Framing an Analytic Narrative: A Tale of Three Crises

1.1 The Bosnian Crisis of 1908–1909

Although the Bosnian crisis of 1908 – 1909 involved almost all the major European powers and several minor powers, the focus here is on the core of the conflict which involved Austria-Hungary and Russia.^[3] The “specific precipitant” of the crisis was Austria’s annexation of Bosnia-Herzegovina, on October 7, 1908 (Snyder and Diesing 1977: 539). Although the two provinces were nominally under Turkish suzerainty, Austria had administered the provinces since 1878 under the terms of the Treaty of Berlin.

Prior to the annexation the Austrian Minister of Foreign Affairs, Alois Lexa von Aehrenthal, and his Russian counterpart, Alexander Izvolsky, had reached a secret agreement in which Russia would recognize the annexation and Austria would support Russia’s goal of opening the Straits of the Bosporus and the Dardanelles to Russian warships. Izvolsky, however, believed that the agreement would be implemented at an international conference at which the Austrians and the Russians would support each other’s position. But he was unpleasantly surprised when Aehrenthal informed him that the annexation would take place on October 7, 1908.

The annexation impinged on the interests not only of Turkey, but also of Serbia which had long coveted the provinces in order to gain access to the Adriatic. Bilateral negotiations between Austria and Turkey resulted in an agreement in which the Turks “got nearly everything which they desired” (Schmitt 1970: 124), but the Serbs were not so easily bought off, especially since the Russian government rejected Izvolsky’s understanding with Aehrenthal and threw its support behind Serbia’s demands for territorial compensation.

Fed up with the intransigence of both the Russians and the Serbians, Aehrenthal eventually turned the tables and demanded that the two Slavic states formally accede to his attempted fait accompli. Putting it to the Serbs, he also insisted that they immediately demobilize and reduce the size of its army. In the meantime, the Austrian military prepared for an invasion. Facing this implicit Austrian threat and backed up by what many considered a German ultimatum, Russia (and an abandoned Serbia) capitulated (Massie 1991: 609). In consequence, the planned Austrian invasion of Serbia was called off, at least until 1914.

1.2 The July Crisis of 1914

As Kagan (1995: 167) notes, “the Bosnian crisis was a crucial step on the road to war” in 1914. Although the outcomes were different, the circumstances surrounding the Bosnian and the July crisis were remarkable similar. In both crises it was the Austrian intention to subjugate and humiliate Serbia that placed Russia in the unenviable position of either allowing Serbia to be dismembered or going to war with Austria. But whether it was a consequence of its rearmament program, an unambiguous French commitment to stand by its ally, a desire to avenge its defeat in 1909, or all the above, this time the Czarist government stood firm. Unlike in 1909 when he decided against any level of mobilization (Schmitt 1970: 191), this time the cautious Czar first agreed to a partial mobilization intended as a warning to Austria. When Austria ignored the threat, Russia completed its mobilization, prompting Germany to attack Russia and, a few days later, France. Even before Great Britain chimed in, Europe was at war.

1.3 The Cuban Missile Crisis of 1962

The specific precipitant of Cuban Missile crisis was the discovery by the United States that the Soviet Union was in the process of installing medium- and intermediate-range missiles in Cuba. The US response was measured: a naval blockade of Cuba, coupled with a stern warning by President Kennedy that it was American policy “to regard any nuclear missile launched from Cuba against any nation in the Western Hemisphere as an attack by the Soviet Union on the United States, requiring a full retaliatory response upon the Soviet Union” (Kennedy 1969: 168). The crisis ended with a political bargain: the Soviets agreed to withdraw the missiles in exchange for a public US promise not to attack Cuba and a private assurance that it would (eventually) dismantle its missiles in Turkey.

