Foreign Policy Appointments

Abstract

How do leaders select their top-level foreign policy appointees? Through a formal model of the domestic and intragovernmental politics surrounding an international crisis, I investigate the trade-offs shaping leaders’ appointment strategies. In the model, a leader selects a foreign policy appointee, anticipating how the appointment will affect the advice he receives in the crisis, the electorate's evaluation of his performance, and ultimately the policies that he and his foreign counterparts pursue as a consequence. The analysis uncovers a fundamental tension in the leader's ability to use appointments to advance his core political and policy objectives of deterring foreign aggression, obtaining accurate advice, and maximizing domestic approval: any appointment that advances one of these objectives invariably comes at the cost of another, and the leader's appointment strategy must balance across these trade-offs. Analyzing cross-national appointment patterns to the offices of ministers of defense and foreign affairs, I find descriptive evidence consistent with the model's predictions: leaders from dovish parties are more than twice as likely as leaders from hawkish parties to select cross-partisan and politically independent appointees, and such appointments are less likely for leaders of either party as they approach re-election.

These administrations need liberals for domestic tasks. … But for foreign policy it is essential to have men who inspire confidence. This liberals do not do.

John Kenneth Galbraith, The Triumph (1968, 58)

When new leaders are elected to office, one of the first tasks awaiting them is to select their top foreign policy advisors. Conventional wisdom among policymakers holds that these decisions carry significant international ramifications. When introducing Lloyd Austin for a confirmation hearing as President Biden's defense secretary, senator Dan Sullivan proclaimed, “Our allies will take comfort in his confirmation, and our adversaries will take pause.”¹ A primary motive behind President Obama's choice to retain Robert Gates as defense secretary, as Gates recounted in his memoir, was that “my staying in place would show foreigners that US resolve would be undiminished.”² Putting a finer point on the matter, Senate Foreign Relations Committee chair Bob Corker described President Trump's foreign policy team as “those people that help separate our country from chaos … [and] make sure that the policies we put forth around the world are sound and coherent.”³ Analysts and scholars, even those of the realist tradition, frequently refer to major US military engagements as belonging to a particular appointee (“McNamara's War” in Vietnam,⁴ “Madeleine's War” in Kosovo,⁵ “Hillary's War” in Libya⁶), crediting the appointee as the “architect” or “father” of the conflicts undertaken during their time in government service.⁷

Yet despite the perceived importance of top-level foreign policy appointments, we have a limited understanding of how leaders decide whom to place in these roles. One standard logic holds that principals can achieve better bargaining outcomes by delegating bargaining authority to an agent with misaligned preferences.⁸ This sort of hands-tying logic has been used to explain the value of other domestic institutions, notably legislatures⁹ and independent central banks,¹⁰ in improving the credibility of a state's international commitments. Other explanations point to the domestic political value of certain appointments: anecdotal accounts suggest that appointees can provide leaders with “political cover”¹¹ in the face of domestic criticism, enabling them to “inoculate themselves against the ‘weak on defense’ charge”¹² or obtain “insurance against recrimination.”¹³ Experimental evidence confirms that public perceptions of a leader's foreign policy performance can be influenced by public communication from appointees.¹⁴

While these accounts provide important insights into the problem at hand, they also raise a number of questions. Appointments would serve a useful hands-tying function if they did in fact tie the leader's hands to a certain course of international conduct. But unlike other widely studied domestic institutions, foreign policy appointees have no de jure autonomy in the policy process; they typically serve in their roles “subject to the direction of the President”¹⁵ and may be removed from office at will.¹⁶ In this context, the credibility of any delegation of authority should be explained, rather than assumed.

Moreover, these accounts generally address only one side of the ledger. If certain appointments can strengthen the leader's hand at the international bargaining table, or bolster her domestic political standing, why would the leader not always select those appointees? What are the trade-offs implied by different appointment strategies? Why do we observe variation in the appointments made by different leaders, or by the same leader at different points in their administration?

To address these questions, I analyze a formal model of the domestic and intragovernmental politics surrounding an international crisis—a high-stakes setting in which appointee influence cannot be explained as a product of rational inattention by a time-constrained leader.¹⁷ I examine the leader's selection of a foreign policy appointee, in anticipation of how the appointee will shape the advice he receives in the crisis, the electorate's evaluation of his performance, and ultimately the policies that he and his foreign counterparts pursue as a consequence. The analysis uncovers a fundamental strategic tension between the appointment incentives discussed earlier: in general, any appointment who advances the leader's international policy objectives will undermine his domestic political standing, and vice versa. This core insight carries implications for how the leader will optimally staff his administration, and how those appointment decisions will affect citizen welfare and international conflict and cooperation.

The model examines the leader's appointment strategy with respect to two appointee attributes: the appointee's ideological bias (her “hawkishness” or “dovishness”), which denotes her willingness to use force to achieve foreign policy objectives; and her political loyalty to the leader, or conversely her independence, which denotes her willingness to publicly criticize a policy decision that she opposes. I consider how the leader's incentives regarding each attribute vary depending on the leader's partisan identity, as a member of either a Hawk party or a Dove party. To microfound the mechanisms of appointee influence, I model appointees as serving a purely informational function: providing private policy advice to the leader to inform his crisis response, and communicating publicly with a domestic audience to inform their assessment of the leader's crisis performance.

With this approach, the analysis yields three primary implications. First, appointees can enhance deterrence against foreign aggression, through two distinct but related mechanisms: an advisory mechanism, and a fire-alarm mechanism. Through the advisory mechanism, the leader selects an advisor with a known bias from whom he will subsequently solicit policy advice. In doing so, the leader can effectively outsource his “resolve” to a more hawkish advisor—but only insofar as it remains credible for the leader to follow her advice once a crisis arises. Through the fire-alarm mechanism, the leader can appoint a politically independent agent who may let it be known publicly when she believes the leader to have acted against the voters’ interest in his crisis response. This serves to mitigate the moral hazard problem between the voters and the leader, as the threat of exposure can discipline leaders into governing more in line with the voters’ interest. While the advisory mechanism can improve deterrence under leaders of either party, only dove leaders receive a deterrent benefit from the fire-alarm mechanism; for Hawk leaders, curbing their excessive willingness to use force serves to weaken rather than enhance deterrence.

Importantly, the formal analysis demonstrates that these two mechanisms of influence are fundamentally interconnected. The appointee's ability to shape policy through the private advice channel depends, in part, on the threat of “going public” with her criticism in the event that her advice is ignored. Conversely, the incentive for the appointee to communicate her criticisms to an outside audience, and thus the credibility of her communication, derive from her (in)ability to pursue her policy objectives through private advisory processes. The analysis reveals the conditions under which each mechanism is operative, and how each depends on the other to function.

A second implication of the analysis pertains to citizen welfare. Insofar as the leader is able to enhance deterrence through the selection of a foreign policy appointee, this deterrent benefit necessarily comes at a cost: in equilibrium, the appointment will either distort the leader's crisis response away from the voter's ideal (undermining policy responsiveness), or muddle the voters’ ability to distinguish between leaders whose true preferences do or do not align with their own (undermining electoral selection)—or both.

Third, in light of these effects on deterrence and welfare, the analysis yields novel implications regarding the optimal appointments that leaders of different parties will select. Contrary to intuition, the optimal appointment is not necessarily one that maximizes deterrence, nor one that provides the most accurate policy advice, nor one that yields the best re-election prospects for the appointing leader; rather, the leader invariably faces trade-offs among these objectives, and must weigh gains on any one criterion against losses on another. Generally I find that leaders of either party will never select dovishly biased appointees, and that both will select hawkishly biased advisors if they place sufficient value on deterrence. A major asymmetry emerges, however, regarding political loyalty: leaders from the Dove party are strictly more likely to select politically independent appointees, as compared to leaders from the Hawk party. Yet their reasons for doing so are highly contingent: an independent appointee may be selected in some cases solely to improve electoral prospects, but in other cases for deterrent purposes, despite the fact that doing so is electorally harmful.

