hungary mirrored the choice facing Challenger at node 3a of the Asymmetric Escalation Game. One option was simply to step back. For example, AustriaHungary could have accepted Grey’s July 24 proposal that “the four nations not immediately concerned—England, Germany, France, and Italy—should undertake to mediate between Russia and Austria” (Kautsky 1924: no. 157). Or Austrian leaders could have accepted the “Halt in Belgrade” proposal made by the kaiser on July 27 that Austria announce its intention to occupy the Serbian capital, but only until Serbia carried out the promises it had made in its response to the Austrian ultimatum (Kautsky 1924: no. 293). Or Austria could have taken Bethmann Hollweg’s strong hint on July 29 to defuse the crisis by entering into direct discussions with St. Petersburg (Kautsky 1924: no. 396). All of which indicates that, as late as July 29, a Limited Con›ict was a distinct possibility, that there was still a way out of the crisis if Austria-Hungary wanted one. But the Hapsburg Empire was not looking for an escape clause. Disregarding Berlin’s last-minute pleas to accept mediation (Kautsky 1924: no. 441), Austria-Hungary decided to plow on. After learning of the Russian partial mobilization, Austria mobilized against Russia while continuing its advance toward Serbia. By resisting mediation and pushing forward militarily, Austria-Hungary clearly intensi‹ed the con›ict. It did not take long for Russia to respond (see node 4). On July 30, the czar consented to full mobilization of Russian forces against both Germany and Austria. As Trachtenberg (1990/91) shows,
118
the games of july
the Russian mobilization decision was a war choice. Before long, German troops were marching into Belgium.12 For generations, historians and political scientists have attempted to explain why. If a limited con›ict was a distinct possibility, why did it fail to materialize? Was there an inevitable slide to an all-out war, as Britain’s chancellor of the exchequer David Lloyd George suggested after the war, or was the general European con›ict preventable? What role did perceptions play in the way the game played out? In what follows, I answer these and related questions by examining the equilibrium structure of the Asymmetric Escalation Game with incomplete information.
6.2. Preferences The Asymmetric Escalation Game has six possible outcomes. The game tree of ‹gure 6.1 shows these outcomes and the choices that lead to them. Note once again that the model admits two distinct con›ict outcomes. Limited Con›ict occurs only when Challenger defects at node 1, Defender responds in kind at node 2, and Challenger chooses not to escalate at node 3a. All-Out Con›ict occurs whenever both players escalate. It is also worth pointing out that the Asymmetric Escalation Game has two distinct paths to All-Out Con›ict. The ‹rst results when Defender escalates immediately at node 2 and Challenger retaliates at node 3b. The second path conforms to a classic escalation spiral: Challenger initiates at node 1, Defender responds in kind at node 2, Challenger escalates ‹rst at node 3a, and Defender counterescalates at node 4. Table 6.1 summarizes the preference assumptions that guide the analysis of the Asymmetric Escalation Game. For the most part, these assumptions are straightforward and fully consistent with those used to analyze all of the other 12. The alert reader will notice that as drawn, the game tree of the Asymmetric Escalation Game provides Challenger with an opportunity to counterescalate (at node 3b) should Defender escalate ‹rst at node 2. In the context of the events of July 1914, this decision node models a possible response to a full mobilization by Russia immediately after the Austro-Hungarian declaration of war against Serbia. By contrast, there is no analogous decision node for Challenger after Defender’s escalatory choice at node 4. The implicit assumption in the latter instance is that a full Russian mobilization simply implies a (German) countermobilization and hence an AllOut Con›ict between the Dual Alliance and Russia-France. This is clearly an inconsistency, but given the assumption made below that Challenger’s threat to counterescalate is highly credible, this inconsistency is of no analytic import. It could easily be eliminated either by adding another decision node after Defender’s escalatory choice at node 4, or by eliminating Challenger’s node 3b decision. I have not made these changes to the underlying game form for a number of reasons. First, as noted, the inconsistency is in no way material. Second, eliminating the inconsistency would likely complicate the narrative and the associated analysis. And third, redrawing the tree to eliminate the inconsistency would obscure (1) the model’s generality and (2) the relationship of the Asymmetric Escalation Game to the family of models that delineate perfect deterrence theory.
La Guerre Européenne
119
models presented here and, therefore, with the broad contours of perfect deterrence theory. For example, the preferences arrayed in table 6.1 presume that both players prefer winning to losing. To re›ect the costs of con›ict, the players are also assumed to prefer to win or, if necessary, to lose at the lowest level of con›ict. Thus, Challenger prefers Defender Concedes (outcome DC) to Challenger Wins (outcome ED), and so does Defender. Notice that the table leaves open several critical preference relationships. These relationships are associated with threats that the players may or may not prefer to execute. Challenger has only one threat: to escalate or not at node 3b. Defender, by contrast, has two threats: a tactical-level threat to respond in kind at node 2, and a strategic-level threat to escalate at node 4. As before, the preferences associated with these threats de‹ne each player’s type. Since Challenger has only one threat, it may be one of two types: hard Challengers prefer All-Out Con›ict to Defender Escalates; Challengers with the opposite preference are called soft. Defenders, by contrast, are more dif‹cult to typecast. A Defender that prefers Limited Con›ict to Defender Concedes is said to be hard at the ‹rst or tactical level while a Defender with the opposite preference is said to be soft at the ‹rst or tactical level. Similarly, a Defender that prefers All-Out Con›ict to Challenger Escalates is said to be hard at the second or strategic level, while a Defender with the opposite preference is said to be soft at the second or strategic level. Thus, Defender may be one of four types: hard at the ‹rst level but soft at the second (i.e., type HS); soft at the ‹rst level but hard at the second (i.e., type SH); of type HH: hard at both levels; or of type SS: soft at both levels. In the analysis that follows, all retaliatory threats are taken to be capable in the sense that the player that initiates con›ict ends up worse off if and when the other player retaliates. In terms of preferences, this means that both players prefer the Status Quo to Limited Con›ict, and Limited Con›ict to All-Out Con›ict. This ‹nal assumption about preferences, however, is neither innocuous nor noncontroversial—as I explain below. TABLE 6.1. Preference Assumptions for Asymmetric Escalation Game Challenger
Defender
Defender Concedes
Status Quo
Status Quo
Defender Escalates
Challenger Escalates
Defender Concedes or Limited Conflict
Limited Conflict
Challenger Escalates or All-Out Conflict
Defender Escalates or All-Out Conflict
120
the games of july
6.3. Some Caveats There are a few devils in the details of the preference assumptions I make that, perhaps, require exorcism. Before proceeding, however, I will comment, brie›y, on the connection between the six theoretical outcomes of the Asymmetric Escalation Game and the real-world events they are meant to represent. The outcome with the clearest meaning is the one labeled Status Quo, which I take to be the existing European order as of July 1914. As the situation stood shortly after Sarajevo, neither Austria-Hungary nor Germany placed a high value on this outcome, which provides further justi‹cation for the identi‹cation of the governments in Vienna and Berlin with the player called Challenger. Speci‹cally, German leaders looked around and saw both a dominating Great Britain and a rising Russia that was tied closely to France, a longtime German rival. For their part, Austria’s policymakers feared that their polyglot empire would implode if Serbian subversives were not soon eradicated. Clearly, both Austria-Hungary and Germany were dissatis‹ed powers as the July Crisis unfolded. Another outcome whose meaning should be clear is Defender Concedes. Defender Concedes is simply a more generic term for the outcome that Levy (1990/91) calls “localized war.” Defender Concedes, therefore, is intended to capture the denouement of a war in the Balkans that pitted Austria-Hungary against “tiny Serbia” (Geiss 1967: 363). Defender Escalates/Wins and Challenger Escalates/Wins refer to one-sided victories for Defender and Challenger, respectively, that come about after an escalatory move by one player and capitulation by the other. In the context of the July Crisis, Russia (i.e., Defender) would clearly have gained a political and diplomatic advantage had it implemented a full mobilization of its army and forced Austria-Hungary and Germany to back off. Similarly, Austria-Hungary and Germany (i.e., Challenger) would have gained the upper hand and probably split the Entente had the partially mobilized Russian army stood down as Belgrade was leveled and Serbia dismembered. The remaining two outcomes of the Asymmetric Escalation Game have names that may mislead the reader. Although in the next chapter I use the term All-Out Con›ict to refer to a world war that includes Germany, Austria-Hungary, Russia, France, and Great Britain, I use it differently here. Within the con‹nes of this chapter, All-Out Con›ict corresponds to the con›ict Levy (1990/91) refers to as a “continental war”—that is, a war in which only the four
La Guerre Européenne
121
major continental European powers and Serbia are involved. This, of course, is the war that broke out in Europe after Germany declared war on Russia on August 1. The ‹nal outcome of the Asymmetric Escalation Game, Limited Con›ict, requires a most careful exegesis. Normally, this term is reserved for an actual war in which two or more nations ‹ght but at least one of the involved nations has either a limited objective or fails to use all the weapons at its disposal. But here, Limited Con›ict refers to any outcome that results after Challenger contests the status quo, Defender measures its response, and Challenger decides not to escalate. In 1914, for example, a Limited Con›ict would have evolved had Austria-Hungary agreed to mediate its dispute with Serbia after Russian troops were mobilized in the Balkans. The broad outlines of this outcome, therefore, correspond closely to the outcome Levy (1990/91) identi‹es as “negotiated peace.” All this said, one might well ask whether the preference assumptions summarized in table 6.1 stand up to empirical scrutiny. After all, these are generic preferences that were developed to represent an interesting and important general case. It is more than possible, therefore, that some critical differences exist between these (posited) preferences and those of the actual players during the July Crisis. For example, the Defender in our model strictly prefers the Status Quo to any other outcome. But was Russia truly a satiated power in 1914? Schroeder (1972: 335) makes a compelling case that it was not,13 that it had designs on large swaths of Austrian territory, and that it was simply waiting for the aging emperor’s death to annex Galicia and other parts of Franz Joseph’s sure-to-disintegrate empire. Clearly, Russia was also a dissatis‹ed power on the eve of World War I. However, Russia’s territorial ambitions are largely immaterial for our purposes. One reason is that in the Asymmetric Escalation Game, Defender never chooses between the Status Quo and any other outcome, so that its relative ranking is theoretically irrelevant. In addition, unless one wishes to argue that Russia-France deliberately provoked the crisis in order to humiliate AustriaHungary and Germany by forcing the Dual Alliance to back down, implying that Defender preferred Defender Escalates/Wins to the Status Quo, the assumption that Russia-France’s highest-ranked outcome was the Status Quo is 13. See also Butter‹eld 1965.