3.1 Perfect Bayesian Equilibria

Why did Russia capitulate in 1909? Why was the US response measured after the missiles were discovered? And why did Austria escalate the crisis in late July 1914? Answers to these and related questions can be derived from an examination of the equilibrium structure of the Asymmetric Escalation Game with incomplete information.

As Table 3 reveals, there are six Perfect Bayesian Equilibria in the special case of the Asymmetric Escalation Game summarized here. They fall into three major groups. The first is a family of several equilibria called the Escalatory Deterrence Equilibria. But since all members of this family are based on beliefs that are implausible, they will be ignored.^[8] This leaves only five possible solutions to the game: the No Response Equilibrium, which always exists, and the four members of the Spiral Family, of which precisely one will always co-exist with the No-Response Equilibrium.

Of the plausible perfect Bayesian equilibria, two are deterrence equilibria (Det₂, and Det₃). These two closely related equilibria are members of the Spiral Family. Although they are based on plausible beliefs, they can also be ignored since they clearly are not possible descriptors of rational play in any one of the three crises. This leaves the No Response Equilibrium, the Constrained Limited-Response Equilibrium CLRE₁, and the Escalatory Limited-Response Equilibrium ELRE₃ as rational strategic possibilities.

As Table 3 shows, under the No-Response equilibrium NRE, Challenger always initiates at node 1 (i.e. the strategic variable x always equals one for both types of Challenger) and Defender always concedes and never responds, either in-kind or by escalating at node 2 (i.e. the strategic variables y and z always equal zero for all four types of Defender), which is why the only outcome that is consistent with rational choice under the No-Response equilibrium is Defender Concedes. As will be seen, the same cannot be said about either the Constrained-Limited Response equilibrium CLRE₁ or the Escalatory Limited-Response equilibrium ELRE₃.

The Constrained Limited-Response Equilibrium CLRE₁ is the only form of its type that exists when Challenger is likely Hard. Under this equilibrium, the Status Quo never obtains; Challenger always initiates at node 1. For its part, Defender responds-in-kind but only 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 first-level, the most likely outcome of play under CLRE₁ is also Defender Concedes. Thus, when Challenger chooses D at node 1, it does so with the expectation that its demands will almost certainly be met. If and when Defender is tactically Hard, however, it will respond with certainty. Defender’s response will surprise Challenger who then decides that discretion is the better part of valor. Under CLRE₁ then, even a Hard Challenger is deterred from escalating first at node 3a; instead, it always chooses D and settles for a Limited Conflict (outcome DD).

As is the case with both the No-Response and the Constrained Limited-Response equilibrium, both types of Challengers always choose D at node 1 under the Escalatory Limited-Response Equilibrium ELRE₃. Defender’s type determines its choice at node 2. Under ELRE₃, Defender is likely to be tactically Soft (i.e. of type SS or SH). Tactically Soft Defenders always concede at node 2. In the less likely event that Defender is Hard at the first level, it would respond-in-kind, with certainty if it is also Hard at the second level (i.e. of type HH) and with a variable probability (v) if it is Soft at the second level (i.e. of type HS)^[9] Given the probabilities, however, a response-in-kind will once again come as a shock to Challenger. Up to this point of surprise and reevaluation, behavior and expectations are similar under CLRE₁ and ELRE₃. What separates these two equilibria are a Hard Challenger’s action choice at node 3a. Under ELRE₃ Hard Challengers, guessing that Defender is strategically Soft (i.e. of type HS), are prone to escalate first at node 3a. At this point, Defender normally backs off and the outcome is Challenger Escalates (Wins). But sometimes Challenger’s guess will be wrong. Whenever this happens, Defender counter escalates, and an All-Out Conflict takes place.