Empirically, I provide descriptive cross-national evidence consistent with this third main implication. Drawing on annual data on cabinet compositions and codings of party manifestos for fifty-eight countries from 1963 to 2021, we observe the patterns of partisan asymmetry just described in appointments to the office of minister of defense: leaders from a Dove party are more likely to appoint hawkish defense ministers than the reverse, and Dove party leaders are more likely to appoint defense ministers who are non-partisan or from outside the leader's party. I also find, consistent with the theory, that co-partisan appointments are more likely as the leader approaches re-election. Patterns of appointments of foreign affairs ministers are similar, though less pronounced.

This study contributes to three primary bodies of literature in international relations and political economy. First, it advances our understanding of the politics of personnel selection in foreign policy. A recent review by Saunders identifies this as the primary component missing from the recent resurgence of literature on foreign policy elites. “Beyond chief executives,” writes Saunders, “the selection of many crucial foreign policy elites remains mostly unexplored.”¹⁸ This study provides both a theoretical framework for understanding the incentives and trade-offs that chief executives face in choosing their most important foreign policy advisors, and the first cross-national evidence of how these incentives bear on the selection of ministers of defense and foreign affairs.

There are three recent studies that most closely resemble the current model, and which provide different but complementary insights into leaders’ incentives regarding foreign policy appointments. Separate studies by Lindsey and Goldfien consider leaders’ optimal choice of an ambassador who negotiates directly with a foreign government.¹⁹ In contrast to the present findings, Lindsey's focus on outward rather than internal communication gives rise to an incentive for leaders to select dovishly rather than hawkishly biased agents. Goldfien's findings regarding the value of appointee loyalty (or “familiarity”) diverge from those of the present model due primarily to his focus on agents who must operate at arm's length from the president's decision-making process—in contrast to the present focus on agents who constitute the president's core advisory team. Alexiadou, Spaniel, and Gunaydin analyze a leader's decision to appoint a political ally or a politically independent technocrat to the office of finance minister, anticipating the reaction of international bond markets.²⁰ In their model, the appointment is purely a “burned money” signal of the leader's policy intentions, where the appointee herself takes no action after being appointed. In contrast, in the present model, the equilibrium is fully pooling at the appointment stage, and I focus instead on how the appointee shapes policy and politics once in office.

The second major contribution of this study is its investigation of two forms of information that feature prominently in the theoretical literature on international crisis bargaining. Canonical models of crisis bargaining assume that a leader observes some piece of private information that bears on their conflict payoff—such as the cost of conflict, or the valuation of the prize in dispute²¹—without interrogating the source of the information or the process by which it reaches the leader. Other domestically oriented theoretical models focus on how voter assessments of leader performance shape leaders’ incentives in crisis bargaining,²² without considering how those perceptions are influenced by elite cues and framing.²³ The present study advances our understanding of both of these information-transmission processes, highlighting the role executive appointees play in providing the leader with information to guide policymaking, and providing the voters with information to assess leader performance.

Finally, this study contributes to the literature on accountability in multilayered governing hierarchies more broadly, beyond the context of foreign policy. The formal literature on political accountability generally studies the delegation of authority from voters to elected leaders,²⁴ and from leaders to bureaucratic subordinates,²⁵ in isolation from one another. Studies that do consider both agency relationships in tandem focus on a setting in which an elected politician appoints a bureaucrat who then sets policy directly.²⁶ Here, I do not assume that the leader can credibly delegate policymaking authority to his agent; the politician can choose only how to make use of the agent's advice. This approach yields novel and generalizable results regarding how electoral incentives condition the leader's willingness to follow his appointees’ advice, and how internal advisory processes in turn color voter evaluations of the leader's performance.

Multilayered Agency Problems in Foreign Policy

The theory centers on the strategic interdependence between two principal–agent relationships—one between the leader and the voter, and another between the appointee and the leader—and how these relationships influence the state's foreign policy behavior. I begin by describing the international context in which the domestic game unfolds, and then discuss the two agency relationships.

Deterrence at the International Level

This study follows a long tradition in the international-conflict literature by analyzing a game of crisis bargaining between a foreign “challenger” state and a domestic “defender” state.²⁷ The foreign challenger wants to take some provocative action against the defender's interests (for example, a territorial incursion against an ally, or the development or testing of a weapon program). If they do, the defender must then choose whether to retaliate against the provocation (for example, by arming the ally, intervening militarily, or imposing punitive sanctions). The defender's goal is to deter the challenger from taking the provocative action, which requires convincing them of two facts: first, that the challenger's provocation will be met by costly retaliation which renders the provocation undesirable; and second, that such costs will not be imposed if the challenger refrains from provocation.²⁸ In the current setting, the challenger faces uncertainty regarding the defender's true preferences, and thus regarding the credibility of the defender's commitment to uphold both conditions. In light of this uncertainty, we examine how an appointee within the defender state might alter the state's crisis response, and the foreign adversary's expectations thereof. Insofar as she can, this will affect the adversary's incentive to instigate the crisis in the first place.

Hawks, Doves, and Electoral Accountability

I theorize domestic politics as consisting of two key principal–agent relationships. Our model of the first relationship—that between the leader and voter—draws from the literature on adverse selection models of electoral accountability, which views elections as opportunities for citizens to select “good types” of leaders to whom to delegate political authority.²⁹ What exactly constitutes a “good type” depends on the context of analysis; a common approach, which I adopt here, is to model politicians as holding differing degrees of ideological (mis)alignment with voters, where voters seek to select candidates who more closely share their primitive preferences over policy.³⁰

Specifically, I follow Schultz in assuming that the leader is known to be from one of two parties—a “Dove” party or a “Hawk” party—with differing reputations regarding their relative willingness to enter into conflict to defend the national interest.³¹ Unknown to the voter (and the foreign challenger) is whether the leader is a “moderate” whose policy preferences align with the voter's, or an “extremist” who is either excessively willing (if an extreme Hawk) or excessively unwilling (if an extreme Dove) to engage in conflict. While extremists hold strong ideological commitments to their preferred policies, moderate leaders are more willing to adjust their policy responses in the face of uncertainty, and to adapt to the facts of the particular situations they confront. As we will see, this makes moderates more intrinsically receptive to expert policy advice when formulating their crisis response.

Given this specification of voter preferences and informational asymmetry, a straightforward implication is that leaders face an electoral incentive to act “against type”: Hawk party leaders benefit from conciliatory foreign policy behavior that signals moderation, and Dove party leaders conversely benefit from demonstrating a willingness to use force. This implication receives empirical support from a number of experimental studies:³² Trager and Vavreck, for instance, find that independent voters harshly penalize Democratic presidents for staying out of a conflict, but reward Republican presidents for doing the same.³³ Mattes and Weeks likewise find that Dove leaders face much more public disapproval than Hawk leaders for seeking a conciliatory policy against a foreign adversary; a key mechanism behind this result is that Hawk leaders are largely perceived as moderates when they pursue such a policy, whereas Dove leaders are perceived as extreme pacifists.³⁴

Incentives and Influence of Executive Appointees

The model of the second agency relationship—between the leader and his appointed advisors—builds on a long tradition of scholarship on bureaucratic politics and foreign policy. Graham Allison's Essence of Decision was most prominent in proposing a “governmental politics” model of foreign policy decision making characterized by “bargaining along regularized circuits among players positioned hierarchically within the government.”³⁵ In the view of this and related work,³⁶ bargaining with subordinates is an inevitable feature of executive governance; it arises by assumption that appointees can and do constrain their leader's international conduct.

