SIGNALING, SECRECY, AND DIPLOMACY

Scholars have long suggested that uncertainty is a fundamental cause of war (e.g., Blainey 1988). Correspondingly, the literature on deterrence and crisis bargaining has searched for credible information-revealing mechanisms that help state leaders overcome uncertainty and hence avoid inefficient fighting (e.g., Powell 1990, 1999). The literature describes how uncertainty about an opponent's resolve can lead to the outbreak of war. In particular, the formal literature on crisis bargaining shows when and how war results from the conscious decisions of the bargainers, even though a mutually preferable peaceful settlement is available (e.g., Fearon 1995; Morrow 1989).

Recent audience-cost models propose the linkage between domestic and international politics as a prominent mechanism that can rationalize a state leader's decision to go to war. By formalizing what is known as tying-hands signals, these models generally posit that the credibility of a threat can be established when leaders go public with their challenge and engage their domestic audiences, so that they can generate political costs that they would have to pay ex post if they fail to carry through on their public commitment (Fearon 1994, 1997; see also Schelling 1966). Because leaders can tie their hands with audience costs, going public helps leaders overcome communication barriers and reveal meaningful information in crises. These models then assert that private, or less public, diplomacy is ineffective because it is relatively costless and nonbinding (e.g., Fearon 1994, 1995; Ramsay 2004; Schultz 1998).

Given this line of reasoning, it is quite logical to conclude that normal forms of diplomatic communication lack the credibility, as secrecy essentially “unlocks” leaders from inefficient outcomes and allows them to maintain their leeway to disavow their commitments. Since Fearon's influential rationalist account of war, diplomacy is routinely cited as “cheap talk” (Fearon 1995; Sartori 2005), which leads to a conventional conclusion that diplomacy by itself is inconsequential or secondary to military might.

Yet the 1970 Cienfuegos crisis challenges this conclusion. This anecdote suggests that President Nixon and Henry Kissinger rationally concluded that staying private was preferable to going public, fully recognizing the potentially adverse effects associated with engaging the American (and Russian) public. Historically, secrecy has been at the heart of modern diplomatic institutions which originate primarily as a conflict-resolution mechanism to overcome the security dilemma (Mattingly 1955; Nicolson 1954).

How can we account for this gap between theoretical expectations and the empirical record? Note that the conventional account of “ineffective” private threats is inferred solely from the informational rationale of public threats. However, as we shall see, the fact that a threat lacks informational efficacy does not necessarily mean that the threat is ineffective in influencing an opponent's crisis behavior and ultimately crisis outcomes. To properly address the puzzle of private threats, it is crucial to fully appreciate the role of audience costs generated by public threats in crisis diplomacy.

To establish the credibility of its threat in a crisis, a state leader can invoke the tying-hands mechanism by raising audience costs. Such credible information-revelation, however, is possible because this mechanism requires “a rational state … to run a real risk of (inefficient) war in order to signal that it will fight” (Fearon 1995, 397). That is, the gist of the tying-hands mechanism is a double-edged sword: generating audience costs by going public simultaneously facilitates separation of types of an informed state (informational effects) and escalates a risk of costly fighting (escalatory effects). Accordingly, the audience-cost story speaks to two closely related puzzles: (1) it helps explain why state leaders can optimally go to inefficient war (Fearon 1995); (2) it also helps explain why some states—typically democracies—can establish the credibility more effectively than others—typically nondemocracies (Fearon 1994).

Although the dual effects of audience costs in crisis bargaining are well recognized, the existing conclusion about private diplomacy focuses exclusively on its informational role (or the lack thereof).²

However, describing the duality of audience costs by itself still falls short of addressing the puzzle of private threats. The fact that some states have incentives against going public says nothing about how private threats in crisis diplomacy works. A satisfactory explanation for public versus private threats should not only account for disincentives to go public, but also how private communication can influence crisis behavior and outcomes. That is, to account for the mixed record of public and private diplomacy in crises, we must be able to explain both the 1962 Cuban Missile and the 1970 Cienfuegos crises. However, the existing audience-cost models are not suited for investigating the choice between a public versus a private threat, because these models do not capture the private aspect of crises that are carried out in the public eye, but instead simply assume that crises are public events (Fearon 1994; Ramsay 2004; Schultz 1998). A notable exception is Snyder and Diesing's (1977, 251–54) comparative analysis of public and private communication methods. However, they presume that a public threat is a better credibility-generating mechanism, and conclude that private communication best serves as a supplement to public threats. As we shall see, my model suggests that exactly the opposite can be the case.