122
the games of july
entirely defensible within the context of the Asymmetric Escalation Game and the set of outcomes associated with it. Theoretically or empirically, no other outcome could be ranked higher. But F. Fischer (1967; 1975) would most certainly object to the relatively low ranking of All-Out Con›ict (i.e., a continental war) by Challenger. In Fischer’s opinion, Germany deliberately provoked war in 1914 in a bid for world power. If Fischer is correct—and some scholars believe he is—All-Out Con›ict was Germany’s highest-ranked outcome.14 In consequence, deterrence by Russia and France (without the assistance of Great Britain) would have been impossible. In other words, Fischer’s assumption about German preferences, in and of itself, constitutes a suf‹cient condition for the outbreak of war in Europe on August 1. Both Schroeder (1972: 336–37) and Remak (1971: 361), however, agree that Fischer’s argument is not necessary for an explanation of the escalation spiral that eventually led to World War I. I later demonstrate why they are correct, but for now, in the tradition of William of Ockham, I simply adopt the less demanding assumption.15 Levy’s (1990/91) contention that both Austria-Hungary and Germany preferred All-Out Con›ict (i.e., a continental war) to Limited Con›ict (i.e., a negotiated peace) also runs counter to the preference assumptions arrayed in table 6.1.16 With respect to Austria-Hungary, Levy’s conclusions are debatable but dif‹cult to establish. To be sure, as the crisis intensi‹ed, Vienna did everything it could to avoid mediation. But did it take this tack because it preferred a continental war or, as I subsequently argue and as Jannen (1996) forcefully contends, because it did not believe that Russia would intervene? With respect to Germany, Levy’s conclusions are even more problematic unless one is willing to discount completely the sincerity of Bethmann Hollweg’s frantic, last-ditch effort to moderate Austria-Hungary’s behavior (discussed below). Even Immanuel Geiss (1967: 88), Fischer’s student and disciple, is unwilling to go that far. Some readers may also ‹nd fault with the fact that the Asymmetric Escala14. Copeland’s (2000: 57) argument that Germany “preferred major war to even a localized war or a negotiated solution” is even more extreme. With respect to Moltke and other German military leaders, Copeland may well be correct. But as Copeland (2000: 59–60) and others (e.g., Williamson and May 2007: 363) note, the military did not control German policy in 1914. Had the military been in control, the war most likely would have come sooner, perhaps as soon as 1875, when Moltke (the elder) ‹rst proposed a preemptive war against France (Förster 1999: 351). 15. Stone (2009: 23–25) also argues that Germany wanted a European war in 1914. 16. Levy (1990/91: 162) claims that Fischer would also argue that both Austria-Hungary and Germany preferred All-Out Con›ict to Limited Con›ict.
La Guerre Européenne
123
tion Game, and the outcomes associated with it, does not include Great Britain as a player and the possibility of a wider, world war, which eventually broke out when Britain declared war on Germany. I have chosen not to model Great Britain’s choices in this chapter in part because I seek to explain, in the simplest possible way, the escalation spiral that led to the outbreak of war on the continent. Including Great Britain among the players would only unnecessarily complicate matters. In addition, I separately analyze the game played between Great Britain and Germany in 1914 in chapter 7. It would be a straightforward exercise to include that game as proper subgames of the Asymmetric Escalation Game.17 But doing so would not alter my argument here. Finally, Challenger’s postulated preference for the Status Quo over a Limited Con›ict (i.e., a negotiated peace) also runs counter to both Fischer’s and Levy’s assessments of Austro-Hungarian and German preferences. In the analysis of the Asymmetric Escalation Game, I assume that both Challenger and Defender possess capable threats at every level of play. Consistency with this assumption requires that Challenger prefer the Status Quo to Limited Con›ict. How critical is this assumption? It plays a relatively minor role in this analysis. After all, the crisis ultimately escalated to the highest level, suggesting that Challenger’s relative ranking of these two outcomes was of little moment. On the other hand, it has important implications for how a hypothetical question about the inevitability of con›ict in 1914 is answered (see section 6.4.2).
6.4. Analysis With two players, each with at least two moves, and six outcomes, the Asymmetric Escalation Game with incomplete information can have many variants. To gain tractability, several assumptions about the players’ preferences have already been made. Nonetheless, these assumptions about preferences are not stringent enough to make the results easy to summarize or intuitive to understand. One more assumption, however, should do the trick. In what follows I focus on the special case in which Challenger is likely hard—that is, when Challenger’s threat to counterescalate at node 3b is highly credible. This special case analysis is easy to justify. First, although this assumption vastly simpli‹es the analysis of the Asymmetric Escalation Game, it does so 17. Notice the plural. To completely extend the Asymmetric Escalation Game, the game developed in the next chapter would have to be substituted for Challenger’s node 3b decision and appended subsequent to Defender’s decision at node 4.
124
the games of july
with no serious loss of information. True, the absolute number of perfect Bayesian equilibria is fewer in the special case analysis, but that is the point.18 All distinct equilibrium forms of the general case are represented in the special case analysis. Moreover, the equilibria that arise when Challenger is likely to be hard exist under conditions that fully exemplify the existence conditions in the general case. In other words, little is gained theoretically by examining the Asymmetric Escalation Game in the absence of this simpli‹cation. The more important justi‹cation, however, is empirical. The assumption that Challenger is likely hard is entirely consistent with the facts on the ground at the end of July 1914. According to Berghahn (1993: 197), the “hard-liners” were in control in Germany. And as the crisis unfolded, both the Russians and French took it for granted (1) that Austria-Hungary had Germany’s backing and (2) that a full Russian mobilization implied a general European war. Neither would be reasonable inferences if Challenger (i.e., Austria-Hungary/Germany) was seen as likely to be soft. With these justi‹cations in hand, I now analyze the equilibrium structure of the Asymmetric Escalation Game with incomplete information. Knowing the theoretical characteristics of the perfect Bayesian equilibria (see table 6.2) TABLE 6.2. Perfect Bayesian Equilibria of the Asymmetric Escalation Game when Challenger Has High Credibility Challenger x xH
Defender
w xS
wH
y wS
qHH
yHH
z yHS
zHH
zHS
zSH
zSS
r
Escalatory Deterrence Equilibria (typical) Det1
0
0
1
1
Small
0
0
1
1
1
1
≤ d1
0
0
0
0
pCh
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
≥ d2 ≥ d2 pCh pCh
No-Response Equilibrium NRE
1
1
Large
Small
0
0
Spiral Family of Equilibria Det2 Det3 CLRE1 ELRE3
0 0 1 1
0 0 1 1
0 d*/r 0 d*/pCh
0 0 0 0
pStr|Tac cq pStr|Tac cq
1 1 1 1
1 v 1 v
18. Sixteen rational behavioral possibilities exist when no restriction is placed on Challenger’s type, but only six exist in the special case. The reader interested in an analysis of the general case should consult Zagare and Kilgour 2000: app. 8. See also Kilgour and Zagare 2007.
La Guerre Européenne
125
that exist when Challenger is likely hard will enable us to address a number of interesting questions about the events that led to the outbreak of war in Europe in early August. As table 6.2 reveals, there are six rational behavioral possibilities when it is highly likely that Challenger is hard.19 These six perfect Bayesian equilibria can be conveniently placed into three major groups: the Escalatory Deterrence Equilibria, the No-Response Equilibrium, and the Spiral Family. At least one member of the ‹rst two categories always exists in the Asymmetric Escalation Game, as does precisely one member of the Spiral Family. Next I brie›y describe the de‹ning features of each distinct equilibrium form.
6.4.1. escalatory deterrence equilibria The ‹rst group of deterrence equilibria is a family of several perfect Bayesian equilibria called the Escalatory Deterrence Equilibria. Typical of this family is Det1 , under which Challenger, regardless of its type, never initiates con›ict and Defender, regardless of its type, always intends to escalate ‹rst at node 2. For a member of this family to exist, Defender must believe that any demand for a change in the Status Quo would be a mistake made by a genuinely soft Challenger.20 There may be situations when this kind of belief is in fact plausible. But our special case analysis, which is based on Defender’s a priori belief that Challenger is very likely hard, is patently not one of them. Indeed, it is extremely dif‹cult to imagine any situation in which Defender, after observing an act of unprovoked hostility, concludes that a Challenger, originally thought to be likely hard, is most likely soft. For this reason, all Escalatory De19. Table 6.2 is excerpted from table A8.1 in Zagare and Kilgour 2000: app. 8, which should be consulted for details of the de‹nitions and interpretations. De‹nitions of the strategic and belief variables appearing in table 6.2 are summarized here for convenience. The probability that Challenger initiates at node 1 of the Asymmetric Escalation Game is denoted x. In fact, this probability can depend on Challenger’s type—if Challenger is hard, the initiation probability is xH; if soft, it is xS. Likewise, wH and wS are the probabilities that hard and soft Challengers, respectively, escalate at node 3a. At node 3b, Challenger always chooses E if hard and D if soft. Similarly, Defender chooses D at node 2 with probability y, E with probability z, and C with probability 1 – y – z. Again, these probabilities can depend on Defender’s type, so they are denoted yHH, zHS, and so forth. It can be proven that ySH = ySS = 0 at any perfect Bayesian equilibrium. At node 4, Defender chooses E if strategically hard (type HH or SH) and D otherwise. Finally, players revise their initial probabilities about their opponent’s type as they observe the opponent’s actions. Of these revised probabilities, the only two that are important to the equilibria are shown in table 6.2. Defender’s revised probability that Challenger is hard, given that Challenger initiates, is denoted r. Challenger’s revised probability that Defender is of type HH, given that Defender chooses D (response in kind) at node 2, is denoted qHH. For the most part, the speci‹c value of the probabilities given in the table (e.g., d1 or cq) does not enter into the analysis that follows. The exception is pStr|Tac, which is the conditional probability that Defender is hard at the second level given that it is hard at the ‹rst level. 20. More technically, Defender’s updated belief (probability r) that Challenger is hard given that Challenger chooses D at node 1 must be relatively small.
126
the games of july
terrence Equilibria are dismissed as plausible possibilities in the Asymmetric Escalation Game and, henceforth, disregarded.