This early wave of scholarship was subsequently criticized for its underestimation of presidential power in foreign affairs. “Since the president has personally appointed the top officials in his administration and can dismiss them at any time,” Bendor and Hammond ask, “and since he has substantial formal authority … to order them to do what he wants, why must the president bargain with them?”³⁷ The present study takes up the challenge posed by Bendor and Hammond: starting from the premise that formal authority is vested in the leader who sits atop the governmental hierarchy, I proceed to interrogate how and under what conditions bureaucratic delegation can constrain the leader's foreign policy behavior.

My explanation focuses on two informational channels through which appointees can influence foreign policy: an advisory mechanism, and a fire-alarm mechanism. I outline each of these mechanisms in turn, and then discuss the strategic interdependence between them.

Appointees as Advisors

One mechanism through which appointees can influence the leader's behavior is their provision of policy-relevant advice. Canonical models of bureaucratic delegation assume that a bureaucratic agent holds an informational advantage over a political principal, which provides the basic rationale for the principal's willingness to delegate.³⁸ The cost of the agency relationship, from the principal's perspective, is that the agent's interests may diverge from the principal's, and the agent can exploit her informational advantage for her benefit and at the principal's expense.

The present analysis adopts this approach. I assume that the agent has expertise with respect to some policy-relevant state of the world, and that she provides private advice to the leader to inform his policy decision.³⁹ The preference divergence between the principal and agent takes the form of the agent holding some degree of ideological “bias,” which leads her to shade her advice in a hawkish or dovish direction. However, I do not treat this misalignment as an inevitable cost of the agency relationship; rather I show that, even when a perfectly unbiased agent is available, the leader may actually prefer to commit ex ante to distorting his own informational environment by relying on biased advice.

Some recent studies have examined the sort of internal advisory processes modeled here. Jost and coauthors show that the decisions coming out of a foreign policy advisory meeting with the US president systematically reflect the hawkishness of the group of advisors in attendance at that meeting.⁴⁰ Saunders argues that the ability of advisors to sway their leader's foreign policy decisions depends on the relative experience and expertise of the leader and his advisors,⁴¹ a result which I recover formally in the present analysis.

But if we believe—as is widely posited in the electoral-accountability literature mentioned earlier—that leaders vary in the rigidity of their ideological commitments, and thus in their receptiveness to expert advice, then the process of foreign policy advising takes on a much more political complexion. Providing perfectly accurate information through private advisory channels may be insufficient to sway a leader's decision when he is strongly predisposed to a particular course of action. Thus the advisory mechanism is most effective when operating in conjunction with the second mechanism of influence, as I discuss next.

Appointees as Fire Alarms

A second channel of appointee influence over foreign policy is their communication with a domestic audience that can hold the leader accountable. Following previous work, I refer to this as the fire-alarm mechanism.⁴² The potential misalignment of interests between the leader and voters, along with the leader's private information regarding the value of the policy options he faces, give rise to a moral hazard problem, whereby the voters cannot perfectly monitor the leader's action to ensure that he is governing in their interest. If the appointee can credibly threaten to expose the leader's crisis mismanagement to the voters—to sound the fire alarm—this can serve to discipline the leader's behavior and mitigate the moral hazard.

Bendor and Hammond suggest a mechanism of this sort as a potential microfoundation for the bargaining between presidents and subordinates that is assumed into Allison's model: “While the president can often order his political appointees and their respective bureaucracies to do his bidding, they may be able to hurt him politically if they disagree with his choices and make their disagreements known to outside supporters.”⁴³ Halperin and Clapp likewise identify “finesse in threatening to leak information or to resign” and “aptitude for mobilizing support outside the bureaucracy” as key factors that determine an official's influence over policy.⁴⁴

Public signaling by appointees can manifest in a variety of ways. In the most extreme instance, an appointee can resign in protest over a particular decision made by the leader. Other forms of protest include issuing public statements, leaking damaging information to the press or to other actors across the government, or simply refraining from expressing support for the leader's policy when given the opportunity to do so. Even these milder forms of protest carry some risk of losing standing with the president—potentially to the point of being forced out of office—so for simplicity I focus on the stark case of resignation in the formal analysis.

Some recent experimental work has found that these sorts of public cues from appointees can affect voter perceptions of a leader's foreign policy performance. Saunders shows that public approval of a president's decision regarding military intervention can be influenced by an advisor's public statement supporting or opposing the intervention.⁴⁵ Jost and Kertzer further show that voters are attuned to quite nuanced differences in appointee attributes—including experience, institutional position, age, and education—when determining how much stock to put in a given official's policy recommendation.⁴⁶

If appointees taking their disagreements public can harm the leader's standing with some relevant domestic audience, then it stands to reason that the leader may opt to accommodate the appointee's policy concerns rather than suffer the consequences of public disapproval—consequences which can range from undermining the leader's ability to advance her policy agenda through congress⁴⁷ or through unilateral executive action,⁴⁸ to ultimately diminishing her prospects of re-election.⁴⁹ Two questions arise, however.

First, while the threat of going public may be useful in advancing the appointee's policy goals, why would the appointee actually want to carry out the threat? Doing so requires overcoming the daunting “effectiveness trap.” As historian and former National Security Council staffer James Thomson writes, “The inclination to remain silent or to acquiesce in the presence of the great men—to live to fight another day, to give on this issue so that you can be ‘effective’ on later issues—is overwhelming.”⁵⁰ Once the appointee has lost on a particular issue, it is not obvious why she would sacrifice future opportunities for policy influence in order to make a statement about the past. But if this incentive is dominant, then the threat of the fire alarm is rendered noncredible, and the leader's behavior is unaffected. Second, if the appointee does go public, why should the voters listen to what she has to say? Why is the appointee's choice to sound the alarm informative of the leader's quality, rather than merely indicative of the appointee's own preferences and priorities?

The formal analysis demonstrates that the answer to these questions lies in the interaction between the two mechanisms of appointee influence. In the model, the primary motive for appointees to remain in government service is the opportunity to influence future policy decisions through the advisory process—the ever-present desire for effectiveness, as Thomson describes. Crucially, however, this potential for influence varies across leaders. If the appointee assesses that the leader is unwilling to make use of expert advice, then her choice to speak out or resign comes at a lower cost than if she thought the leader was receptive to expertise. It is this differential costliness that renders appointee protest an informative signal of the leader's quality: the decision to protest is indicative of what sort of future opportunities the appointee believes she is giving up. This in turn explains why the audience would revise their appraisal of the leader in response to the appointee's action.

We can see this logic exemplified in two high-profile appointees’ decisions over whether to resign in the face of disagreements with their principals. When President Trump announced his intention to withdraw American troops from the anti-ISIS coalition fighting in Syria, the unexpected decision presented defense secretary James Mattis with “the most urgent crisis of his nearly two years in the Cabinet.”⁵¹ Mattis pleaded with the president to reverse course, but he was firmly rebuffed. In response, Mattis resigned, releasing a public letter in which he outlined his core beliefs regarding “treating allies with respect and also being clear-eyed about both malign actors and strategic competitors.” “Because you have the right to have a Secretary of Defense whose views are better aligned with yours on these and other subjects,” Mattis wrote, “I believe it is right for me to step down from my position.”⁵²

William Perry also confronted a major policy disagreement with President Clinton during his time as defense secretary, in this case regarding the rapid accession of several Eastern European countries into NATO. “In the strength of my conviction, I considered resigning,” Perry later reflected. “But I concluded that my resignation would be misinterpreted … President Clinton had given me just what I had requested—an opportunity to state my case—and unfortunately, I had not been persuasive enough.”⁵³

These examples illustrate the forward-looking assessments appointees engage in when deciding whether to speak out against a leader. Perry's policy disagreement with Clinton was just that—a disagreement. He emerged from the experience still believing Clinton to be receptive to expert advice, so he saw value in continuing to serve in the administration. Mattis's disagreement with Trump, however, revealed a deeper defect. Mattis had previously “operated under the illusion that he could change Trump's views, or at least some of his foolish ways.”⁵⁴ Trump's handling of the Syria withdrawal indicated to Mattis that there was little potential for him to contribute constructively to future policy decisions. It was this inference, I argue, that drove Mattis's decision to resign, and that in turn shaped the audience's interpretation of his decision.