4.1 Bosnia Explained

The standard interpretation of the Bosnian crisis paints Izvolsky as having been played by Aehrenthal. But as Mercer (1996: 115) argues, convincingly, Izvolsky knew exactly what he was doing. Unable to prevent the annexation, he hoped to extract Austrian support for opening the Straits in return for his acquiescence to what otherwise would have been an uncontested Austrian fait accompli. In other words, the Russian Foreign Minister hoped to “get something for nothing.” But he had been dealt a very weak hand – and he knew it. In 1908 Russia was still recovering from its devastating defeat by Japan in 1905, and the Tsar’s hold on power had recently been shaken by political upheaval. It was generally understood in St. Petersburgh that his autocratic regime would not survive another war, but a losing one in particular. In the terms of the model, Russia was clearly of type S/S, that is, in the short run it was not prepared to forcefully oppose the annexation of the provinces either diplomatically or militarily,^[10] especially since Izvolsky “never doubted Austrian or German resolve” (Mercer 1996: 137). Again, in the terms of the model, Russian policy makers believed that Austria was Hard and certain to escalate at either node 3a or 3b unless they acceded to the annexation.

Aehrenthal also knew what he was doing. With only a modicum of diplomatic support from Great Britain, the Russians were clearly outgunned. Throughout the crisis, its main ally, France, took a hands-off approach. Aehrenthal clearly believed that Russia could not, and would not, forcefully resist what he was sure to be a diplomatic coup (Miller 2012: 129–135). As well, he possessed embarrassing documents that revealed Izvolsky’s “complicity in the annexation and his willingness to surrender Serbian and Montenegrin interests” (Kagan 1995: 161). It was taken for granted not only in Vienna, but also in London, Paris, and Berlin, that Russia’s diplomatic isolation, its military weakness, and Aehrenthal’s threat to release the incriminating documents made it all but impossible for Izvolsky not to fold. So, when push came to shove and Germany demanded a “Yes or No” response to its demand that the Russian government recognize the annexation, Izvolsky capitulated.^[11]

The defining characteristics of the No-Response equilibrium helps to explain why. Recall that under this equilibrium Challenger (i.e. Austria) always issues a demand at node 1 and Defender (i.e. Russia) always complies at node 2. Defender complies because Challenger is very likely, if not certainly, Hard and, therefore, more than likely to escalate first at node 3a or to counter-escalate at node 3b.^[12] Few diplomatic historians would dispute the fact that the beliefs and the action choices associated with the No-Response equilibrium align well with those of both Russian and Austrian decision-makers during the end game of the Bosnian crisis.

Given the stark choice Izvolsky faced and his strong (and accurate) belief about Austria’s intention to move against Serbia unless he acceded to the annexation, a simple game-theoretic model could also be used to explain his decision. For example, in the Unilateral Deterrence Game with complete information, from which the Asymmetric Escalation Game is derived, Defender Concedes is the unique subgame perfect equilibrium when it is played by a Challenger whose threat is credible (i.e. is Hard) and a Soft Defender (Zagare and Kilgour, 2000: 140). Nonetheless, to facilitate cross case comparisons, it will be useful to situate the Bosnian crisis in the wider context of the Asymmetric Escalation Game, especially since, except for the lack of French support in 1909, the players and the underlying circumstances of the Bosnian crisis and those that existed in 1914 were almost identical. To wit, in 1909 and in 1914 Germany issued Austria a blank check; and in both crisis the British were noncommittal, at least initially.

It is important to point out that the outcome Defender Concedes is also the most likely outcome regardless of which of the three plausible perfect Bayesian equilibria is taken to be in play, that is, the action choices of Austria and Russia are consistent with those of the No-Response equilibrium, the Constrained Limited-Response equilibrium, and the Escalatory Limited-Response equilibrium. Nonetheless, it seems safe to associate the resolution of the Bosnian crisis with the No-Response equilibrium. One reason is that it provides a more parsimonious interpretation of the crisis. Another is that it is the only equilibrium under which there is no chance that Defender (i.e. Russia) would resist. The No-Response equilibrium, then, is the only solution candidate that is fully consistent with both the historical record and consensus interpretations of it.^[13]

4.2 The July Crisis Explained^[14]

“The Bosnian crisis was a crucial step on the road to war” (Kagan 1995: 167). Indeed, it was virtually a dress rehearsal. In 1914, Austria’s intent to subjugate Serbia presented Russia with the same difficult choice it faced in 1909: to either accept or resist the dismemberment of a client state. This time around, however, a sustained rearmament program coupled with staunch French support made it possible for Russia to contemplate a war with both members of the Dual Alliance.