The key to successful private crisis diplomacy is that, given a challenger's incentives to go private, a defender must also agree to capitulate in private, rather than dismiss a private threat as a bluff. Hence, to understand the puzzle of private threats, we need to know the incentives that the defender faces in deciding whether to capitulate in public or in private. Although Fearon (1994) and Schultz (2001b) also consider audience costs for both states in their models of international crises, these models are not suitable for the purpose here, because they postulate audience cots as signaling costs in crises. In Fearon's model, audience costs are automatically raised by the onset of a crisis, but not by a rational choice of any states. In Schultz's model, the act of signaling generates audience costs only for the sender, but not for the receiver, of signals. In effect, these models implicitly assume that a signaling action does not affect the subsequent bargaining environment because it has no consequences (or payoffs) on the opponent's side (see also Slantchev 2005). As a result, none of the existing audience cost models capture the fact that both the challenger and the defender have domestic political audiences who observe how crises are carried out and evaluate the performance of their leadership.

The remainder of this paper is an attempt to explore these issues and to look for the credibility condition of private signals in crisis bargaining. The model studied here is a natural extension of common audience cost models, building on a canonical crisis bargaining game that shares basic strategic elements with previous models. I take a standard model and relax two common assumptions, on which the conventional conclusion crucially depends. Because the model is simple, one could easily add further complications. But this simplicity is designed to highlight the questions above, to clarify some of the less intuitive consequences of the strategic problems, and to facilitate comparisons with previous studies.

THE MODEL

A crisis game involves two states—the challenger (C) and the defender (D)—in a dispute over some good whose value to both is normalized to 1.³

Sequence. In making a challenge, C can choose whether to go public (Pub) or stay private (Pri). Without loss of generality, C has no option of retaining the status quo (SQ) at the outset of the game because the focus here is on C's incentives to make a private threat and the associated credibility condition.⁴

Outcomes and Payoffs. When D makes a public concession, C obtains the value for the good, and D not only loses the good but also incurs audience costs from suffering “diplomatic humiliation” (Fearon 1994; O'Neill 1999). Thus, the payoffs are 1 for C and −a_D≤0 for D. If C backs down in public when resisted, C pays audience costs −a_C≤0 and D keeps the status quo payoff of 1. When C makes a private threat, neither side incurs audience costs by backing down or conceding. Hence, if D makes a private concession, C gets 1 and D gets 0. If C backs down in private, the game ends as if the crisis never happened, yielding the status quo payoffs 0 for C and 1 for D. In the event of a war, C's payoff is given by his expected value for war w_C=p−c_C, where p∈[0, 1] and c_C≥0 represent C's probability of victory and expected costs, respectively. Notice that the costly lottery assumption underlies the definition of war payoffs, and that the c_C term captures C's costs of war relative to the value of the disputed good. Similarly, D's war payoff is given by w_D=1−p−c_D. Figure 1 illustrates the sequence of moves and the payoffs associated with each outcome in the crisis game.

Crisis Diplomacy with Public versus Private Threats

Information and Beliefs. The crisis game involves two-sided uncertainty: each state has private information about its value for war, w_i. To generate this uncertainty, assume that nature randomly selects c_C and c_D from independent distributions on intervals

and

, respectively. This assumption simplifies the expression for w_i and implies that the

are distributed according to the cumulative distribution function F_i(x)=Pr(w_i≤x), whose support is the interval

for C and

for D. Each state observes the value of its own w_i, but neither observes the other's value for war. The probability distributions are common knowledge, so each state forms precrisis beliefs about w_j, j≠i. C's threat is said to be genuine or credible if he follows through with it. Define p_pub and p_pri as D's precrisis beliefs that C's public and private threats are genuine, respectively.

EQUILIBRIA

All equilibria to this game can be described by a set of cutpoints along the continuum of possible types in the range

, i=C, D. I first define these cutpoint strategies, and then turn to the formal characterization of two equilibria in this game.

By subgame perfection, C would stand firm at the final decision node if and only if his expected payoff from war is greater than or equal to that of backing down. Provided that C made a public threat, this condition holds when w_C≥−a_C≡α, where α denotes a unique type that is indifferent between standing firm and backing down in public. All types with w_C≥α stand firm (resolved types) in public, and all other types with w_C<α back down in public if resisted (public bluffers). Similarly, provided that C made a private threat, he would stand firm if and only if w_C≥0≡β, where β denotes the critical type that is indifferent between standing firm and backing down in private. All types with w_C≥β stand firm (resolved types), and all other types with w_C<β back down when resisted in private (private bluffers).