Implications for Leaders’ Appointment Strategies

Anticipating these mechanisms through which appointees can influence the leader's behavior, and the ensuing effects on the foreign adversary's incentives, the leader must decide how to optimally staff his administration. The model considers appointees selected from a two-dimensional attribute space, corresponding to the two mechanisms of influence described earlier: the appointee's hawkish or dovish bias; and the appointee's loyalty or independence (her reluctance or willingness to speak out against the leader).

Though it may be most intuitive to think of these attributes as highly correlated—dovish appointees of Dove leaders will be loyal, hawkish ones disloyal, and vice versa for Hawk leaders—the correlation is far from perfect. Some examples are instructive. Robert Gates, the defense secretary originally appointed by George W. Bush and retained for the first two years of the Obama administration, proudly identified himself as a hawk on national security matters.⁵⁵ Yet he also proved to be a politically independent actor with respect to both presidents under whom he served. In his confirmation hearing in December 2006, when asked “Do you believe that we are currently winning in Iraq?” Gates gave the surprisingly blunt response, “No, sir”⁵⁶–—directly contradicting President Bush's answer of “Absolutely we're winning” just two months prior.⁵⁷ Later, when President Obama was deliberating over options for a troop surge in Afghanistan in late 2009, Gates was one of the chief proponents of a larger troop presence. Beyond his internal advocacy of his preferred policy, it was his willingness to resign in protest—commonly understood, if not threatened explicitly⁵⁸—that made his position especially persuasive.

Robert McNamara was likewise a hawkish defense secretary brought into an administration intent on demonstrating that it was not as dovish as its party image suggested. In contrast to Gates, however, McNamara—though a registered Republican, and a consistent advocate for a more interventionist foreign policy (at least in the early years of his tenure)—was a fiercely loyal subordinate to both Presidents Kennedy and Johnson.⁵⁹ As one telling example, McNamara developed a practice of categorizing his internal memoranda as “draft recommendations,” which he would then revise to reflect the president's ultimate decision, “so that there would be no record for history of any difference between the Secretary of Defense and the President.”⁶⁰

Another Kennedy appointee, Undersecretary of State Chester Bowles, fits the opposite model of a dovish agent with relatively weak political loyalties to his co-partisan president. Bowles, as an intellectual leader of the party's liberal wing prior to his appointment, came to be seen by his allies as a “litmus paper of the Administration,” someone who could inform them if the administration was veering too far from the party's values.⁶¹ This proposition was tested early on, when Bowles advocated internally against the ill-fated Bay of Pigs invasion; after the mission's failure, Bowles leaked to the press his earlier memorandum opposing the decision—a move which precipitated his ultimate dismissal from his prominent post at State.⁶²

These examples suggest that leaders have a rich set of appointment options across the two-dimensional space of ideological bias and political loyalty. When weighing their appointment options along these dimensions, leaders face a set of complex and potentially conflicting incentives. They want to induce certain behaviors from foreign actors (deterring aggression, in the present model); they want to elicit the best advice to guide policymaking; and they want to maximize their domestic political standing as a result of their foreign policy conduct. To evaluate leaders’ optimal appointment strategies in light of these various incentives, we turn to the formal analysis.

Formal Model

The model examines the domestic politics within a home (“defender”) country, in the context of an international crisis against a foreign (“challenger”) country. To fix terms, we say that a crisis occurs when the foreign government takes some provocative action against the home country's interests. The home government must then choose whether or not to take a costly retaliatory action. We refer to the foreign government's action as initiating a challenge, and the home government's action as either fighting or conceding.

Within the defender country are three players: a leader L, an appointee (or agent) A, and a representative voter V. The foreign government is treated as a unitary actor, F. The domestic leader is either from a Dove party (j = D) or a Hawk party (j = H), which is fixed at the start of the game and known by all players.

The game begins with the leader selecting his appointee, characterized by two attributes which I will describe later: a bias parameter $\pi _{\rm A}^k $, and a loyalty parameter λ. Following the appointment, the foreign government's conflict valuation ω _F is realized, and F decides whether to initiate a challenge (a _F = 1) or not (a _F = 0) against the home country. The home country's conflict valuation ω is then realized, and the appointee offers private advice to the leader in support of (s = 1) or against (s = 0) fighting. The leader chooses whether to fight (a = 1) or not (a = 0), and the appointee chooses whether to protest the leader's decision (z = 1) or remain silent (z = 0). Finally, the domestic voter chooses whether to retain the leader (r = 1) or remove him (r = 0) in favor of his electoral opponent. The game sequence is summarized in Figure 1.

FIGURE 1. Game sequence

The foreign government's incentives are specified sparsely, so as to focus attention on the domestic politics within the home country. F has a valuation ω _F for the issue in dispute,⁶³ which is distributed uniformly on the interval $[ \underline{\omega }_{\rm F}, \;\bar{\omega }_{\rm F}] $, where $[ \underline{\omega } _{\rm F}, \;\bar{\omega }_{\rm F}] \supseteq [ 0, \;1] $. They prefer to win the issue uncontested, and they incur a cost normalized to 1 if the home government fights back (or if the latter initiates conflict unprovoked). Thus F's payoff is

(1)$$U_{\rm F} = a_{\rm F}\omega _{\rm F}-a $$

There are two sources of informational asymmetry between the domestic leader and voter. First, the leader has an informational advantage with respect to a policy-relevant state of the world, ω ∈ {0, 1}. This variable represents, in simplified terms, the net value of engaging in conflict with the foreign adversary: taking account of the valuation of the issue, the costs of conflict, and the likelihood of success in conflict, we say that it is either in the national interest to fight (ω = 1) or it is not (ω = 0). The state ω is drawn randomly by nature, with probability

$${\rm Pr}\left( {\omega = {\rm 1}|a_{\rm F}} \right) = \left\{ {\matrix{ {\tau ,} \hfill & {a_{\rm F} = 1} \hfill \cr {\tau _0 < {\rm }\tau ,} \hfill & {a_{\rm F} = 0} \hfill \cr } } \right.$$

—meaning that F's challenge serves to increase (from τ ₀ to τ) the probability that fighting is in the home country's interest.⁶⁴

To simplify the exposition, we will focus on the limiting case of τ ₀ → 0:⁶⁵ the leader has the option to engage in unprovoked aggression,⁶⁶ but it is very rarely in the national interest to do so.⁶⁷ Following F's action, the leader does not observe ω directly but receives a private signal x ∈ {0, 1}, with ${\rm Pr}( x = \omega \vert \omega ) = \phi \;\in \;\left({{1 \over 2}, \;1} \right)$ (along with the private advice given by the agent, as described later).

The second source of informational asymmetry is the leader's privately known policy preference, which is either congruent (θ = 1) or incongruent (θ = 0) with that of a representative voter. The voter wants the leader to match his action to the state of the world—fighting if and only if the value of conflict is high (ω = 1)—which we can express as

$$W_{\rm V} = a\omega + ( {1-a} ) ( {1-\omega } ) = {\mathbb 1}[ {a = \omega } ] $$

Following Schultz,⁶⁸ we assume that the congruent leader (alternatively referred to as a “moderate” type) shares the voter's policy preference, while the incongruent leader (an “extreme” type) has a state-independent preference for taking one action over the other: an incongruent leader of the Dove party prefers conceding (a = 0) regardless of ω, and an incongruent Hawk prefers fighting (a = 1) regardless of ω.