6.4.2. no-response equilibrium The second major equilibrium form of the Asymmetric Escalation Game with incomplete information is the No-Response Equilibrium. Under the No-Response Equilibrium, Challenger always initiates and Defender always capitulates, as the Russians did in 1909 at the conclusion of the Bosnian crisis. Defender gives in because Challenger is very likely hard and therefore is prone to escalate ‹rst at node 3a or to counterescalate at node 3b. To support its choice at node 3a, however, Challenger must believe that a Defender who unexpectedly responds in kind at node 2 is more likely to be of type HS than of type HH. This is a plausible belief since, ceteris paribus, type HH Defenders are more likely to escalate than are type HS Defenders.
6.4.3. the spiral family The Spiral Family contains four perfect Bayesian equilibria. Two are deterrence equilibria; there is also one member of the Constrained Limited-Response Equilibrium (or CLRE) group, and one representative of the Escalatory Limited-Response Equilibrium (or ELRE) group. As noted above, the members of this set are mutually exclusive. At most, one member of the Spiral Family may exist at any one time in the Asymmetric Escalation Game with incomplete information.21 The two closely related deterrence equilibria in the Spiral Family (Det2 and Det3) are called the Limited-Response Deterrence Equilibria. Under either equilibrium form, Challenger never initiates, and the outcome of the game is always Status Quo. As their name implies, equilibria of this category do not require Defender to escalate ‹rst. In fact, the form of deterrence that emerges under either Det2 or Det3 rests entirely on the more limited threat of responding in kind at node 2. This characteristic alone sets the Limited-Response Deterrence Equilibria apart from the Escalatory Deterrence Equilibrium (Det1). In addition, since the Limited-Response Deterrence Equilibria are based on plausible beliefs, they are not so easy to dismiss. Indeed, one would expect that, over time, one or the other of these equilibria might come into play in the Asymmetric Escalation Game with incomplete information. However, Det2 is much more likely. The existence of a Limited-Response Deterrence Equilibrium depends 21. By contrast, the Escalatory Deterrence Equilibrium Det1 (which has been dismissed as a rational strategic possibility) and the No-Response Equilibrium always exist.
La Guerre Européenne
127
solely on Challenger’s beliefs about Defender’s type. (Defender’s a priori beliefs are completely immaterial to the existence of a Limited-Response Deterrence Equilibrium.) Speci‹cally, for Det2 or Det3 to exist, both Defender’s ‹rstand second-level threats must be highly credible: Challenger must believe it quite likely that Defender is tactically hard, and given that Defender is tactically hard, Challenger must place a fairly high probability on Defender also being strategically hard.22 Given these beliefs, Challenger generally intends not to escalate at node 3a because it believes that Defender will likely counterescalate at node 4. And because Challenger believes that Defender will almost certainly respond in kind at node 2—most likely forcing Challenger to back down at node 3a—Challenger decides not to initiate at node 1. The third member of the Spiral Family, CLRE1, is the only form of Constrained Limited-Response Equilibrium that exists when Challenger is likely hard. Under this equilibrium, the Status Quo never obtains; Challenger always initiates. For its part, Defender responds in kind if it is tactically hard (i.e., of type HH or HS). Otherwise, Defender capitulates. Since this member of the Spiral Group of perfect Bayesian equilibria exists only when Defender is likely soft at the ‹rst level, (i.e., when pTac is low and pHS is not too large), the most likely outcome of play under CLRE1 is Defender Concedes. Thus, when Challenger chooses D at node 1, it does so with the expectation that its demands will almost certainly be met. Thus, should Defender respond in kind, Challenger will be surprised. In this unlikely event, Challenger will be forced to update its beliefs about Defender’s type. Clearly, Challenger will conclude that Defender is tactically hard, since only tactically hard Defenders can rationally choose D at node 2. Moreover, under any Constrained Limited-Response Equilibrium, if Defender is hard at the ‹rst level, then it is likely hard at the second level as well—that is, more likely to be of type HH than of type HS. Fearing this possibility, Challenger is understandably deterred from escalating at node 3a; instead, it always chooses D at node 3a, settling for a Limited Con›ict. The Constrained Limited-Response Equilibrium group is strategically signi‹cant, if only because a Limited Con›ict is most likely when a member of this group is in play. In the next section, particular attention is paid to the conditions under which CLRE1 exists. For now, I simply observe that the existence of a Constrained Limited-Response Equilibrium may help to explain why states, at times, abruptly shift gears and adjust their behavior in midcrisis, an explanation that is fully consistent with G. Snyder and Diesing’s (1977: 397) 22. More precisely, pStr|Tac must be large.
128
the games of july
empirical observation that “strategy revision is initiated when a massive input of new information breaks through the barrier of the image and makes a decision maker realize that his diagnosis and expectations were somehow radically wrong and must be corrected.” Limited Con›ict is also possible under ELRE3, the only form of Escalatory Limited-Response Equilibrium that exists when Challenger is likely hard, but that possibility is remote, at best. In fact, the most likely outcome of a game played under this equilibrium form is again Defender Concedes. Whenever ELRE3 is in play, Challenger, whatever its type, always chooses D at node 1, thereby upsetting the Status Quo. What happens next depends on Defender’s type. Under ELRE3, Defender is likely to be tactically soft (i.e., of type SS or SH). Such Defenders always concede at node 2, which is why the Defender Concedes outcome is the most likely outcome under any Escalatory Limited-Response Equilibrium. In the less likely event that Defender is hard at the ‹rst level, it would respond in kind, with certainty if it is also hard at the second level (i.e., of type HH) and probabilistically if it is soft at the second level (i.e., of type HS). Given the probabilities, however, a response in kind will again surprise Challenger. Up to this point of surprise, behavior and expectations are similar under ELRE3 and CLRE1. What separates these two equilibria are Challenger’s expectations should Defender unexpectedly choose D at node 2. Recall that under CLRE1, Defender responds in kind only if hard at the ‹rst level, and if Defender is hard at the ‹rst level, then it is likely hard at the second level as well. This is why Challengers never escalate ‹rst under a Constrained Limited-Response Equilibrium. Under ELRE3, though, a Defender that responds in kind is much more likely to be of type HS than of type HH. For this reason, a hard Challenger, the focus of our attention, often escalates at node 3a. If it does and Defender is strategically hard, the heretofore limited con›ict then spirals to the highest level. In other words, ELRE3 describes one possible path to All-Out Con›ict in the Asymmetric Escalation Game. When Challenger is seen as likely hard, it is the only way this outcome can occur.
6.5. Discussion To this point I have described the perfect Bayesian equilibria of the Asymmetric Escalation Game with incomplete information for the special case when Defender thinks that Challenger is likely to be hard. Next I use the equilibrium
La Guerre Européenne
129
structure of the game to address a number of questions about the outbreak of war in Europe.
6.5.1. what were they thinking? One question that arises immediately concerns the expectations of both Germany and Austria during the period immediately following the Hoyos mission on July 6 and the delivery of the Austrian ultimatum on July 23. What, in other words, were the leaders in Vienna and Berlin thinking during the crisis’s socalled silent period? The equilibrium structure of the Asymmetric Escalation Game with incomplete information gives a very strong suggestion about the likely content of their thoughts. We know from the last chapter that, in the wake of the archduke’s assassination, the Austrian request for support and Germany’s strong commitment were implicit in the structure of the Protégé-Defender Subgame. We also know that the equilibrium that came into play in that game implied an Austrian hard-line policy toward Serbia. Thus, it seems safe to conclude that after the blank check was issued, there was little or no chance that Vienna would not cash it. All of which implies that, after Sarajevo, neither of the two plausible deterrence equilibria was likely to come into play. Once these deterrence equilibria are eliminated as empirically unlikely, only three theoretical options remain, and one of them is the No-Response Equilibrium. A strong possibility exists that leaders at both the Ballhausplatz and the Wilhelmstrasse anticipated that play would occur under this equilibrium form. Were that the case, they would have fully expected a one-sided victory. Recall that Defender Concedes is the only possible outcome under the NoResponse Equilibrium. But much the same could be said for the two remaining theoretical possibilities, CLRE1 and ELRE3. Under either of these two perfect Bayesian equilibria, Defender Concedes, though not certain, is the most likely outcome. Thus, regardless of which of the three nondeterrence equilibria AustriaHungary and Germany believed to be in play, one would expect, theoretically, that they had the clear expectation that Russia and France were unlikely to offer any meaningful resistance. Of course, we know empirically that such was indeed their expectation, at least initially, so our answer should come as no surprise. Still, it is encouraging to know that the equilibrium structure of the Asymmetric Escalation Game with incomplete information is fully consistent with the facts as they are known. Were this not the case, the explanatory and predictive power of the model would be more than suspect.