To complete the definition of C's strategy, consider his initial decision. Suppose that there is a unique type that is indifferent between going public and staying private when making a threat. Let κ denote this unique type's value for war such that all types with w_C≥κ make a public threat in equilibrium, and all other types with w_C<κ make a private threat.

To define D's cutpoint strategy, suppose there exists a critical type that is indifferent between resisting and conceding, on seeing a public threat. Let γ denote such a type, so that all types with w_D≥γ resist in public and all other types with w_D<γ optimally make a public concession. Likewise, let δ be the type that is indifferent between resisting and conceding when a private threat is observed. All types with w_D≥δ resist in private, and all types with w_D<δ optimally make a private concession.

Because α and β are determined only by the realized values of the exogenous parameters, there are six configurations of C's cutpoints, for each of which an equilibrium may exist. But because α≡−a_C is bounded above by β by definition, we need only to look for solutions for three of these cutpoint configurations: (i) κ<α<β; (ii) α<κ<β, and (iii) α<β<κ.

I define the public equilibrium as a unique equilibrium with κ<β, which encompasses two cases (i) and (ii) and takes a different form for each case. This equilibrium, however, is not mutually exclusive with the private equilibrium when it exists, which is defined for case (iii). As such, the solution for this crisis game is not unique, and these two equilibria exhaust all possible cutpoint configurations described previously.

The Public Equilibrium

In the public equilibrium, a private threat is ineffective in crisis bargaining: all resolved types go public, and the status quo always prevails as a result of a private threat. Upon seeing a private threat, D always detects that the threat is a bluff and hence resists. In response, C always backs down quietly without being caught by his domestic audience. Observe that backing down in private has no consequences different from the status quo. Hence making a private threat is inconsequential in the public equilibrium. As such, C must go public to compel D to concede. Once C goes public, however, equilibrium behavior in the public equilibrium is equivalent to the crisis dynamics generally captured by common audience cost models (e.g., Fearon 1997; Ramsay 2004; Schultz 1998). That is, by going public, C can attract the attention of his domestic audiences and thereby enhance the credibility of his threat by tying his hands. Because a public threat creates audience costs that C would suffer ex post if he backed down, credibility enhancement through hand-tying may have some perverse side effects that increase the risk of inefficient outcomes such as fighting unwanted wars and costly public concessions.

To understand this dynamic, consider the players' equilibrium strategies. A cutpoint strategy of C takes the following form in the public equilibrium. First, all types that make a private threat will back down if resisted because these types have low values for war (i.e., w_C<β). Although types below

will never go public in equilibrium, when α<κ, there exist some types in

whose off-the-equilibrium-path strategy is to stand firm in public. Second, all types above

make a public threat. If α<κ, all public threats are genuine because these types above

will carry through on the threat. If κ<α, on the other hand, the mid-valuation types with

will back down if resisted. Hence, public threats are genuine only if they are made by types above α, and the rest of public threats made by types in [κ, α] are bluffs. However, when κ<α, some “publicly genuine” types in [α, β] would back down privately off the equilibrium path if resisted.

Given C's cutpoint strategy, whenever D observes a private threat, she forms the posterior belief that the threat is a bluff, and hence she always resists. When D receives a public threat, on the other hand, she will resist if and only if her expected payoff from doing so is greater than that from conceding. This means that all types of D above

will resist the public threat, whereas all types below

will make a public concession. D chooses her optimal resistance rate

so that that the type of C with

is indifferent between public and private threats. This analysis is illustrated in Figure 2, and summarized in the following proposition.

The Public Equilibrium

Proposition 1. If κ<β, there exists a unique perfect Bayesian equilibrium of the crisis game with the following strategies. C makes a public threat if

, and a private threat otherwise. When

, C always stands firm in public if resisted. When

, on the other hand, C stands firm in public if w_C≥α, and backs down otherwise. D resists all private threats and resists a public threat if

.