The leader knows his own type, but all other players only know the prior probability of each type, Pr(θ = 1) = π ∈ (0, 1). We can denote the leader's policy payoff as

$$W_{\rm L} = \theta W_{\rm V} + ( {1-\theta } ) \left\{{\matrix{ {1-a,\quad j = {\rm D}} \cr {\,\,\,\,\,\,\,\;\;a,\quad j = {\rm H}} \cr } } \right\}$$

In addition, the leader suffers a cost of γ > 0 when the foreign government challenges, and enjoys an office-holding benefit of β > 0 when the voter retains him in office. All together, the leader's total payoff is

(2)$$U_{\rm L} = -a_{\rm F}\gamma + W_{\rm L} + r\beta $$

The appointee, like the leader, has policy preferences which are either congruent (θ _A = 1) or incongruent (θ _A = 0), in a predictably hawkish (k = H) or dovish (k = D) direction. Specifically, at the appointment stage, the leader chooses $\pi _{\rm A}^k $, which denotes the direction (k ∈ {D, H}) and magnitude (π _A ∈ [0, 1]) of the agent's “bias.” After being appointed, the agent's type θ _A ∈ {0, 1} is realized and observed privately by A, with the commonly known probability Pr(θ _A = 1) = π _A. The agent's policy payoff is

$$W_{\rm A} = \theta _{\rm A}W_{\rm V} + ( {1 - \theta_{\rm A}} ) \left\{{\matrix{ {1-a, \;\quad k = {\rm D}} \cr {\,\,\,\,\,\,\,\;\;a, \;\quad k = {\rm H}} \cr } } \right\}$$

The agent observes the state ω, and offers advice to the leader in the form of a private, cheap-talk message expressing support for fighting (s = 1) or conceding (s = 0).

After the leader acts, the agent can either resign (z = 1) or remain in the administration (z = 0) (alternatively referred to as “protesting” or “remaining silent”). The agent has an “outside option” payoff of $ y \in { {\underline{y} , \;\bar{y}} } $, with $\underline{y} < 0 < \bar{y}$; this represents her value of leaving the administration relative to remaining, taking account of all factors other than policy concerns—income, prestige, ego rents, personal reputation, and so on. This value is observed privately by the appointee and the leader (after the appointment and before the leader's crisis response), while the other players hold a prior belief Pr(y = y) = λ (where λ, the appointee's “loyalty,” is chosen by the leader at the appointment stage). The policy value of remaining in the administration is f _A(θ), with f _A(1) > f _A(0) = 0, reflecting the idea that serving under a congruent leader provides the appointee with greater scope for influence in future policy decisions.⁶⁹ All together, the agent's total payoff is

(3)$$U_{\rm A} = W_{\rm A} + zy + ( {1-z} ) f_{\rm A} ( \theta ) $$

Finally, consider the election stage. On observing the leader's appointment decision $\alpha = ( {\pi_{\rm A}^k , \;\lambda } ) $, the foreign government's action a _F, the leader's action a, and the appointee's action z, the voter forms a posterior belief of the leader's quality, μ ^h = Pr(θ = 1|h) for history h. We assume that the voter wishes to re-elect the incumbent if and only if he is congruent.⁷⁰

Specifically, suppose the voter faces the choice between retaining the incumbent or replacing him with a domestic opponent O, where O has privately known congruence θ _O ∈ {0, 1}, with Pr(θ _O = 1) = π _O. V's payoff from retaining (r = 1) or replacing (r = 0) the leader is

$$U_{\rm V} = r\theta + ( {1 - r} ) ( {\theta_{\rm O} + \varepsilon } ) $$

where $\varepsilon $ captures V's preferences on all dimensions other than foreign policy, and $\varepsilon \sim U( {\underline{\varepsilon } , \;\bar{\varepsilon }} ) $. For simplicity, we assume that $\left[{\underline {\varepsilon}} , \;\bar{\varepsilon }\right] = \left[-\pi_{\rm O}, \;1-\pi_{\rm O} \right] $, which implies that any perfect Bayesian equilibrium to the game features V re-electing with probability μ ^h71; the probabilistic nature of re-election reflects the fact that the leader's (perceived) foreign policy congruence is only one of a wide array of considerations which could become more or less salient to voters come election time.

For reference, all parameters and variables are listed in Table A4 in the online appendix. The solution concept employed throughout is perfect Bayesian equilibrium (henceforth “equilibrium”). Technical assumptions regarding parameter restrictions and equilibrium refinements are discussed in Appendix 7.

Analysis

Before proceeding with the substantive analysis of the game, we should pin down some preliminary concepts we will build on going forward.

It will be useful to divide the game into three stages: the appointment stage; the crisis subgame (the subgame following F's action of a _F = 1); and the non-crisis subgame (the subgame following F's action of a _F = 0).

Behavior in the non-crisis subgame is straightforward, with a formal characterization given in the online appendix. Dove leaders and moderate Hawk leaders do not fight when unprovoked. Extreme Hawks may initiate unprovoked aggression, with a probability that is (weakly) decreasing in the strength of electoral incentives (and thus in the value of mimicking a moderate Hawk). In all cases, the leader's behavior in the absence of a crisis is unaffected by the appointment, so the bulk of the analysis focuses on the crisis subgame.

Within the crisis subgame, let us first consider the agent's strategy in providing private advice to the leader. Because the agent's advice takes the form of a cheap-talk message, many reporting strategies are available in principle. The most intuitive strategy, and the one we will focus on throughout the analysis, is as follows:

Definition 1 (Sincere reporting strategy)

We say that the agent plays a sincere reporting strategy when she provides advice to L that reflects her true policy preferences: that is,

$$s = \hat{\omega }_{\rm A}: = \left\{ {\matrix{ {\omega ,} \hfill & {\theta _{\rm A} = 1} \hfill \cr {1,} \hfill & {\theta _{\rm A} = 0\ \&\ k = {\rm H}} \hfill \cr {0,} \hfill & {\theta _{\rm A} = 0\ \&\ k = {\rm D}} \hfill \cr } } \right.$$

From the perspective of the other players, who know $\pi_{\rm A}^k $ but not θ _A, a sincere reporting strategy generates advice that satisfies

$$\left\{\!\matrix{{\rm Pr}( s = 1\vert \omega = 0) = 0 \hfill \cr {\rm Pr}( s = 1\vert \omega = 1) = \pi_{\rm A} \hfill} \!\right\}{\rm \;if\;}k = {\rm D}, \;{\rm \;and\;}\left\{\!\matrix{{\rm Pr}( s = 1\vert \omega = 0) = 1-\pi_{\rm A} \hfill \cr {\rm Pr}( s = 1\vert \omega = 1) = 1 \hfill} \!\right\}{\rm \;if\;}k = {\rm H}$$

We can now define the equilibrium of interest in the crisis subgame.

Definition 2 (Congruent-responsive equilibrium)

The CRE is the perfect Bayesian equilibrium to the crisis subgame in which the congruent leader attempts to match his action to the state: that is,

$$\sigma _1^{x, s} = {\rm Pr} ( a = 1\vert \theta = 1, \;x, \;s) = \left\{{\matrix{ {1, \;} & {{\rm Pr} ( \omega = 1\vert x, \;s) \ge \displaystyle{1 \over 2}} \cr {0, \;} & {\rm otherwise}\hfill \cr } } \right.$$

In Appendix 7, we introduce some technical assumptions regarding off-path beliefs and equilibrium selection. The most consequential among these is that we select the CRE with sincere reporting whenever it is available. Intuitively, this restriction focuses our attention on the equilibrium that maximizes information transmission between the advisor and leader, and that maximizes the policy payoffs for both congruent and incongruent leaders. With these assumptions imposed, we can state the following:⁷²

Proposition 1.