130
the games of july
6.5.2. was war avoidable? A second important question is whether the crisis in Europe was inevitable, whether Austria-Hungary and Germany could have been deterred from instigating a crisis in Europe. One answer is that, after the Hoyos mission, the die had been cast, that the prevailing status quo was no longer sustainable. But to accept this answer, one must also hold to the view that in the period before Austria-Hungary ‹nally issued its ultimatum, no new information about Russia’s, France’s, and perhaps Great Britain’s attitude could have stayed the Dual Alliance from its appointed rounds. This is a dif‹cult position to sustain. Assuming, then, that Austro-Hungarian and German perceptions were subject to revision, the answer is clear. The existence of two distinct LimitedResponse Deterrence Equilibria in the Asymmetric Escalation Game attests to the theoretical possibility that the crisis could have been averted. Of course, what is theoretically possible is not necessarily likely. Such is the case in the Asymmetric Escalation Game when Challenger is likely hard. To understand why, consider now ‹gure 6.2, which depicts in three-dimensional space the existence conditions associated with the Spiral Family of perfect Bayesian equilibria. Recall that, along with the No-Response Equilibrium, exactly one member of this family will exist at any one time. Defender’s credibilities determine which Spiral Family equilibrium exists. In ‹gure 6.2, every possible combination of Defender’s credibilities is represented as a point in the tetrahedron shown in the center of this ‹gure. The right horizontal axis represents the probability that Defender is of type HH, the lower-left (horizontal) axis the probability that Defender is of type SH, and the vertical axis the probability that Defender is of type HS. Thus, any point in the three-dimensional triangle, or simplex, has a combination of nonnegative coordinates, (pHH, pHS, pSH), with a sum less than or equal to 1. The fourth credibility, pSS, equals the difference between this sum and 1; this amount is also the (perpendicular) distance between the point (pHH, pHS, pSH) and the front face of the tetrahedron. For example, the point (0,0,0) represents the combination pHH = pSH = pHS = 0, pSS = 1. Speaking more informally, ‹gure 6.2 can be visualized as a corner of a room with two walls and a ›oor, all at right angles; the fourth face of the simplex is the downward-sloping plane. The side wall is light gray, the back wall is medium gray, and the ›oor is dark gray. Of course, to enable us to peer into this corner, the front face must remain transparent. Observe that the two Limited-Response Deterrence Equilibria occupy a
La Guerre Européenne
131
PHS 1
ELRE3
Det3
c*
Det2
cq c2 PHH 1
PSH
1
CLRE1
Fig. 6.2. Existence Regions for Equilibria of the Spiral Family
relatively small area of the simplex. Thus, ceteris paribus, it is not all that likely that either would come into play in the Asymmetric Escalation Game.23 Of course, not all things were equal in 1914. Given that deterrence was theoretically possible in July, what would have had to occur for this possibility to become a reality? The conditions associated with the existence of Det2 and Det3 provide a succinct answer. Notice from ‹gure 6.2 that the two closely related Limited-Response Deterrence Equilibria occupy a small region in the right-hand side of the tetrahedron, where pHH is large, pHS is not too large, and pSH and pSS are small. In this region, Defender is likely tactically hard; this explains its propensity under either Det2 or Det3 to respond in kind at node 2, whatever its actual type. But this tendency alone is not suf‹cient to deter Challenger. Defender’s willingness to respond in kind also rests on its ability to dissuade Challenger from escalating at node 3a. To obtain this result, Defender’s second-level threat must be highly 23. This statement can be veri‹ed simply by comparing the relative size of the cutouts associated with the two Limited-Response Deterrence Equilibria with those of CLRE1 and ELRE3.
132
the games of july
credible as well; in other words, for deterrence to succeed under either Det2 or Det3, Defender must likely be both strategically and tactically hard—that is, pHH must be large (both of Defender’s threats must be fairly credible). It is clear, however, that this condition, which is necessary for deterrence success in the Asymmetric Escalation Game, was not satis‹ed during the July Crisis. There is ample documentary evidence that both German and AustroHungarian leaders formulated their policies with the expectation that Russia, even with the backing of France, would not offer anything but token resistance if and when Austria moved against Serbia. The kaiser, for example, initially believed that the risk of a war was minimal since neither Russia nor France was prepared for one (Geiss 1967: 71, 77; Massie 1991: 862). Bethmann Hollweg (1920: 126) shared the kaiser’s opinion. Vienna’s views were similar. According to the Italian ambassador in St. Petersburg, by mid-July, Austria “was capable of taking an irrevocable step with regard to Serbia based on the belief that, although Russia would make a verbal protest, she would not adopt forcible measures for the protection of Serbia against any Austrian attempts” (Albertini 1952: 2:184). Astonishingly, not even the full mobilization order issued by the czar on July 30 altered Vienna’s expectations. In a telegram that Franz Conrad von Hötzendorf, the chief of Austria’s General Staff, sent to his German counterpart, Helmuth von Moltke, on the night of July 31, Conrad matter-of-factly observed that the Austrian leadership was “not sure yet whether Russia is merely threatening, therefore we could not allow ourselves to be diverted from the action against Serbia” (quoted in F. Fischer 1975: 507). To summarize: deterrence was a distinct but not highly likely possibility in July 1914. An overly aggressive act against Serbia could have been averted, at least in theory. But—and this is a big but—given that Russian and French credibility was negligible, the theoretical possibility could not be realized. Thus, while a European war was certainly not inevitable, the status quo was not very likely to survive, even in the short run. Parenthetically, it should be noted that this conclusion rests directly on the assumption, discussed previously, that each member of the Dual Alliance preferred the Status Quo to Limited and All-Out Con›ict—that is, both threats of Russia and France were capable. Of the two, Austria-Hungary’s and Germany’s preference for the Status Quo over Limited Con›ict is the more problematic. Would Austria-Hungary and Germany still have provoked a crisis if they believed (1) that they could avoid a continental war but (2) that the issues
La Guerre Européenne
133
that separated Vienna from Belgrade would be subject to multilateral mediation? If one concludes, as does F. Fischer (1967) and, by extension, Levy (1990/91), that both Austria-Hungary and Germany ranked the status quo relatively low, the answer would have to be yes. Most revisionist historians, including those who hold that World War I was in some sense inadvertent, would strongly disagree.
6.5.3. why did it happen? Finally, why did the policies of Germany and Austria-Hungary, coordinated at the onset of the crisis, diverge so dramatically after the Russian partial mobilization on July 29? To address this question, I refer once again to the evolving beliefs of the leadership groups in Vienna and Berlin and the equilibrium structure of the Asymmetric Escalation Game with incomplete information. The common policy, reached early in July after the kaiser, with Bethmann Hollweg’s concurrence, issued the blank check, is easy to explain. As noted above, both the Wilhelmstrasse and the Ballhausplatz initially believed that Russia was unlikely to respond if Austria moved aggressively against Serbia. To be sure, German leaders preferred that Austria-Hungary act quickly. (In the kaiser’s words, it was “now or never.”) But they did so precisely because they also believed that any delay would decrease the probability that a fait accompli could be pulled off. In consequence, as the days passed in July, Berlin continued to press Vienna to act—not that it mattered. While Austria was determined to strike, it was also determined to strike at a time and at a place of its own choosing. Timing aside, however, both members of the Dual Alliance preferred action to inaction and were clearly intent on taking it. Until mid-July, Germany and Austria-Hungary had policies and expectations that were, for all intents and purposes, identical. All of which helps to explain why, in 1914, these two closely aligned nations are generally considered a single unit in the theoretical and empirical literature of international relations. The commonality of purpose and expectations, however, would not hold forever. As rumors of Austria’s intentions began to circulate in European capitals, German of‹cials started to fear that Austria might not be able to have its way with Serbia before Russia could react (Berghahn 1993: 210). The equilibrium structure of the Asymmetric Escalation Game with incomplete information helps to explain what happened next. Previously I suggested that of the ‹ve possible perfect Bayesian equilibria in the Asymmetric Escalation Game, the two Limited-Response Deterrence
134
the games of july
Equilibria were inconsistent with German and Austrian expectations and, therefore, were not likely to come into play. The same, however, cannot be said of the remaining three rational strategic possibilities. The three still viable perfect Bayesian equilibria share a number of important characteristics. Whether play takes place under a No-Response Equilibrium, under CLRE1, or under ELRE3, Challenger always begins play by initiating at node 1. In other words, under any of the three, Challenger’s initial action choice is always the same. There is also no chance that Defender will respond by escalating immediately (i.e., zHH = zHS = zSH = zSS = 0) at node 2. Finally, a one-sided victory (i.e., Defender Concedes) is the most likely outcome under each equilibrium form. Under most real-world circumstances, therefore, it may not be possible to determine empirically which of the three had actually come to de‹ne play. However, one critical difference may sometimes provide a clue, even before the game is played out. Recall that under the No-Response Equilibrium, Defender never intends to respond, either in kind or by escalating, regardless of its type. By contrast, under either CLRE1 or ELRE3, tactically hard Defenders always respond in kind when they are strategically hard and sometimes or always respond in kind when they are strategically soft.24 Thus, up to the point at which German leaders came to fear that Russia might act to protect Serbia, the No-Response Equilibrium was the only perfect Bayesian equilibrium consistent with the expectations of decision makers in Berlin (and Vienna). And had the events of July 1914 unfolded as both Germany and Austria-Hungary initially hoped, World War I would have never occurred; rather, the Third Balkan War would have been a localized con›ict between Austria-Hungary and Serbia. Of course, everything did not go according to plan. As Austria ‹ddled about, German beliefs about Russian intentions changed. These expectations are clearly inconsistent with the existence of the No-Response Equilibrium, which can now con‹dently be eliminated as a potential descriptor of the events of late July. This leaves (for now) just two theoretical possibilities, CLRE1 and ELRE3. As shall be seen, until July 30, each provides a plausible explanation of the unfolding crisis. In fact, until push came to shove, German leaders acted as if CLRE1 was in play. Unfortunately, the beliefs and action choices of Austria’s leaders were consistent with the existence of ELRE3. If such had not been the case, the crisis would have been resolved differently. To explain why, I return now to our narrative. Once German leaders began 24. Tactically Soft Defenders never respond in kind.