This crisis dynamic and its outcome depend on the relative magnitude of C's audience costs. If the audience costs for C are high, irresolute types with w_C<α will shy away from a public commitment, because a “punishment mechanism” effectively counteracts C's incentives to misrepresent his type. As Figure 2a illustrates, a public threat fully separates types of C in this case. If the audience costs are low, on the other hand, irresolute types will have an incentive to bluff and run a risk of backing down in public. As Figure 2b depicts, this gamble will work if D's valuation of war is low

, but otherwise it will result in costly diplomatic humiliation. Hence, as C's audience cost increases, a public threat conveys more credibility, and thereby the probability of D's resisting decreases.⁵

Define r_pub as D's probability of resisting a public threat. Then,

, where

.

The next result summarizes this argument.

Corollary 1.1. In the public equilibrium, bluffing may occur when

.

The introduction of audience costs for D gives rise to the comparative statics regarding informational effects of a public threat. A threat is said to be informative if it increases D's belief that C will follow through on his threat. Then, the next result follows.

Corollary 1.2. A public threat becomes more informative as audience costs for D get larger. Further, a public threat is fully informative when

.

When audience costs of a public concession are engaged, D finds it harder to concede. So the increase in the magnitude of D's audience costs raises the probability that D is induced to resist if challenged publicly. Facing a higher probability of resisting, only resolute types with w_C≥α can afford to make a public threat. Accordingly, imposing audience costs on D improves the ability of higher types, for which

, to distinguish themselves from lower types. This will cause less bluffing than if there were no audience costs for D (i.e., a_D=0), as is the case with conventional audience-cost models. In fact, Corollary 1.2 shows that bluffing will never occur when D's audience costs are large enough (i.e.,

). Hence, when D suffers from greater audience costs ex post, C can convey greater credibility.

The Private Equilibrium

There exists another equilibrium, the private equilibrium, to the crisis game, in which a private threat can compel D to concede under certain conditions. In principle, because signals essentially become costless and nonbinding when going private, the credibility problem would be a major obstacle that C faces in attempting to convey information through private communication. In this equilibrium, however, improving the credibility is not necessary for a private threat to persuade D to concede. In fact, a private threat induces D to revise downward her beliefs about the credibility of the threat. Nevertheless D concedes privately under broad conditions.

The driving force of the private equilibrium is the behavior of the midrange valuation types of C, both on and off the equilibrium path. In essence, these types could have gone public in making a threat to enhance the credibility of their threat, but in equilibrium they instead forgo this public option and go private. This “deviation” from a putative equilibrium path implicitly signals that some private threats are genuine and hence induces D to concede as effectively as a public threat.

To see this dynamic more formally, suppose κ>β. Because α≤0 by definition, there is only one feasible cutpoint configuration that satisfies this condition: α≤β<κ. With this configuration, all public challenges are genuine in equilibrium because, by subgame perfection, C backs down privately if w_C<β, and backs down publicly if w_C<α. Consequently, if q_pub denotes D's posterior belief that C will stand firm if resisted conditional on a public threat, C can induce D to form q_pub=1 and hence eliminate uncertainty about his type by going public.

In contrast, complete revelation of information does not occur with a private threat in this equilibrium. Recall that C goes private if w_C<κ. Then, because α≤β<κ, there exist some types that make a private threat and stand firm if resisted in equilibrium. Hence, there is a positive probability that a private threat to use force is genuine. This leaves residual uncertainty about C's types, and therefore, on seeing a private threat, D updates her beliefs such that q_pri=(F_C(κ)−F_C(β))/(F_C(κ))>0.

Given these beliefs, D resists in public if w_D≥γ, and in private if w_D≥δ. Because q_pub=1, D knows that war will ensue when she resists in public; hence her choice reduces to a public concession or war. Subgame perfection implies that the highest type that concedes publicly is

. Likewise, because q_pri>0, D can either concede privately for a certain payoff of 0 or resist for a gamble that C is bluffing. If C turns out to be a bluffer (i.e., w_C<β), this gamble pays off; if he is genuine (i.e., w_C≥β), private resistance results in costly fighting. This dilemma then forces unresolved types of D with w_D<δ to concede because they cannot afford to gamble. Hence, letting

denote the highest type of D that concedes privately in equilibrium, it follows that in equilibrium

must solve the indifference condition between a private concession and a private resistance: q_pri(w_D)+(1−q_pri)=0.

Given D's optimal strategy,

and

, C decides whether to go public or private by choosing

so that the critical type w_C=κ is indifferent between a public and private threat in equilibrium. Because κ>β, this indifference condition for w_C=κ is given by

. In equilibrium C's optimal

must solve this condition. This indifference condition also implies that D must concede at the same rate both in public and in private (i.e.,

)). This analysis is summarized in the next proposition, and illustrated in Figure 3.