Austria delivered its uncompromising ultimatum to Serbia on July 23, 1914. In response, the Russian Council of Ministers was convened where 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). Depending on circumstances, then, Russia was 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). In the terms of the model, the agreed upon contingency plan established Russia type, that is, Russia was of type H/H, Hard at both the tactical and the strategic level.

Austria declared war on Serbia on the 28th of July. Early the next day Austrian gunboats began shelling Belgrade. To the great surprise of both Austrian and German decision-makers, Russia reacted according to its contingency plan: a partial mobilization against Austria was ordered.^[15] After learning of the Russian partial mobilization, Austria responded by counter-mobilizing against Russia. It then escalated the crisis by continuing to advance into Serbia. As had already been decided, Russia then completed its mobilization. It was well understood in St. Petersburg that this was a war choice. But why did the crisis escalate, why did what was seemingly intended to be a local war confined to Austria and Serbia suddenly morph into what would eventually become a world war?

To understand why, I refer once again to the equilibrium structure of the Asymmetric Escalation Game with a particular focus on the existence conditions associated with the Spiral Family of perfect Bayesian equilibria. Recall that a group of implausible deterrence equilibria and the No-Response equilibrium always exist along with exactly one member of the Spiral Family. The existence of two closely related plausible Deterrence equilibria (DET₂ and DET₃) in the Spiral Family suggests that there were conditions, albeit remote, under which the July crisis could have been avoided.^[16] More likely, though, one of the two remaining perfect Bayesian equilibria, the Constrained Limited-Response equilibrium (CLRE₁) and the Escalatory Limited Response equilibrium (ELRE₃), would come to define play. Figure 2 provides a graphic description of the existence conditions of the Spiral Family of perfect Bayesian equilibria.^[17]

Figure 2:

Existence regions for equilibria of the spiral family (source: Zagare and Kilgour, 2000).

Every possible combination of Defender’s type probabilities (or credibilities) is represented as a point in the tetrahedron shown in the center of Figure 2. These probabilities determine which member of the Spiral Family exists. 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 non-negative co-ordinates, (p_HH, p_HS, p_SH), with a sum less than or equal to 1. The fourth credibility, p_SS, equals the difference between this sum and 1; this amount is also the (perpendicular) distance between the point (p_HH, p_HS, p_SH) and the front face of the tetrahedron. For example, the point (0,0,0) represents the combination p_HH = p_SH = p_HS = 0, p_SS = 1.

As Figure 2 shows, both the Constrained Limited-Response equilibrium CLRE₁ and the Escalatory Limited-Response equilibrium ELRE₃ exist only when Defender is very likely Soft at the first or tactical level. Thus, if and when a Defender responds-in-kind at node 2, it will come as a surprise to Challenger regardless of which of these two rational strategic possibilities is believed to be in play. Such was the case in 1914.

For most of July, neither Kaiser Wilhelm II, the German emperor, nor Theobald von Bethmann Hollweg, the German Chancellor, believed that war was likely (Geiss, 1967: 71, 77; Massie 1991: 862; Bethmann Hollweg, 1920: 126). But German beliefs were largely immaterial. What really mattered was the thinking in Vienna. Unfortunately, until July 29, the German and the Austrian governments were completely in sync. 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, II: 184). Austrian leaders would not waver from their belief until it was too late to matter.