La Guerre Européenne
135
to become concerned about the possible involvement of other powers, they intensi‹ed their pressure on Austria to act. At the same time, they also made a concerted effort to deter Russian interference. For example, on July 19, Jagow placed a notice in the North German Gazette, a quasi-of‹cial publication, that expressed his government’s position that “the settlement of differences which may arise between Austria-Hungary and Serbia should remain localized” (Geiss 1967: 142). The notice was a thinly veiled threat directed at the Entente: Germany would back Austria in a war with Russia and France. The notice was followed up on July 21 by a cable from Bethmann Hollweg that instructed Germany’s ambassadors in Russia, France, and Great Britain to convey the same message of‹cially. The cable reiterated the chancellor’s desire for “localization of the con›ict” (Kautsky 1924: no. 100), a euphemism for the Defender Concedes outcome, that is, a bilateral war between Austria-Hungary and Serbia. The dispatch was quite revealing, implying that Vienna was ‹nally about to take an irrevocable step. As the chancellor explained to his ambassadors, Austria “had no other course than to enforce its demands upon the Serbian Government by strong pressure, and if necessary, to take military measures.” But a localized con›ict also required that other governments remain uninvolved. To make this outcome more likely, Bethmann Hollweg instructed his ambassadors to warn, this time less subtly, that “the intervention of any other Power would, as the result of the various alliance obligations, bring about inestimable consequences” (Kautsky 1924: no. 100). Clearly, German policy now focused on precluding intervention in, and deterring escalation of, the dispute. It would remain so for the rest of the month. Two days after this broadside, at 6:00 P.M. on July 23, Austria’s ambassador to Serbia delivered to authorities in Belgrade the ultimatum that was designed to be rejected. Not only were the terms of the ultimatum harsh, but its deadline was exceedingly short. Serbia would have two days, just forty-eight hours, to respond. It is noteworthy that on July 25, the day the ultimatum was set to expire, Germany’s foreign secretary still believed that “neither London, nor Paris, nor St. Petersburg wants war” (Pogge von Strandmann 1988: 102). Jagow’s strong belief helps to explain the continuing pressure the German Foreign Of‹ce put on Vienna to take decisive action. Diplomatic and political foreplay came to an end on July 28, when Austria ‹nally declared war, much to the relief and the surprise of the German government. Whether it was the result of wishful thinking or of a sober but misguided calculation, the leaders of both countries still believed that localization of the con›ict remained the most likely outcome—that is, that a fait accompli
136
the games of july
could still be accomplished. Their illusion, however, would not last much longer. Later that day, Russia informed Germany and the other powers that “in consequence of Austria’s declaration of war on Serbia,” it would declare a partial mobilization “in the military districts of Odessa, Kiev, Moscow and Kazan tomorrow.” Signi‹cantly, the partial mobilization announcement also underscored “the absence of any intentions of a Russian attack on Germany” (Geiss 1967: 262). The Russian response, while a direct consequence of Austria-Hungary’s challenge to the status quo, was clearly measured. As Fromkin (2004: 190–91) explains, “‘Partial mobilization’ consisted of a number of measures, some feasible and others not, none of which would have signi‹cantly helped to defend Russia and most of which put Russia in a less advantageous position than before. It was an essentially political concept, muddled and unclear, intended to convey the message that Russia was resolved to act if necessary, but did not wish to alarm or provoke Germany or Austria as a full mobilization—a real mobilization—would have done.” To put this in a slightly different way, the intent of the Russian decision to respond in kind (i.e., to choose D at node 2) was deterrence (Fay 1966 [1928]: 2:439; Trachtenberg 1990/91: 130; Williamson and May 2007: 348). The Russian leadership wanted to send the message that, if necessary, it was prepared to escalate the con›ict, that it was not only tactically hard but strategically hard as well. The message was received loud and clear, at least in Berlin, where the Russian decision came as a shock (Williamson 1991: 208). As Massie (1991: 870) concludes, Germany now faced the growing likelihood of war with Russia. German policy had been to encourage a localized Balkan war, punish a regicide state, and restore the fortunes of a crumbling ally. Russian intervention had been discounted. The Tsar’s army was considered unready and the Kaiser and his advisors had expected Russia to give way, as she had ‹ve years earlier in the Bosnian Crisis. The prospect was glittering: localization accomplished; general war avoided; Serbia crushed; Austria reborn; Russia stripped of her status as a Great Power; the balance of power in the Balkans and Europe realigned. Russian mobilization against Austria demolished this dream.
Not surprisingly, the Russian partial mobilization brought about a stunning reversal of Germany’s approach to the crisis.25 Whereas Bethmann Hollweg 25. Fay (1966 [1928]: 2:402–16) traces this turnaround to the late afternoon of July 27.
La Guerre Européenne
137
had previously sought to localize the con›ict by (1) encouraging Austria-Hungary to move against Serbia and (2) discouraging other powers from intervening, he now began to urge restraint on his ally and to encourage a political (i.e., a negotiated) resolution of the crisis. To this end, the ‹rst in a series of what F. Fischer (1975: 495) calls the “world-on-‹re” telegrams was sent to Count Heinrich von Tschirschky, the German ambassador to Austria, on July 28 at 10:15 P.M. Berlin time. In this message, Bethmann Hollweg urged Austria to moderate its policy lest it “incur the odium of having been responsible for a world war.” He speci‹cally asked that Vienna accept the kaiser’s “Halt in Belgrade” proposal, made earlier in the day, by announcing (1) that it had no interest in acquiring Serbian territory and (2) that its occupation of Belgrade was temporary and contingent on Serbian compliance with the terms of the ultimatum. At the same time, the German chancellor wanted it made clear, probably because he was in the process of attempting to do so himself, that Tschirschky was “to avoid very carefully giving rise to the impression that we wish to hold Austria back” (Kautsky 1924: no. 323).26 Albertini (1952: 2:477–79), F. Fischer (1975: 72), Geiss (1967: 223), and B. Schmitt (1930: 2:171) interpret this telegram (no. 174) in the worst possible light, claiming that Bethmann Hollweg’s motivation was simply to place the blame for war, should it come, on Russia.27 They argue that the chancellor had delayed transmitting the kaiser’s proposal to Vienna and had intentionally undermined it by subtly altering its substance. But as Kagan (1995: 200) notes, “That judgment seems unduly harsh.” Although Bethmann Hollweg’s injunction to Tschirschky may have lacked a sense of urgency, most historians now hold that the chancellor sincerely sought to moderate Austria’s behavior on the night of July 28 (Langdon 1991: 180; Lebow 1981: 136).28 In any event, Bethmann Hollweg shortly thereafter followed up with another telegram (no. 192) that was less ambiguous. This telegram, transmitted at 26. Some scholars (e.g., Geiss 1967: 223) have pointed to this injunction as conclusive evidence of Bethmann Hollweg’s disingenuousness. But in a July 30 memorandum to Grey, the British ambassador in Berlin reported that the German chancellor was concerned that too much pressure on Vienna might make matters worse (Gooch and Temperley 1926: 11: no. 329). 27. By contrast, Williamson and May (2007: 361) conclude that the German chancellor “scuttled” Wilhelm’s proposal because he continued “to believe that he could keep the war local.” Similarly, Clark (2000: 209) concludes that “the view that [Bethmann Hollweg] had already begun to harness his diplomacy to a policy of preventive war cannot be supported from the documents. It is more probable that he was simply already committed to an alternative strategy that focused on working with Vienna to persuade Russia not to overreact to Austrian action.” 28. See also Nomikos and North 1976: 156–57. Mombauer (2001: 286, 185) tacitly accepts the sincerity of Bethmann Hollweg’s efforts, claiming that while Moltke and other German military leaders pushed for a preventive war at the end of July, both the kaiser and the chancellor got “cold feet.”
138
the games of july
2:55 A.M. Berlin time, “urgently and impressively” urged Vienna to enter into direct negotiations with Russia. Five minutes later, after learning that this suggestion had previously been rejected, he again telegrammed Tschirschky, in apparent desperation, reiterating that a “refusal to hold any exchange of opinion with St. Petersburg” would be a “serious error” and a “direct provocation” of Russia. Bethmann Hollweg concluded by directing his ambassador to “talk to Count [Leopold] Berchtold at once with all impressiveness and great seriousness” (Kautsky 1924: nos. 395, 396). As Nomikos and North (1976: 156–62) observe, “Taken in sum, [the telegrams] represent a by no means inconsiderable effort to slow down, if not alter, the course of events now unfolding.” Some observers have attributed this dramatic policy shift to a strong warning from the German ambassador in London, Prince Karl von Lichnowsky, that Great Britain was very unlikely to stand aside in any war that involved France. (See, e.g., Albertini 1952: 2:520–22; Massie 1991: 871.) But Trachtenberg (1990/91: 136) argues persuasively that “it was the news from Russia about partial mobilization that played the key role in bringing about the shift in Bethmann’s attitude.” As Trachtenberg notes, “The evidence strongly suggests that the decisive change took place before the Chancellor learned of Grey’s warning, but after he had found out about Russia’s partial mobilization.” In the terms of the model, Germany had on July 28 updated its prior estimate of Russia’s type. Until Russia announced its partial mobilization, both Berlin and Vienna were operating on the premise that Russia would almost certainly capitulate, as it had previously—that is, that it was soft at both the tactical and strategic levels. The Russian partial mobilization, however, is inconsistent with this assessment. In the Asymmetric Escalation Game with incomplete information, only tactically hard defenders rationally respond in kind. It should come as no surprise, then, that both Germany and Austria would revise their initial beliefs in light of the new information obtained. This pattern of surprise and reevaluation is consistent with the existence of either of the two remaining solution candidates, CLRE1 and ELRE3. As ‹gure 6.2 suggests, both CLRE1 and ELRE3 exist only when the credibility of Defender’s ‹rst-level threat is insuf‹cient to sustain either of the two Limited-Response Deterrence Equilibria. In both cases, this reduction in Defender’s credibility gives even a soft Challenger an incentive to initiate at node 1. After all, since Defender believes that Challenger is likely hard, Defender is completely deterred from escalating ‹rst (under any perfect Bayesian equilibrium of the Spiral Family, including CLRE1 and ELRE3). Cal-
La Guerre Européenne
139
culating that Defender is likely to concede at node 2, Challenger therefore takes decisive action. It is in this sense that a response in kind comes as a surprise to Challenger regardless of which of these two perfect Bayesian equilibria is actually in play. What distinguishes CLRE1 from ELRE3, however, is the inferences that are made when Defender unexpectedly responds. Notice from ‹gure 6.2 that the upper face of the CLRE1 region slopes upward away from the bottom edge of the left-side wall. This sloping “ceiling” means that under this equilibrium form, the probability that Defender is of type HS is always small relative to the probability that it is of type HH. By contrast, under ELRE3, Defender is less likely to be of type HH and much more likely to be of type HS than of type HH, than under CLRE1. These critical differences lead to different behavioral patterns whenever Defender unexpectedly responds in kind. Under CLRE1, Challenger believes it more likely than not that Defender will counterescalate at node 4. As a result, Challenger never escalates ‹rst under this equilibrium form, and Limited Con›icts, if they occur at all, are most likely to occur under the conditions associated with the existence of CLRE1. Play under ELRE3, however, is another story. Here, since Challenger believes that it is unlikely that Defender will counterescalate at node 4, it may rationally decide to escalate at node 3a.29 If Challenger’s belief is incorrect, the result will be tragic, the escalation spiral complete. In the Asymmetric Escalation Game with incomplete information, then, the conditions that support the existence of ELRE3 uniquely describe the path to All-Out Con›ict. What is striking about the reevaluation process in 1914 is that German and Austrian leaders drew diametrically opposite conclusions from the measured Russian response. German leaders looked into the abyss and did not like what they saw. The July 29 warning from London may have had an impact here. But the Russian response in kind likely was critical. After all, unless the Russians further pressed the issue, the question of what Britain would do was moot. In any event, after the Russian partial mobilization, Bethmann Hollweg came to believe that Russia would not back down if Austria proceeded with its invasion. He also realized that to protect its western ›ank, Russia would also have to implement a general mobilization, which would clearly threaten Germany. And if Russia mobilized against Germany, Germany would be compelled to mobilize as well, and any German mobilization implied a two-front war 29. Unless it is soft, which is unlikely.