The Private Equilibrium

Proposition 2. If α≤β<κ, there exists a unique perfect Bayesian equilibrium with the following strategies. C makes a public threat if

and a private threat otherwise. C always follows through with a public threat if resisted; he stands firm with a private threat if w_C≥β and backs down in private otherwise. D resists, on receiving a threat, whether it be public or private, if

.

This cutpoint strategy in the private equilibrium generates four sets of types for C, three of which send a private threat. These four sets of types and their corresponding behavior are summarized in Figure 4. First, if

, C always goes public and never backs down from his public threats. This type is so willing to fight that he has no interest in diplomatic solutions whatsoever. I label this type of C “hard-liner.”

Challenger Types and Equilibrium Strategy from Proposition 2

Second, if

, C makes a private threat but stands firm (SF_pri) when resisted. I call types in this range “moderate.” Although moderate types' level of resolve is high enough to stand firm both in public and in private, they instead seek private channels that enable low types of D, for which

, to concede privately. Because κ>β, no types in this range would do better by making a public threat.

Third, if w_C∈[α, β], C makes a private threat and backs down (BD_pri) when resisted. I call this set of types “conciliatory.” Although conciliatory types never make a public threat in equilibrium, subgame perfection implies that their off-the-equilibrium-path strategy is to stand firm (SF_pub) should they face a choice between standing firm and backing down in public. The appendix shows that no conciliatory types can profitably deviate by making a public threat.

Finally, if w_C<α, the “low” types of C behave in exactly the same manner as the conciliatory type in equilibrium: stay private and back down. But their off-the-equilibrium-path behavior is to back down had they made a public threat instead.

Recall that for private threats to work D must resist a public and private threat with the same probability. This ensures that resolved types (i.e., all types above β) are indifferent between public and private threats. Otherwise, pooling among resolved types occurs, resulting in a higher probability that D faces the choice between a costly public concession and certain war. Yet, if all resolved types are indifferent between public and private threats, why does the cutpoint

partition these types into two regions: moderates going private and hardliners going public?

Observe that

is chosen to include some resolved types in the private-threat-pool so that private resistance still entails a risk of war. Facing the choice between a risk of war and a private concession, D does not always resist in private as she would in the public equilibrium. This means that

splits up resolved types to ensure that D receives a “proper” message that private resistance carries some risk while a private concession incurs no audience cost. Hence, this split among resolved types by

is necessary to send D a private signal that induces beliefs that rationalize private concessions.

We now consider when a private threat works. The next proposition summarizes the conditions for existence of the private equilibrium.

Proposition 3. The private equilibrium exists if the following conditions are met: (i) a public threat is always credible

; (ii) D's audience costs are reasonably high

.

The first condition follows immediately from the cutpoint configurations: α≤β<κ. It is only “hard-liner” types that make a public threat, and therefore every public threat in the private equilibrium must be credible. This condition further implies that a private threat can be effective only in the shadow of a credible public threat.⁶

This result is analogous to Austen-Smith and Banks' (2000) result that the availability of costly signals (i.e., burned money in their model) renders cheap talk more informative.

The second condition states that D must incur reasonably high costs in the event of a public concession in front of her domestic audience. This condition highlights the key to the private equilibrium: D's sensitivity to her audience costs. As long as D incurs audience costs when conceding in public, a private concession becomes preferable for her, and consequently C can compel more types of D to concede optimally by relinquishing the credibility enhancement device of the public option. It is interesting to note that a viable domestic audience exists on D's side, although it need not exist on C's side (i.e., a_C≥0, a_D>0). Hence, if a public concession does not impose audience costs on D (i.e., a_D=0), the private equilibrium collapses, regardless of C's audience costs.

Moreover, the lower bound on a_D illustrates the fact that the public and private equilibria are not mutually exclusive. Recall there is no upper bound on a_D for the existence of the public equilibrium, whereas

partitions the public equilibrium into the bluffing and nonbluffing cases. Because the relative magnitudes of this threshold for the public equilibrium and the lower bound of the private equilibrium cannot be determined, the existence condition for any of the two cases of the public equilibrium can overlap with that of the private equilibrium as long as

. Hence, the two equilibria simultaneously exist except for

.