Since only tactically Hard Defenders choose D at node 2, the Russian partial mobilization of July 29 surprised both members of the Dual Alliance and forced them to reassess their initial beliefs. In Berlin, where the intent of the Russian message was received loud and clear, the reevaluation led to the obvious conclusion that the Russians were not only tactically hard, but strategically Hard as well. As Figure 2 shows, the German reassessment is consistent with the existence of the Constrained Limited-Response equilibrium CLRE₁, that is, with the belief that a tactically Hard Defender is more likely to be strategically Hard as well, more likely to be of type H/H than of type H/S. Understandably, German leaders changed course and urged restraint on its only reliable ally in a series of what Fisher (1975: 495) calls the “world-on-fire” telegrams. Vienna was encouraged to not only accept the Kaiser’s “Halt in Belgrade” proposal but also to moderate its demands and enter into direct negotiations with Russia. In one of the final telegrams, Bethmann Hollweg warned that a “refusal to hold any exchange of opinion with St. Petersburg,” would be a “serious error” and a “direct provocation” of Russia (Kautsky 1924: no. 396).

Unfortunately, the Austrians read things differently. As Vienna’s saw things, while the partial mobilization revealed that Russia was tactically hard, it did not follow that it was also strategically hard. In fact, the Austro-Hungarian leadership drew exactly the opposite conclusion (Albertini 1952, II: 388). As Jannen (1996: 263, 249) writes, “the Austrians simply did not take the threat of Russian intervention seriously.”

At the height of the crisis, the Austrian foreign minister, Count Leopold “Berchtold continued to believe that he could keep Russia talking while [the chief of Austria-Hungary’s General Staff General Franz Conrad von Hötzendorf] crushed Serbia.” In the terms of the model, he believed that Russia was strategically Soft.

The belief that a tactically Hard Defender is much more likely to be of type HS than of type HH is the defining characteristic of the Escalatory Limited-Response equilibrium ELRE₃ (see Figure 2). Indeed, no other equilibrium in the Asymmetric Escalation Game with incomplete information is consistent with this updated belief and the action choice it gives rise to. Given Challenger’s belief that Defender will most likely back off at node 4, Hard Challengers tend to escalate first at node 3a under ELRE₃. All of which helps to explain why Berchtold deflected all of Bethmann Hollweg’s last minute pleas and instead decided to plow on. By refusing to compromise, Austrian leaders escalated the crisis.

Sometimes, however, Challenger’s belief about Defender’s type will be wrong. Whenever this happens, Defender counter escalates at node 4, and an All-Out Conflict breaks out. Such was the case on July 30, 1914. World War I was now underway.

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 late July and early August 1914. Several outcomes, including a one-sided victory (i.e. localization) and a limited conflict (i.e. a negotiated settlement) are possible under this equilibrium form. Unfortunately, escalation spirals are also real and distinct possibilities. The defining characteristics of the Escalatory Limited-Response equilibrium ELRE₃ succinctly explains why and when they take place.

4.3 The Cuban Missile Crisis Explained^[18]

Although some have argued otherwise (e.g., Sorensen 1965), there is a wide consensus in the strategic literature that the bargain that ended the missile crisis was a political compromise (Gaddis 1997: 261). Since it failed to escalate, it can also be thought of as a limited conflict. In the Asymmetric Escalation Game with incomplete information, limited conflicts are possible under either the Constrained Limited-Response equilibrium CLRE₁ or the Escalatory Limited-Response equilibrium ELRE₃. In the latter case, however, the possibility is remote, at best. It should not be surprising to learn, then, that the beliefs that support the action choices that most often lead to a limited conflict are to be found in the defining characteristics of CLRE₁.

The key to this possibility is Challenger’s initial and updated beliefs about Defender’s type. Recall that when CLRE₁ is in play, Hard Defenders always initiate at node 1; and that Defender is most likely to be of type SS and, therefore, prone to capitulate at node 2. Thus, if and when a Defender responds-in-kind, it will shock Challenger. Since only tactically Hard Defenders choose D at node 2, Challenger will be forced to update its beliefs about Defender’s type. Since under CLRE₁ tactically Hard Defenders are also more likely to be strategically Hard, even Hard Challengers will choose not to escalate first at node 3a. Instead, they settle for a limited conflict that must, per force, be settled politically.