140
the games of july
against both Russia and France. Of course, an attack against France might bring Great Britain into the con›ict and, as Grey had just warned, British participation would bring about “the greatest catastrophe that the world has ever seen” (Kautsky 1924: no. 368). Bethmann Hollweg considered the latter eventuality as nothing less than a “leap into the dark.” All of which is to say that, after the partial mobilization, German political leaders concluded that Russia was not only tactically hard, but likely strategically hard as well, and they drew the proper inferences from this new assessment of Russia’s type. Accordingly, consistent with the de‹ning characteristics of a Constrained Limited-Response Equilibrium, Bethmann Hollweg quickly reversed course and urged moderation on Germany’s only real ally. His purpose was clearly to avoid the consequences of the escalation spiral that is implied by play under a Escalatory Limited-Response Equilibrium such as ELRE3. It is unfortunate, indeed, that on July 29 the critical decisions were not being made in Berlin. At this time, the locus of decision making was in Vienna, as Bethmann Hollweg obviously realized—hence the desperate tone of his telegrams to Tschirschky. Signi‹cantly, Vienna’s reaction to the Russian partial mobilization differed starkly from Berlin’s reading. The Austrian leadership in general, but Berchtold in particular, did not believe that Russia would further escalate the con›ict. In Vienna’s view, although the partial mobilization revealed that Russia was tactically hard, it did not follow that Russia was strategically hard as well. In fact, Austro-Hungarian leaders drew exactly the opposite conclusion (Albertini 1952: 2:388). According to Jannen (1996: 263, 249), “The Austrians simply did not take the threat of Russian intervention seriously.” At the height of the crisis, “Berchtold continued to believe that he could keep Russia talking while Conrad crushed Serbia.” In consequence, policymakers in Vienna acted as if Russia did not exist. Possibly they were overcon‹dent about the deterrent effect of Berlin’s “blank check”; possibly they exaggerated Romanov adherence to the principle of monarchical solidarity and the need to avenge the Sarajevo murders. Certainly they failed to pay even elementary attention to the danger signals of [a] Russian military response. Until late in the whole process, the senior leadership blissfully directed its attention only southward. . . . Berchtold, Conrad von Hötzendorf and the others, now programmed for action against Serbia, disregarded any information that might require them to mod-
La Guerre Européenne
141
ify their plans—and ambitions. They would do what they wanted and, of course, preferred to do: ‹ght Serbia. (Williamson 1983: 27)30
Based on his belief that Russia would stand aside, Berchtold de›ected all of Bethmann Hollweg’s frantic last-minute injunctions. When beseeched by Tschirschky to “be satis‹ed by the occupation of Serbian territory” (Kautsky 1924: no. 465), Berchtold delayed by claiming that he would have to consult with Franz Joseph before replying. And, with respect to Grey’s proposal for four-power mediation, Berchtold accepted Count Tisza’s suggestion to say that he was “ready to approach it in principle but only on the condition that [Austrian] operations in Serbia be continued and the Russian mobilization stopped” (Geiss 1967: 321). As Albertini (1952: 2:677) observes, “this was tantamount to outright rejection.” Berchtold was clearly drawing a line in the sand. On July 30, he approached the emperor for permission to proceed toward general mobilization that, when carried out, would directly threaten Russia. This move, according to Albertini (1952: 2:659), was “another big step in the direction of war.” All the while, Berchtold insisted that all Austrian “demands must be accepted integrally and we cannot negotiate about them in any way” (Geiss 1967: 320). In essence, by failing to moderate his government’s policy, Berchtold escalated the con›ict (B. Schmitt 1930: 2:155–56). Of course, since Berchtold’s beliefs about Russia’s likely response were incorrect, the results of his hard-line policy were entirely predictable.31 Throughout the crisis Russia had been on the verge of a general mobilization. On July 29, for example, the same day that the partial mobilization was implemented, the czar had in fact agreed to a full mobilization, only to rescind his order after learning that the kaiser was attempting to mediate the dispute. But by the next day, Nicholas could no longer resist the pressure from his foreign minister. It is very likely, as Turner (1968: 85–86) contends, that (1) the news that Austria was shelling Belgrade and (2) Bethmann Hollweg’s warning on July 29 to stop mobilizing turned Sazonov around and prompted him to push for full mobilization. But in his July 31 telegram to the kaiser justifying the action (Kautsky 1924: no. 487), the czar, whose consent was critical, tied his decision 30. Albertini (1952: 2:686) expresses a similar view. 31. In one sense, Berchtold was correct. Discussions with St. Petersburg would continue until August 6, when Austria ‹nally declared war on Russia.
142
the games of july
to sign the general mobilization ukase directly to his mistaken belief that Austria had already mobilized against Russia (Albertini 1952: 2:576). Berchtold’s in›exible policy, lacking as it did any conciliatory gestures, had made it all too easy for the czar to believe the worst about Austrian behavior.32 As the saying goes, the rest is history. At 5:00 P.M. on July 30, an hour after the czar had given his consent, the orders for a general mobilization were issued. At noon the next day, Austria-Hungary, following suit, mobilized against Russia. Shortly thereafter, Germany issued ultimatums to Russia and France, demanding that all of their mobilization efforts be canceled. Of course, neither St. Petersburg nor Paris responded in the af‹rmative. Consequently, Germany declared war on Russia on August 1, 1914, and on France two days later. Continental Europe was now at war. The theoretical characteristics of ELRE3 help to explain why. War broke out in Europe in 1914 because both Austria-Hungary and Germany believed that Russia would stand aside when Austria moved aggressively against Serbia. Localization was not only their objective but also their ‹rm hope and expectation. Of course, both members of the Dual Alliance were mistaken. Russian policymakers had already decided that Russia could not abandon Serbia and still survive as a great power. Fearing war, however, they declined to escalate the crisis. Instead, Russian leaders settled on a limited response, a partial mobilization against Austria that was intended to serve as a warning: “Stop or we will shoot!” Decision makers in Berlin quite clearly got the message. Unfortunately, their counterparts in Vienna did not. And it was in Vienna that the crucial choice not to pull back was made. By refusing to compromise, Austrian leaders escalated the crisis. Russia responded by mobilizing the rest of its forces against Germany. Of course, it was well understood in St. Petersburg that this act of counterescalation “almost certainly meant war” (Fay 1966 [1928]: 2:479). Sadly, this was just about the only belief con‹rmed by events. To conclude, it should be noted that ELRE3 is the only perfect Bayesian equilibrium of the Asymmetric Escalation Game that is consistent with both the expectations and the action choices of the key players in July 1914. Several outcomes, including a one-sided victory (i.e., localization) and a limited 32. Sazonov’s (1928: 203) memoirs con‹rm the importance the czar placed on the untenable position he believed Russia to be in as a consequence of Austria’s (as yet undeclared) general mobilization. It is unclear exactly how the czar came to be misinformed. But there is no indication that Sazonov went out of his way to set the record straight. In his foreword to the diary of Baron Schilling (1925: 9), the head of the chancellery in the Russian Foreign Ministry, Sazonov cites the fact that “Austria’s mobilization was in full swing” as one of several factors that led to Russia’s general mobilization. Sazonov clearly was carefully parsing his words: Russia was the ‹rst major power to fully mobilize.
La Guerre Européenne
143
con›ict (i.e., a negotiated settlement), are possible under this equilibrium form. Unfortunately, escalation spirals are also real and distinct possibilities. Testimony to this distressing fact is the continental war that broke out in early August 1914, la guerre européenne.
6.6. Coda The war was no accident. In 1915, Bethmann Hollweg explained it by noting that a number of factors had forced Germany “to adopt a policy of utmost risk, a risk that increased with each repetition, in the Moroccan quarrel, in the Bosnian crisis, and then again in the Moroccan question” (Jarausch 1969: 48). Clearly the German chancellor realized that he had rolled the dice one time too many. As Remak (1971: 366) insightfully observes, sometimes “it happens.” The laws of probability guarantee it.
Although we cannot bind ourselves under all circumstances to go to war with France against Germany, we shall also certainly not bind ourselves to Germany not to assist France. sir edward grey, april 1912
chapter seven
Britain’s Strategic Dilemma
Under considerable pressure from his political and military advisers, Czar Nicholas II ‹nally consented to a full mobilization of Russian armed forces against both Austria-Hungary and Germany on Thursday, July 30, 1914. The czar’s decision clearly placed Germany’s policymakers in an untenable position, for the Russian army vastly outnumbered its German counterpart (Kennedy 1987: 203). To offset the numeric de‹ciency, Germany’s war plans called for a rapid mobilization, a quick strike against France through Belgium and, with Austria’s help, a defensive campaign against Russia in the east until the French were subdued. At that point, the German army was to do an about-face and turn east to take on the Russians.1 France, however, would have to be defeated quickly, before the Russian mobilization could be completed. Otherwise, the Russian “steamroller” would likely overwhelm German forces on the eastern front. Unless it reacted immediately, or unless all Russian mobilization efforts were suspended, Germany would lose its most important strategic advantage: speed. For German military planners, time was of the essence; delay, they believed, was tantamount to defeat. Decision makers in Berlin were well-aware that “Germany could not win a protracted war [of attrition] against Britain, France, and Russia” (Herwig 2003: 186). Understandably, then, it did not take long for German leaders to respond to the Russian mobilization decision. On the next day, Friday, July 31, the kaiser authorized the proclamation of a state of “imminent danger of war,” which 1. This, the standard interpretation of German war planning, is challenged in Zuber 1999; 2002a, b; 2004.