A CRE to the crisis subgame always exists. At the appointment stage, the leader always selects an appointee whose sincere advice can be followed in a CRE.

The remainder of the analysis of the crisis subgame focuses on equilibrium behavior within a CRE in which the agent reports sincerely and the congruent leader follows the agent's advice.

A core quantity of interest through the analysis is the appointee's influence.

Definition 3 (Appointee influence)

The appointee's influence on policy is the probability that L's crisis action with the appointee differs from what it would be without the appointee in place.⁷³

This definition allows us to quantify precisely the extent to which the appointment impacts the leader's foreign policy behavior, and how that impact varies as a function of appointee attributes and exogenous parameters.

Benchmark Model Without Domestic Politics

To build intuition for the main results, we begin by analyzing a simplified version of the model in which the leader's quality is common knowledge.⁷⁴ We will refer to this variant as the game without domestic politics: because the voter is forward looking, her electoral decision is fully determined by her knowledge of the leader's congruence θ and is thus unaffected by any action taken by the other players. This benchmark model effectively shuts down the fire-alarm mechanism and restricts attention to the advisory mechanism.

Result 1 (Game without domestic politics).

Consider a variant of the full model where L's type is common knowledge, taking the appointee's attributes as given.

• • F is less likely to challenge a congruent leader than an incongruent leader of either party.

• • When facing a congruent leader of either party, F's likelihood of challenging is decreasing in the appointee's hawkishness.

• • Given a challenge from F:

  • − The congruent leader follows the agent's advice, and the incongruent leader ignores it and plays his ideologically preferred action: that is, incongruent Hawks always fight, and incongruent Doves always concede.

  • − The appointee's influence is ${\pi \over 2}( {1-\pi_{\rm A}( {2\phi -1} ) } ) $, which is increasing in her own bias, decreasing in the leader's expertise, and limited to congruent leaders.

In the absence of any electoral considerations, each type of leader takes the action they prefer for policy reasons alone. For incongruent leaders, this simply means taking their ideologically consistent action: incongruent Hawks always fight, and incongruent Doves never fight. In the language of principal–agent models, the incongruent leader always “shirks” his responsibility to govern in the voter's interest,⁷⁵ as he has no incentive to do otherwise.

The preferences of congruent leaders are more complicated. They wish to fight if and only if it is in the voter's interest (that is, when ω = 1), but they face uncertainty as to the true state of the world. In navigating that uncertainty, a congruent leader can draw on two sources of information: his own noisy but unbiased signal x, and his agent's more accurate but potentially biased signal s. In the event that the two conflict, the leader will follow the agent's advice if her bias is bounded relative to the leader's own expertise—a condition that we define (in the appendix) as the agent's being informative. In equilibrium, the congruent leader always selects an informative appointee, so he always does better by following her advice once a crisis arises.

Given this crisis behavior by each leader, it is straightforward to quantify the appointee's influence over policy: the appointee can influence only the congruent leader's behavior, and only by providing him with advice that differs from whatever his private information would have led him to do otherwise. The probability of this happening is decreasing in the precision of the leader's private signal (that is, his expertise ϕ), and increasing in the appointee's bias.

The incongruent leader's unresponsiveness to the underlying state ω has important implications for deterrence. Let $\hat{a}( \alpha ) $ denote the probability that L fights back when challenged, given appointment $\alpha = ( {\pi_{\rm A}^k , \;\lambda } ) $; and let $\hat{a}_0$ denote the probability that L fights even when F does not initiate a challenge.⁷⁶ It follows directly from F's payoff function that they will challenge if and only if $\omega _{\rm F} > \hat{a}( \alpha ) -\hat{a}_0$, or in words, if

$$\hbox{F}{\unicode{x2019}}\hbox{s resolve} > \hbox{Pr ( L fight}\, \vert\, {\hbox{F challenge},\, \alpha ) -\hbox{Pr( L fight}} \,\vert\, \hbox{F not challenge)}.$$

The standard logic of deterrence relies on establishing, in the mind of the challenger, a link between the challenger's provocative action and an adverse consequence of that action. When the domestic leader is incongruent, that link is broken. In the absence of electoral incentives, incongruent leaders of either party are unresponsive to the policy value of fighting, and thus unresponsive to F's decision to initiate a challenge or not. (Formally, for incongruent leaders, $\hat{a}( \alpha ) = \hat{a}_0$ for all α.) This creates the weakest possible disincentive to F challenging.

In contrast, when facing a congruent leader, F understands that only their own aggression can provoke a hostile response from L. (That is, $\hat{a}_0 = 0$ for a congruent leader.) The higher the expectation that F's aggression will be retaliated against (that is, the higher is $\hat{a}( \alpha ) $), the greater the disincentive against F's challenging in the first place. Because the congruent leader follows his appointee's advice, he can credibly enhance deterrence by selecting a hawkishly biased appointee who shades her advice in favor of fighting when a crisis arises.

The model without domestic politics provides a useful benchmark against which to compare the full model, and to demonstrate how electoral incentives open up a new channel of appointee influence. It also clarifies an important point regarding the leader's incentives at the appointment stage of the full model: because revealing oneself to be incongruent at the appointment stage yields the worst deterrent against F's aggression, and because it yields the worst electoral prospects, an incongruent leader must fully mimic a congruent leader's appointment strategy. As a result, we may observe appointees serving under an incongruent leader, whom the latter would prefer not to have in office ex post but could not avoid appointing ex ante.

Full Model: How Appointees Influence Crisis Behavior

We can now expand the analysis to incorporate domestic politics. As originally stated in the model setup, we now assume that the leader's quality θ is his own private information, with all other players holding a common prior belief Pr(θ = 1) = π ∈ (0, 1). This section will analyze the “crisis subgame”—the subgame following F's decision to initiate a challenge—taking the appointee attributes as given. After analyzing the appointee's impact within the crisis subgame, we will examine the leader's optimal appointment strategy.

The following result summarizes the crisis subgame equilibrium, focusing on the case of the Dove leader; results for the Hawk leader are symmetrical (simply switching the preferred action of the incongruent leader).

Result 2 (Crisis subgame).

Consider the full model, with a Dove leader in office, taking the appointee's attributes as given. On the equilibrium path of play, given a challenge from F:

• • The congruent leader always follows the agent's advice.

• • When the agent advises fighting, the incongruent leader follows that advice with a probability that is (weakly) increasing in (i) the appointee's independence, and (ii) the strength of electoral incentives, per Equation (4).

• • The voter rewards the leader for fighting relative to conceding, and maximally punishes the leader when the appointee protests.

The appointee's influence is ${\pi \over 2}( {1-\pi_{\rm A}( {2\phi -1} ) } ) + ( {1-\pi } ) \sigma _{\rm A}\sigma _0^1 $, where $\sigma _0^1 $ is given by Equation (4) and $\sigma _{\rm A} = {\rm Pr} ( {s = 1} ) $ is given by Equation (5)—see the appendix.

• • Appointee influence is decreasing in the leader's expertise, (weakly) increasing in his electoral incentives, (weakly) increasing in the appointee's independence, and increasing in her hawkishness for $\pi _{\rm A}^{\rm H} \le 1$.

Results under a Hawk leader are symmetrical.

As in the benchmark model, the moderate leader remains incentivized to follow the agent's advice. The leader is aware of the agent's bias, and knows that she may be recommending action that the leader would not pursue if he had access to the same information the agent has; but the leader still determines that he is better off, on balance, following her advice rather than disregarding it, because the benefit of her expertise outweighs the cost of her bias. Thus the first term in the value of appointee influence, ${\pi \over 2}( {1-\pi_{\rm A}( {2\phi -1} ) } ) $, remains as it was in the benchmark model.