Given the above, the clear theoretical expectation is that the brokered agreement that ended the missile crisis would have been preceded by a series of events that led the Soviet Premier Nikita Khrushchev to reevaluate his initial beliefs about the likely consequences of his actions. Otherwise, the crisis’s resolution is simply inexplicable. It is not difficult to substantiate this expectation.

When the Soviet decision to install the missiles was made, Khrushchev believed that the United States would not respond, either because it would be too late to do so if and when they were discovered, or because Khrushchev thought that the Kennedy administration would be unwilling to respond forcefully.^[19]

The blockade, of course, shattered this belief. Unlike the Austrians in 1914, the Soviets got the message implicit in the blockade and the other signals, intended or not, sent by the United States. Clearly, a “strategy revision” (Snyder and Diesing 1977: 397) was in order.

The Soviet reevaluation process, which began even before a personal letter from the President and a copy of his televised address was delivered to the Kremlin on October 22, did not take very long. However, “no single piece of information seems to have moved Khrushchev to his new position” (Fursenko and Naftali 1997: 260).^[20] So while “there is little evidence to explain exactly why Khrushchev reversed his assessment of American intentions” (Allison and Zelikow 1999: 125), there is no doubt and very little dispute that, for one reason or another, he became “convinced that the Soviet Union could not keep ballistic missiles in Cuba without going to war” (Fursenko and Naftali 1997: 259). And it was war that Khrushchev (1990: 176) “didn’t want.”

Khrushchev’s strong belief that a war was likely should the Soviets “inflame the situation”^[21] and escalate the conflict by running the blockade and pushing forward with the installation of the missiles is consistent with the beliefs necessary to support a limited conflict under CLRE₁, but inconsistent with the beliefs associated with the existence of ELRE₃. Thus, CLRE₁ is the only perfect Bayesian equilibrium in the Asymmetric Escalation Game that can explain why the crisis was resolved as it was.

Consistent with his beliefs about the consequences of an escalatory choice, Khrushchev did a strategic about face and decided to “conduct a reasonable policy.”^[22] “The decision to end the crisis through diplomatic means was made on the night of Wednesday October 25” (Fursenko and Naftali 2006: 616, fn. 69) at a meeting of the Soviet Presidium.

Khrushchev began that meeting by explaining why he thought that the missiles should be withdrawn: “The Americans say that the missile installations in Cuba must be dismantled. Perhaps this will need to be done. This is not capitulation on our part. Because if we fire, they will also fire.” But he did not back down entirely. He wanted to bargain: “We have to give the opponent a sense of calm and, in return, receive assurances concerning Cuba.” Then he suggested his terms: “Kennedy says to us: take your missiles out of Cuba. We respond: ‘Give firm guarantees and pledges that the Americans will not attack Cuba.’ That is not a bad [trade.]”^[23]

Not surprisingly, his proposal was unanimously supported by the Presidium. But it was left up to Khrushchev to decide when and how to seal the deal. That moment came soon after the Presidium met. “Early on Friday, October 26, Khrushchev received a stream of information indicating the likelihood that the Americans were readying an attack for October 27” (Fursenko and Naftali 2006: 486). Time was obviously running out, or at least he believed. Hence his long rambling letter to Kennedy of October 26 outlining the bargain that, eventually, ended the crisis (May and Zelikow 1997: 485–91; Stern 2012: 139). Most of what occurred afterwards, including Khrushchev’s infamous second letter of October 27 in which he roiled Kennedy by publicly demanding the removal of the missiles in Turkey, was simply diplomatic haggling. It would take a few more days to work out the details.