144
Britain’s Strategic Dilemma
145
was but a prelude to the general mobilization ordered the following afternoon. At the same time, the French government was given just eighteen hours to declare whether it intended to remain neutral in a Russo-German war. Finally, the German ambassador, Count Friedrich von Pourtalès, delivered an ultimatum to Russia demanding a halt to all mobilization activities, including those directed against Austria-Hungary. On Saturday, August 1, Pourtalès returned to the Russian Foreign Of‹ce and asked three times whether Russia would cease and desist. And three times the Russian foreign minister, Serge Sazonov, replied negatively. At that point, “with trembling hands,” Pourtalès “presented S. D. Sazonov a note containing a declaration of war” (Schilling 1925: 77). Shortly thereafter, German troops crossed the Luxembourg border. All of Europe was not yet at war, but it was well on its way. The short delay is best attributed to a last-ditch German effort to preclude Great Britain’s entry into the war on the side of Russia and France. The attempt was not necessarily futile; as late as August 1, decision makers in Berlin had good reasons to hope that Britain might still stand aside. Just the previous day, for example, Sir Edward Grey, Great Britain’s foreign secretary, all but acknowledged that conditions existed under which Britain would allow the four continental powers to settle their dispute among themselves. According to Grey, he had told the German ambassador, Prince Karl Max von Lichnowsky, “that if Germany could get any reasonable proposal put forward which made it clear that Germany and Austria were striving to preserve European peace, and that Russia and France would be unreasonable if they rejected it, I would support it at St. Petersburg and Paris, and go the length of saying that if Russia and France would not accept it His Majesty’s Government would have nothing more to do with the consequences” (Gooch and Temperley 1926: 11: no. 340). At the same time that German leaders were working to gain Britain’s neutrality, French of‹cials were desperately trying to get the British to do exactly the opposite. But Grey would not bite. On July 31, the French ambassador pressed Grey to make a public statement of support, but he politely but ‹rmly declined. Great Britain, he said, could not “undertake a de‹nite pledge to intervene in a war.” Of course, he also indicated that he would “certainly consider the situation again” should there be “a new development” (Gooch and Temperley 1926: 11: no. 352). A German declaration of war against France on August 3 apparently did not quite qualify as a “new development,” at least for the majority of the British
146
the games of july
cabinet.2 But the invasion of Belgium on August 4 certainly did. When Germany failed to respond to a British ultimatum later that day demanding that it respect Belgium’s neutrality by withdrawing its forces, Great Britain declared war on Germany. The localized con›ict that escalated ‹rst into a continental war now included all the major powers of Europe. Several other states, including Italy, Turkey, Japan, and the United States, eventually would be drawn in. After the war, the Russian foreign minister blamed the con›ict on his British counterpart: I cannot refrain from expressing the opinion that if in 1914 Sir Edward Grey had, as I insistently requested him, made a timely and equally unambiguous announcement of the solidarity of Great Britain with France and Russia, he might have saved humanity from that terrible cataclysm, the consequences of which endangered the very existence of European civilization. (Sazonov 1928: 40)
Sazonov was not the only one to point an accusatory ‹nger at the British foreign secretary. In an extended comment appended to a telegram Lichnowsky sent to the German Foreign Of‹ce on July 29, the kaiser avowed that if Grey “were to say one single, serious, sharp and warning word at Paris and St Petersburg, and were to warn them to remain neutral, that both would be quiet at once. But he takes care not to speak the word, and threatens us instead! Common cur! England alone bears the responsibility for peace and war, not we any longer!” (Kautsky 1924: no. 368). Was Sazonov correct? Did the kaiser have a point? Could Germany have been deterred in 1914? Could France and Russia have been restrained? Was Grey’s entente policy fatally ›awed, or did it express, as Crawford (2003: 82) has argued, “a coherent strategic logic”? If a ‹rm commitment to France and Russia was possible, why did Grey fail to issue one? In short, can blame for the war be placed squarely at the feet of Sir Edward Grey, as both his friends and foes have so baldly asserted? To address these and related questions, I return again to the Tripartite Crisis Game, this time using a somewhat different set of information and utility assumptions. These new assumptions are easy to defend, empirically, and easy to justify, theoretically. They are also fully consistent with the set of simplifying assumptions that have guided all previous analyses of the Tripartite Crisis 2. For the internal political dynamic, see K. M. Wilson 1975.
Britain’s Strategic Dilemma
147
Game. But they have the added bene‹t of bringing into bold relief the particular strategic conundrum faced by Grey not only in 1914 but throughout his tenure as foreign secretary. Indeed, the vexing problem that faced Grey after he assumed of‹ce in December 1905 also confronted his immediate predecessor, Lord Lansdowne. It is instructive that both Grey and Lansdowne reacted to it in almost the same way. Grey’s implementation of the entente policy, however, was even more artful than Lansdowne’s. Still, Kagan (1995: 149) is correct when he characterizes Grey’s policy as “tortuous.” How else can one describe a policy that “seemed to offer [the French] a commitment with one hand and take it back with the other?”3
7.1. Modeling the Deterrence versus Restraint Dilemma G. Snyder and Diesing (1977: 432) call the strategic problem that both Grey and Lansdowne faced the “deterrence versus restraint” dilemma.4 G. Snyder (1984, 1997: 37) was the ‹rst to examine this conundrum game-theoretically, offering two related but theoretically isolated models, the Alliance Game and the Adversary Game. In the Alliance Game, Defender either supports (C) or withholds support (D) from Protégé. In the Adversary Game, Defender either stands ‹rm (D) or conciliates (C) Challenger. In general, the choice of (C) in one game implies a choice of (D) in the other, and vice versa. The linkage between the choices in the two games makes it dif‹cult for Defender to choose optimally in both—that is, simultaneously to deter the challenger and restrain the ally. Snyder clearly views the deterrence versus restraint dilemma as a dif‹cult cross-game maximization problem. As Crawford (2003: 18) observes, Snyder’s analysis of the deterrence versus restraint dilemma is informal and his separation of the two games unduly arti‹cial. In consequence, his conceptual synthesis, while provocative and insightful, remains intuitive and needlessly imprecise. The Tripartite Crisis Game was designed to overcome these limitations. The Tripartite Crisis Game begins with Challenger’s node 1 choice. De3. The policy predicament resurfaced after the war, albeit in a slightly different context. In the late 1930s, an important objective of British foreign policy was to preclude the absorption of Czechoslovakia by Germany, even though (signi‹cantly) Czechoslovakia was not considered vital to British interests. British prime minister Neville Chamberlain’s approach was the same as Lansdowne’s and Grey’s. As Lord Halifax, who served as Britain’s foreign secretary in 1938, put it, the British were “perpetually telling [Czech president Eduard] Beneš of what we might not do in the event of trouble, and of tactfully reminding the Germans of what we might do” (Newman 1968: 184). 4. In later works, G. Snyder (1984; 1997) refers to it as the “composite security dilemma.” The remainder of this chapter draws on Zagare and Kilgour 2006.
148
the games of july
pending on how play develops, Protégé may choose second, and Defender may choose third. If Defender has an opportunity to make a choice (at decision node 3), its choice determines whether Challenger or Protégé makes the last move. The set of possible choices and the outcomes associated with them, described in chapter 4, are captured graphically by the extensive-form representation of the Tripartite Crisis Game in ‹gure 7.1. As discussed previously, a distinguishing feature of the game is its complex network of threats. All three players have one. Although Challenger might view it as an opportunity or as a promise of sorts, a choice by Protégé at node 4a to realign threatens Defender’s interests. Similarly, Defender’s node 3 choice to support or to withhold support from Protégé threatens Challenger and Protégé, respectively. Finally, at node 4b, Challenger’s choice to press its demands and precipitate Con›ict threatens both Defender and Protégé. In the Tripartite Crisis Game, therefore, the choices of each player can be sanctioned by at least one other player. In addition, the players’ threats are dynamically interdependent. None of the players can afford to ignore the choices of any of the others.
7.2. Preference Assumptions This analysis retains all preference assumptions ‹rst described in chapter 4. A number of these assumptions are implied by the names or the roles of the players. For example, a player whose highest-ranked outcome is the Status Quo cannot properly be considered a Challenger. Similarly, a player who does not place a relatively high value on both the status quo and the welfare of a third state fails to meet the minimal requirements of a Defender. And ‹nally, since a Protégé, by de‹nition, depends on the assistance of another for its survival, it must be the case that Protégé prefers those outcomes that occur when it is supported by a third party to those that come about when it is left to fend for itself. Within the con‹nes of the Tripartite Crisis Game, however, a number of critical preference relationships are less obvious. In these instances, both theoretical and pragmatic considerations have been determinative. An important theoretical consideration is the need to highlight and intensify the deterrence versus restraint dilemma that conditioned Britain’s policy choices during Grey’s tenure as foreign secretary. Pragmatically, a focus is placed on a special case that makes the analysis tractable. No claim is made that these preference
Britain’s Strategic Dilemma
Challenger Not Demand (1 – x)
Node 1
Demand (x)
Protégé
Status Quo
Concede (1 – y)
Node 2
Hold Firm (y)
Defender
Challenger Wins Not Support (1 – z)
Protégé
Node 4a
Realign
Protégé Realigns
149
Support (z)
Node 4b
Not Realign
Protégé Loses
Node 3
Challenger
Back Down
Challenger Concedes
Press On
Conflict
Fig. 7.1. Tripartite Crisis Game
assumptions are those of all challengers, all defenders, and all protégés involved in extended deterrence relationships. Rather, they serve as boundary conditions that delimit the empirical circumstances to which the model applies, of which the run-up to the Great War is clearly one. Table 7.1 summarizes these assumptions and assigns hypothetical cardinal utility values to each outcome. (Later these hypothetical values are used to illustrate the theoretical implications of the model and its assumptions.) All preference assumptions are consistent with the axiomatic base of perfect deterrence theory. This means that the players are assumed to prefer, ceteris paribus, winning to losing. To re›ect the costs of con›ict, the players are also assumed to prefer to win or, if necessary, to lose at the lowest level of con›ict.
150
the games of july
Finally, those preference assumptions that have been left open represent threats that the players may or may not prefer to execute. Challenger’s, Protégé’s, and Defender’s preferences—which are rank ordered from most to least preferred in the table’s ‹rst, third, and ‹fth columns, respectively—have been justi‹ed at length in chapter 4. Recall that each player’s type is established by those critical preference relationships that remain unspeci‹ed. Speci‹cally, a determined Challenger prefers Con›ict to Challenger Concedes; a hesitant Challenger has the opposite preference. Similarly, when Protégé is loyal, it prefers Protégé Loses to Protégé Realigns; a Protégé with a preference for realignment is called disloyal. Finally, a staunch Defender prefers Con›ict to Protégé Realigns, while a per‹dious Defender has the opposite preference. To recapitulate brie›y, the version of the Tripartite Crisis Game considered here is de‹ned by the game tree appearing in ‹gure 7.1 and the preference assumptions listed in table 7.1. This version of the model was designed to capture the essence of the deterrence versus restraint dilemma that confronted not only the British but also the Russians (Joll 1992: 125), the French (Keiger 1995: 127), and the Germans (Carr 1964: 111–12) in July 1914. Of course, Defender is not the only player with a strategic dilemma in the Tripartite Crisis Game. Protégé’s threat to realign will tend to elicit Defender’s support. But when a Challenger is motivated to disrupt a strategic partnership, as Bismarck feared Russia was in 1879, Protégé’s threat to desert Defender may also prompt a crisis. Indeed, Germany’s desire to split the Franco-British entente helps to explain, in part, not only the First (1905–6) and the Second (1911) Moroccan Crises but also the July Crisis.