EFFICIENT SECRECY

I have shown that there exist two equilibria and they exhaust three possible cutpoint configurations in the crisis game. Each equilibrium captures a distinctive equilibrium mechanism, through which state leaders can signal their private information in international crises: going public and staying private. Given the multiplicity of the equilibria, although the model does not allow us to predict which mechanism state leaders will choose in crisis diplomacy, we could at least ask the question of which mechanism will provide a more efficient solution to crises. I address this question by comparing the welfare values of the two equilibria.

Proposition 4 presents a simple efficiency result. From both an ex ante and “interim” perspective (Holmström and Myerson 1983), the private equilibrium is Pareto superior to the public equilibrium. In general, both C and D are always (at least weakly) better off with the private equilibrium (when it exists) than the public equilibrium regardless of types. In particular, the private equilibrium weakly interim dominates the public equilibrium, in which all types of C above β strictly prefer the private equilibrium.⁸

Proposition 4 (Efficient Secrecy). For any types of C and D, the private equilibrium is ex ante efficient. For all types of C with w_C<β and for any types of D, the private equilibrium strictly interim dominates the public equilibrium. If w_C≥β, C is indifferent between the two equilibria.

Intuitively, when a private threat works, it can expand the range of peaceful settlements that are mutually acceptable. Comparing Figures 2 and 3, it is obvious that peaceful outcomes are possible under broader conditions in the private equilibrium than in the public equilibrium where only a public threat is credible. This result is driven by two facts. First, a private threat makes a private concession possible, which is unobtainable in the public equilibrium. Second, going private reduces the ex ante risk of war, compared to a public threat.

Turning to the equilibrium probabilities of war in the two equilibria, note that the war outcome is limited to the higher types (i.e., w_C≥β) in the private equilibrium, although it can occur even if C has a relatively low valuation of war in the public equilibrium (i.e.,

, where

). To have a war outcome in the private equilibrium, C's value for war must be greater than or equal to his status quo valuation, but it is relatively easier to have war in the public equilibrium. Moreover, a public threat is more likely to induce D to resist at a higher rate overall in the public equilibrium. Together, the expected probability of war across the entire range also falls in the private equilibrium. The next corollary establishes this result.

Corollary 4.1 (Risk of War). The ex ante probability of war is strictly greater in the public equilibrium than in the private equilibrium.

Clearly, a private threat has advantages for both C and D. On the one hand, because staying private allows C to pretend as if nothing had happened, secret diplomacy can generally secure leeway for irresolute types of C to disavow the threat if D resists and avoid domestic political costs associated with backing down. This leeway also allows the midvaluation types of C to avoid getting “locked into” costly fighting due to his relatively high audience costs.

On the other hand, a private threat helps irresolute types of D (i.e.,

) escape from a costly public concession and therefore helps her avoid being forced to fight an unwanted war in order to protect her honor from public humiliation. For this reason, a private concession is attractive for D, as it effectively lowers the costs of a concession. As such, secrecy allows D to rationally make a private concession, which is not attainable in public diplomacy.

Historical norms are that state leaders employ secrecy as face-saving tactics in order to facilitate cooperative outcomes and tension reduction. For example, during the final phase of the Cuban Missile Crisis, President Kennedy wanted to make sure that “Every opportunity was to be given to the Russians to find a peaceful settlement which would not diminish their national security or be a political humiliation” (Kennedy 1969, 81).

Contrary to the popular perception that transparency or “open diplomacy” carries beneficial effects in the age of democracy (Finel and Lord 1999; Nicolson 1963), efficient secrecy posits that the private equilibrium is a more valuable mechanism for almost any type of players, as it can lead to better bargaining outcomes than the public equilibrium can. Moreover, inefficient bargaining failures due to players' incentives to “go public” and signal to an outside audience are ubiquitous not only in international disputes (Kydd 2006) but also in labor bargaining (Cai 2000) as well as legislative politics (Groseclose and McCarty 2001). Likewise, as Louis XIV observed about 3 centuries ago, public diplomacy feeds bargainers with incentives for manipulative political “posturing”:

ALASKA BOUNDARY DISPUTE, 1903

It is useful to illustrate how the logic of efficient secrecy operates in an actual case. I present a historical episode from the Alaska Boundary Dispute, where President Theodore Roosevelt's “speak softly and carry a big stick” foreign policy was most evident.