In the benchmark, the extreme leader had no incentive to take action contrary to his true policy preference. With the introduction of domestic politics, however, the extreme Dove confronts a trade-off: he prefers a conciliatory policy response for ideological reasons, but taking that action may reveal his incongruence to the voters and harm his re-election prospects. How he navigates this trade-off depends on the strength of his electoral incentives and the political loyalty of his appointee. Specifically, recall that the appointee faces an “outside option” value of $y\in { {\underline{y} , \;\bar{y}} } $ (where $\underline{y} < 0 < \bar{y}$), which she compares against her value of continued service f _A(θ) (where f _A(1) > f _A(0) = 0). Intuitively, the agent is more likely to remain in the administration when she believes the leader is more likely to be congruent, and thus receptive to her policy advice in the future.

When the agent recommends conceding, the extreme leader faces no trade-off: his preferred action is conveniently the one that will elicit no protest from his appointee. Likewise, when the agent recommends fighting but has a low outside option value (y = y), the extreme leader's decision is easy: he knows that if he takes a conciliatory action in the crisis, the appointee will disagree with that decision, but she will not make her disagreements known to an audience that can hold the leader accountable. However, when the agent's outside option is high—meaning she is willing to publicly protest the leader's decision—her recommendation to fight pits the leader's policy interests against his political concerns: he is ideologically inclined toward conceding, but doing so will invite politically damaging public criticism from within his own administration.

Why would voters punish the leader upon observing internal policy disagreements aired out publicly? Fundamentally, the voter's responsiveness to the appointee's protest depends on two factors: the alignment of interest between the voter and appointee with respect to the leader's quality; and the appointee's insider information. Appointees want to serve under congruent leaders, who are intrinsically motivated to follow their appointees’ expert advice; voters want to retain congruent leaders, who are intrinsically motivated to set policies that match the voters’ preference. The appointee's willingness to resign in protest, or to otherwise criticize the leader and risk her future standing in the administration, reflects her assessment that continued service in the administration is of little value—due to the leader's unwillingness to make use of expertise. Because this assessment is informed by the appointee's insider knowledge of internal policy deliberations, the voters rationally incorporate the appointee's protest into their own assessment of the leader's quality, and hence into their electoral decision.

Anticipating this political penalty as a result of public protest by the appointee, the incongruent leader must decide whether to endure that cost, or to instead forgo his preferred policy in order to maintain public approval. As we derive in the appendix, the best response to this dilemma gives rise to behavior which we can characterize as:⁷⁷

(4)$$\sigma _0^1 = {\rm Pr}( a = 1\vert \theta = 0, \;s = 1) = \max \left\{{0, \;{\mkern 1mu} \min \left\{{1-\lambda , \;\displaystyle{{\pi ( {\beta -1} ) } \over {1-\pi }}} \right\}} \right\}$$

Intuitively, the extreme Dove leader follows the agent's advice to fight with a probability that is (weakly) increasing in the agent's independence (1 − λ) and in the strength of his own electoral incentives (β): when the agent is more likely to make her disagreements known to the public, and when the leader is more sensitive to the costs of public disapproval, he is more likely to accommodate the agent's policy concerns so as to preempt any electoral punishment. If electoral incentives are sufficiently large (β > 1) and the appointee is not fully loyal (λ < 1), then the appointee's influence on policy is strictly greater than it was in the benchmark model without domestic politics (by the amount of $( {1-\pi } ) \sigma _{\rm A}\sigma _0^1 $, as stated in Result 2).

Implications for Deterrence and Voter Welfare

Having formally articulated the mechanisms through which appointees can influence leaders’ behavior, we can proceed to assess their impact on some substantively and normatively important political outcomes. We will consider three in particular: deterrence, policy responsiveness, and electoral selection. For convenience, we denote an appointment profile as $\alpha = ( {\pi_{\rm A}^k , \;\lambda } ) $.

Deterrence

Let $\hat{a}( \alpha ) $ denote, from F's perspective, the equilibrium probability of the domestic leader's fighting back when challenged, given appointment α; and likewise, let $\hat{a}_0$ denote the probability of the leader's initiating conflict unprovoked. Recall from the discussion surrounding Result 1 that F challenges if and only if their resolve ω _F is greater than $\hat{a}( \alpha ) -\hat{a}_0$, where $\hat{a}( \alpha ) $ varies as a function of appointee attributes but $\hat{a}_0$ does not. Thus we can say that an appointment improves deterrence if it increases $\hat{a}( \alpha ) $, which discourages F from initiating a challenge.

Responsiveness

An appointment improves policy responsiveness if it increases the probability that, conditional on a challenge from F, the leader's action matches the state of the world: that is, the appointment increases E[W _V|a _F = 1] = Pr(a = ω|a _F = 1). Responsiveness is undermined when the extreme leader shirks his responsibility to the voter, or when either leader seeks to serve the voter's interest but errs due to incorrect beliefs about the state.

Selection

An appointment improves electoral selection if it increases the difference between the re-election rates of congruent and incongruent leaders. We can measure electoral selection, conditional on appointment α and the emerge of a crisis, as

$$\Delta _r( \alpha ) = E\left[r\vert \theta = 1, \;a_{\rm F} = 1, \;\alpha \right] - E \left[r \vert \theta = 0, \; a_{\rm F} = 1, \;\alpha \right] $$

Intuitively, voters wish to retain congruent leaders and remove incongruent leaders, in the face of uncertainty over the leader's true quality; the electoral selection metric quantifies the extent to which they are able to achieve this goal.

Let us first consider the appointee's effect on deterrence:

Result 3 (Deterrence).

Under a Dove leader:

• • The likelihood of F challenging is decreasing in the appointee's hawkishness and (weakly) decreasing in her independence.

Under a Hawk leader:

• • The likelihood of F challenging is decreasing in the appointee's hawkishness and (weakly) increasing in her independence.

The basic logic of deterrence in the case of the Dove leader was outlined in the context of Results 1 and 2. As the appointee becomes more hawkish, she is more likely to induce the leader to fight via the advisory mechanism. The impact of her advice is further amplified when accompanied by the fire-alarm mechanism, which pressures extreme Doves to fight when they otherwise would not. Thus we can see that hawkishness and independence have complementary effects on deterrence.

The advisory mechanism operates in equal measure under a Hawk leader, likewise serving to enhance deterrence. The fire-alarm mechanism, however, works in the opposite direction. Under a Dove leader, the moral hazard problem between the voter and incongruent leader implies that the leader fights less often than the voter would want to, if the voter were in the leader's position; the threat of the fire alarm disciplines the leader into governing more in line with the voter's interest, by fighting more frequently. Under a Hawk leader, in contrast, the moral hazard problem implies an excessive combativeness on the global stage—entering into conflicts where the voter would prefer to stay out, if fully apprised of the expected costs and benefits. Mitigating this form of moral hazard, though desirable for the voters ex post (after a crisis has arisen), has the unfortunate effect of undermining deterrence ex ante.

While some appointments can enhance deterrence against foreign aggression, there is inevitably an offsetting cost.

Result 4 (Welfare implications).

Consider a fully loyal and unbiased appointee, π _A = λ = 1. Any marginal change in appointee bias or loyalty that improves deterrence must weaken either policy responsiveness, or electoral selection, or both.

The result is fairly straightforward when considering the effect of appointee bias on policy responsiveness. For the agent's hawkishness to improve deterrence, it must be the case that she induces the leader to fight in some instances where he (and the voter) would prefer not to, were they themselves privy to the same information that the agent observed. This distortionary effect on policy is core to the deterrent logic of the advisory mechanism.