TABLE 7.1. Initial Preference Assumptions for Tripartite Crisis Game Challenger
Protégé
Preferences
Utilities
Challenger Wins
cCW = 10
Protégé Realigns Protégé Loses
Preferences
Defender Utilities
Preferences
Utilities
Status Quo
eSQ = 10
Status Quo
dSQ = 10
cPR = 8
Challenger Concedes
eCC = 8
Challenger Concedes
dCC = 8
cPL = 6
Conflict
eC = 4
Challenger Wins
dCW = 4
Status Quo
cSQ = 4
Challenger Wins
eCW = 2
Protégé Loses
dPL = 2
Conflict (when Determined)
cC+ = 1
Protégé Realigns (when Disloyal)
ePR+ = 1
Conflict (when Staunch)
dC+ = 1
Challenger Concedes
cCC = 0
Protégé Loses
ePL = 0
Protégé Realigns
dPR = 0
Conflict (when Hesitant)
cC– = –1
Protégé Realigns (when Loyal)
ePR– = –1
Conflict dC– = –1 (when Perfidious)
Britain’s Strategic Dilemma
151
7.3. Information and Utility Assumptions As should be clear from its discussion in chapter 4, the decision-making environment of the Tripartite Crisis Game is theoretically rich but conceptually complex. To gain additional tractability, therefore, two additional sets of simplifying assumptions are made. The ‹rst set is composed of several assumptions about the game’s information structure. First, Defender is assumed to be staunch, and this information about Defender’s type is taken as common knowledge. It is also assumed that all remaining information of the Tripartite Crisis Game is common knowledge, with the exception that the players are uncertain about Challenger’s and Protégé’s types. In other words, the assumption is that the players are aware of the game in extensive form as given by the game tree of ‹gure 7.1 and the utility assumptions summarized in table 7.1. Since the Tripartite Crisis Game is intrinsically more interesting when Defender is known to be staunch than when Defender is known to be per‹dious, the former assumption is easy to defend. Against a per‹dious Defender, Protégé’s node 4a threat to realign is unduly diminished and Challenger’s node 4b threat to press on is unnecessarily enhanced. Thus, the assumption that Defender’s preference for Con›ict over Protégé Realigns is common knowledge equalizes the impact of Challenger’s and Protégé’s threats directed against Defender. In consequence, the deterrence versus restraint dilemma is intensi‹ed.5 The second set of simpli‹cations contains several plausible albeit arbitrary assumptions about the players’ utilities. These assumptions, which are unlikely to be of any theoretical moment, are made here solely to facilitate a description of the game’s strategic structure. Before proceeding, however, the notation for the utility values given for Challenger, Protégé, and Defender in the second, fourth, and sixth columns of table 7.1, respectively, must be explained. In general, the utility of outcome O to Challenger is denoted by cO, to Protégé by eO, and to Defender by dO. For example, Challenger’s utility for the Status Quo is cSQ. But to account for the different player types, two different values are assigned to the utility of one critical outcome for each player as follows: 5. This assumption differs from the simplifying assumption made in chapter 4 that Challenger is known to be determined. Since in chapter 4 the goal was to explain not only Bismarck’s motivation to draw closer to Austria, but also the success of his policy over time, the worst was assumed (from Defender’s point of view) about Challenger’s preferences. By contrast, this chapter seeks to explain the failure of British policy to deter a German attack on Belgium and France. To avoid an all-too-easy explanation, Defender is assumed to be staunch. This is the strongest assumption about Defender’s preferences (from Challenger’s point of view) that is consistent with the existence of the deterrence versus restraint dilemma.
152
the games of july
cC+ is a determined Challenger’s utility for Con›ict cC– is a hesitant Challenger’s utility for Con›ict ePR+ is a disloyal Protégé’s utility for Protégé Realigns ePR– is a loyal Protégé’s utility for Protégé Realigns dC+ is a staunch Defender’s utility for Con›ict dC – is a per‹dious Defender’s utility for Con›ict. The hypothetical utilities given in table 7.1 range from a high of 10 to a low of –1. Thus, in the numerical example, both Protégé (eSQ = 10) and Defender (dSQ = 10) place a high value on maintaining the Status Quo relative to the other possible outcomes of the game. Challenger, by contrast, is comparatively dissatis‹ed with the existing order (cSQ = 4). Similarly, both types of Challengers (cC+ = 1, cC– = –1) and both types of Defenders (dC+ = 1, dC– = –1) have relatively low utilities for Con›ict. It is entirely understandable, however, that Protégé’s evaluation of Con›ict (eC = 4) is comparatively higher than that of both Challenger and Defender. After all, Protégé stands to lose the most if it is forced to accommodate Challenger.
7.4. Analysis As in previous chapters, this section’s analysis of the strategic properties of the Tripartite Crisis Game under incomplete information is informal, and many ‹ne points are omitted.6 The goal is to describe in the simplest possible terms the most important features of all perfect Bayesian equilibria of the game, given the utility and information assumptions outlined earlier. Recall that in the Tripartite Crisis Game, Protégé’s and Challenger’s choices at nodes 4a and 4b, respectively, are strictly determined by each player’s type. Speci‹cally, at node 4a, a disloyal Protégé always chooses to realign, and a loyal Protégé always chooses not to; similarly, at node 4b, a determined Challenger always presses on, while a hesitant Challenger always backs down. Notice also that if Protégé faces a decision at node 2, it will have observed Challenger’s prior action choice at node 1. And should the game reach node 3, Defender will have observed previous action choices by Challenger and Protégé. Protégé can use its observations to update its belief about Challenger’s type before making 6. For a formal treatment with all relevant technical details, see the appendix to Zagare and Kilgour 2005, available at http://pluto.fss.buffalo.edu/classes/psc/fczagare/Articles/Appendix.pdf.
Britain’s Strategic Dilemma
153
its choice at node 2; similarly, Defender has an opportunity to update its beliefs about Challenger’s and Protégé’s types prior to its node 3 decision. Given these considerations, it follows that a perfect Bayesian equilibrium of the Tripartite Crisis Game will consist of a seven-tuple of probabilities: [xD, xH , r; yD, yL, q; z] where: xD = the probability that a determined Challenger makes a demand at node 1 xH = the probability that a hesitant Challenger makes a demand at node 1 r = the updated probability that Challenger is determined, given that Challenger makes a demand at node 1 yD = the probability that a disloyal Protégé will choose to hold ‹rm at node 2 yL = the probability that a loyal Protégé will choose to hold ‹rm at node 2 q = the updated probability that Protégé is disloyal, given that Protégé holds ‹rm at node 2 z = the probability that (a staunch) Defender will choose to support Protégé at node 3. Five of the probabilities—xD , xH , yD, yL, and z—are strategic variables that specify Challenger’s, Protégé’s, and Defender’s choices at nodes 1, 2, and 3, respectively, contingent on their type. (Since the assumption is that Defender is known to be staunch, its node 3 choice is represented by a single variable, z.) The remaining variables, r and q, are the a posteriori probabilities that are updated after certain action choices by Challenger (at node 1) and Protégé (at node 2) are observed. As it turns out, there are exactly four plausible nontransitional perfect Bayesian equilibria in the numerical example of the Tripartite Crisis Game examined presently:7 Deterrence, Separating, Weak Crisis, and Strong Crisis. Table 7.2 summarizes their technical properties; ‹gure 7.2 contains a graphical summary of their existence conditions. When only plausible equilibria are considered, each perfect Bayesian equilibrium of the Tripartite Crisis Game is unique.8 To both describe the equilibria and model the consequences of the players’ 7. Hereafter the quali‹cations are dropped. 8. For further details, see Zagare and Kilgour 2005.
154
the games of july
uncertainty about Challenger’s and Protégé’s types, assume that Challenger is determined with probability pCh (where 0 < pCh < 1), and, as in chapter 4, that Protégé is disloyal with probability pPro (where 0 < pPro < 1). Also assume the values of pCh and pPro are known to all the players, and that the realized value of pCh (pPro) is known to Challenger (Protégé)—that is, that each player is aware of its own type.
7.4.1. deterrence equilibrium There is only one plausible deterrence equilibrium in the Tripartite Crisis Game when all the players know that Defender is staunch. It is referred to here as the Deterrence Equilibrium. Under the or any other (i.e., implausible) deterrence equilibrium, Challenger never makes a demand at node 1—that is, xD = xH = 0. As table 7.2 shows, under the Deterrence Equilibrium, both loyal and disloyal Protégés always intend to hold firm at node 2 (yD = yL = 1), and Defender always intends to support Protégé at node 3 (z = 1). Thus, it is should not be very surprising that Challenger never issues a demand at node 1. Whenever the Deterrence Equilibrium is in play, extended deterrence success is assured. Because there is no overt behavioral marker, deterrence success is, in some sense, impossible to observe. This explains why, under the Deterrence Equilibrium, there is no opportunity for Protégé or Defender to update its beliefs about Challenger’s type, why Protégé’s and Defender’s updated beliefs (r) about Challenger’s type are the same as their initial beliefs, and why Defender’s a posteriori belief (q) is always equivalent to its a priori belief (pPro) about Protégé’s type. Defender’s dif‹cult strategic choice at node 3 is clearly avoided if and when TABLE 7.2. (Plausible) Perfect Bayesian Equilibria of the Tripartite Crisis Game Challenger
Deterrence Equilibrium Separating Equilibrium Weak Crisis Equilibrium Strong Crisis Equilibrium
Protégé
Defender
xD
xH
r
yD
yL
q
z
0
0
arbitrary
1
1
pPro
1
1
0
1
1
pPro
½
½
1
0 < xH < 1
pCh < r < 1
1
1 – pPro 0 < yL < 1
pPro < q < 1
0