In 1902 the United States claimed the Canadian territory adjacent to Alaska along the Pacific coast. The origin of the dispute dated back to the Anglo-Russian Treaty of 1825 drawing the boundary between Britain and Russian territory. This boundary rendered the Russian territory, which the United States later purchased, valueless because it was encompassed by mountains and irregular coastlines. The border became strategically important when gold was claimed to be discovered there. This dispute was eventually resolved in October 1903 in the favor of the United States—and the United States gained a town now known as Juneau—with an appearance of reasonable compromises. In fact, Canada had conceded in the face of Roosevelt's private threats of waging war.

In March 1902, when Secretary of State John Hay warned Roosevelt of the risk of a miner's uprising if gold were discovered in Alaska, Roosevelt decided to send troops in to southern Alaska “as quietly and unostentatiously as possible … to prevent any possible disturbance along the disputed boundary line” (Collin 1985, 174–78). In the meantime, Roosevelt sent a message to Ottawa about the possibility of violence, implicitly challenging Canada with a territorial demand. Knowing that Roosevelt would not pull back the troops from the disputed area, Prime Minister Laurier wanted to make a private concession so as to avoid an apparent surrender of territory to Roosevelt's threats to draw the boundary. So Laurier's government proposed an arbitrational settlement so that his government could at least save face with Canadians. Laurier reportedly “pleaded to Henry White, the head of the American Embassy, that he would like to ‘save his face’ with Canadians by an arbitration” (Nevins 1930, 192–93).

Roosevelt from the beginning, however, refused to arbitrate this dispute or to consider any settlement whatsoever short of a complete victory. Still, he was willing to gesture a compromise as long as the United States obtained the claimed territory (Beale 1956, 115–16). Having been briefed on White's meeting with Laurier, Roosevelt wrote to Hay,

Britain did not have a strong interest in the territory, and by 1903 it was having difficult foreign relations with France, Germany, Japan, and Russia. So Britain did not want her relations with the United States to suffer from the contestation over the Alaskan border. Britain's decision therefore was primarily based not on Alaska but on the essential need to maintain friendship and détente with the United States (Collin 1985, 183; Francis, Jones, and Smith 1992, 115–16; Penlington 1972, 92). As for the Canadian leaders, they really had only two choices, given Roosevelt's high resolve: either to conclude the tribunal favorably for the American case and save face with fellow Canadians, or to surrender territory to the U.S. forces in public and be humiliated.

Roosevelt chose to threaten Britain rather than Canada because of the six tribunal members the British representative, Lord Alverstone—Chief Justice of England and President of the Alaska Boundary Tribunal, was a pivotal voter on the tribunal. The United States kept sending messages to the British leaders to convince them that Britain's “self-interest would be better served by aligning with America rather than Canada” (Collin 1985, 182), and threatened that should they fail to win the American case, the United States would draw the boundary using military force (Francis, Jones, and Smith 1992, 116; Penlington 1972, 89–90).

Roosevelt gave Senator Henry Lodge not an “official and authoritative” instruction but a private letter to be shown to British leaders including the Prime Minister Arthur Balfour, the Foreign Minister Joseph Chamberlain, and the Liberal Party leader William V. Harcourt, as well as to Justice Alverstone whose vote was decisive in settling the case with a complete American victory. This letter was the first of a series of Rooseveltian threats to intimidate British authorities.

What Roosevelt later identified as “one of the decisive elements in the eventual American victory,” was another personal letter he had delivered through Supreme Court Justice Oliver W. Holmes to Chamberlain at a private meeting (Collin 1985, 179–80). Its message was: settle or fight. Roosevelt instructed Justice Holmes in this letter dated July 25, 1903 that:

In the end, the British representative Alverstone sided with the Americans and accepted their position of the boundary and territorial control as the United States claimed. Although Alverstone had committed himself to a compromised division of the disputed territory, including Canada's ownership of four islands, he reneged on this commitment only five days later and made an arrangement with the three Americans. Alverstone was reportedly instructed either by Prime Minister Balfour or by Foreign Minister Chamberlain to side with the American demands (Penlington 1972, 90–99).

On the surface, the dispute appeared to be resolved through a tribunal settlement. But it was actually Britain's ceding the Canadian territory under the private threats Roosevelt repeatedly issued. During the course of this boundary dispute, Roosevelt spoke softly by publicly holding a tribunal, but he carried a big stick by quietly dispatching troops and privately blackmailing Britain and Canada. That way, Roosevelt made it easier for Laurier to surrender the territory.