The effects of appointee bias on deterrence and responsiveness under a Dove leader are depicted in the top panel of Figure 2.⁷⁸ Both the congruent and incongruent leaders’ fighting probabilities are increasing linearly in the appointee's hawkishness, with the incongruent leader's strategy responding at a slower rate (given 0 < λ < 1, as is the case in the figure). Deterrence is simply a weighted average of the congruent and incongruent leader's fighting probabilities, reflecting the foreign challenger's uncertainty as to which type of domestic leader they are facing. Policy responsiveness peaks in the center of the figure, when the agent's advice is perfectly unbiased (π _A = 1), and decreases with the agent's bias in either direction.

FIGURE 2. Effects of appointee bias, under a Dove leader

The effect of appointee hawkishness on electoral selection is more subtle, operating through both a direct effect on leader behavior and an indirect effect on voters’ beliefs. The lower panel of Figure 2 depicts these effects in the case of a Dove leader. As the appointee becomes more hawkish, this leads the moderate leader to fight more frequently, creating greater separation in the behavior of the two leader types (increasing Pr(a = 1|θ = 1) − Pr(a = 1|θ = 0)). This in turn leads voters to assess the leader more negatively when they see him concede in the crisis (decreasing Pr(θ = 1|a = 0)). These effects combine to make voters better able to distinguish moderate Doves from extreme Doves, thus improving electoral selection.

Under a Hawk leader, however, appointee hawkishness induces convergence rather than separation in the behavior of extremists and moderates. The only way for Hawk leaders to improve deterrence is by appointing a hawkishly biased agent, but doing so undermines both policy responsiveness and electoral selection.

In addition to appointee hawkishness, the Dove leader can improve deterrence through appointee independence, which activates the fire-alarm mechanism. As discussed earlier, this mechanism improves deterrence through its effect on the incongruent leader's behavior; and by pressuring the incongruent leader to fight more often than he otherwise would, the appointee improves policy responsiveness along with deterrence (solid line in Figure 3).⁷⁹ However, as the incongruent leader is induced to behave more like the congruent leader, the voter becomes less able to differentiate the two. Electoral selection is weakened as a consequence (dashed line).⁸⁰

FIGURE 3. Effects of appointee independence, under a Dove leader

Leader's Optimal Appointment Strategy

Having examined the various effects that appointees can have on foreign policy and domestic politics once in office, as summarized in Table 1, we turn to the question of which appointee the leader will optimally select. Two preliminary points are worth noting. First, as mentioned, the incongruent leaders of both parties are incentivized to fully mimic the appointment strategy of their congruent counterparts. Second, each outcome discussed in the preceding section—deterrence, policy responsiveness, and electoral selection—corresponds to a goal that the congruent leader wishes to advance through his appointment strategy (the terms −a _Fγ, W _L, and rβ, respectively, in the leader's payoff function, Equation (2)). Thus the equilibrium appointment will be the one that best serves the congruent leader's interests, taking account of the trade-offs among these three objectives.

TABLE 1. Effects of appointee hawkishness and independence

Note: All changes are relative to an unbiased and fully loyal appointee (π _A = λ = 1).

We have the following result with respect to appointee bias.

Result 5 (Biased appointments)

• • Leaders of either party will never appoint a dovishly biased agent.

• • Leaders of either party will appoint a hawkishly biased agent if the value of deterrence is high.

• • More experienced leaders are less likely to appoint biased agents.

The first point is straightforward: dovishly biased agents clearly offer no benefit to a Dove leader, and while they may benefit Hawk leaders electorally, that benefit is always outweighed by the costs of distorting policy and weakening deterrence. In contrast, hawkishly biased agents can prove optimal for a leader of either party: though a hawkish agent distorts policy responsiveness, and may (for a Hawk leader) harm re-election prospects, a leader of either party is willing to incur these costs if they place sufficient value on deterring foreign aggression.

We also uncover a more nuanced result regarding leader experience (insofar as experience is reflected in the leader's “expertise,”⁸¹ or the accuracy of the leader's private signal ϕ). For the appointee to influence policy through the advisory mechanism, it must be credible for the leader to follow advice that he knows to be biased; this requires that the cost of the bias be offset by the benefit of the agent's expertise—and in particular, the difference between the agent's expertise and the leader's own. More experienced leaders are less able to credibly commit to being influenced by biased advice, and thus have less use for biased appointments.

Turning to appointee independence, we find the following.⁸²

Result 6 (Independent appointments)

• • A Dove leader will appoint an independent agent if the value of deterrence is high.

• • Leaders of either party will appoint an independent agent if and only if electoral incentives are low.

• • A Hawk leader may appoint an independent agent, even when doing so will undermine deterrence; but he is less likely than a Dove leader to appoint an independent agent (under otherwise symmetrical conditions⁸³).

While Dove leaders may optimally appoint independent agents under a broad range of conditions, their reason for doing so will depend on the strength of electoral incentives, β. Counterintuitively, Dove leaders select independent appointees for their electoral benefits precisely when electoral incentives are low. When β < 1, an independent appointee will not affect the incongruent leader's behavior, and therefore cannot improve deterrence; but she can help the congruent leader's re-election prospects, as her decision not to protest can provide “political cover”—validating actions by the leader which would otherwise be viewed unfavorably by the voter. Conversely, when electoral incentives are high, the motive for selecting an independent agent is purely its deterrent value: when $\beta > {1 \over \pi }$, the appointee's independence induces the extreme leader to fully mimic a moderate, thus improving deterrence but undercutting the moderate's electoral advantage as a consequence. It is only under a fairly narrow range of conditions that an independent appointee can advance both electoral and deterrent objectives simultaneously.⁸⁴

The effect of electoral incentives on the Hawk's appointment strategy follows a similar logic. Independent appointees are always desirable when they don't affect the incongruent leader's behavior. As electoral incentives increase, and the threat of appointee protest starts to induce moderation in the extreme Hawk's behavior, appointee independence becomes less valuable for the congruent Hawk. Somewhat surprisingly, this logic implies that congruent Hawks may still be willing to suffer some diminution of deterrence in exchange for the electoral benefits an independent appointee provides. Yet ultimately we do find that Doves are more willing than Hawks to appoint independent agents, and this divergence in appointment preferences grows as the value of deterrence increases.

Empirical Implications: Partisan Asymmetry in Appointments

To assess the empirical plausibility of the theory's implications, we can consider some descriptive patterns of partisan appointments in top-level foreign policy positions.

The incentives and institutional arrangements assumed in the theory are most directly modeled after the postwar US presidency, and this appears to be a context in which the theory provides considerable empirical purchase. Table 2 presents a striking pattern: since the creation of the office of the US secretary of defense in 1947, Democratic presidents have filled the position with a Republican appointee for nearly half of their thirty-seven years in office. No Democrat has been appointed to the position in forty years of Republican administration. “Partisan” Democrats—those who previously held elected office or worked in party politics—served in the role for only four of Democrats’ thirty-seven years; the analogous figure for partisan Republicans is twenty-three of forty years.

TABLE 2. US defense secretary appointments by party, 1947–2024

Note: See Table A5 in the online appendix for details.

The model presented here provides a framework to help make sense of this asymmetry. Taking Democrats and Republicans to represent the Dove and Hawk parties, respectively, a central implication of the model is that neither Democrats nor Republicans will ever appoint dovishly biased agents to high-level national security positions; that both have an incentive to appoint hawkishly biased agents; and that Democrats are more likely than Republicans to prefer appointees with weak political loyalties to themselves. Each of these tendencies is borne out in the data. The appendix provides further discussion of several cases of co-partisan appointments that also seem to fit the logic of the model—Democratic defense secretaries who were perceived as more hawkish than their co-partisan presidents—as well as other high-profile appointments to different foreign policy positions.

To examine whether these patterns also appear beyond the context of the United States, we turn to two cross-national data sets: WhoGov,⁸⁵ which records the portfolios and party affiliations of the global sample of leaders and cabinet members at an annual frequency; and the Manifesto Project,⁸⁶ which codes time-varying party positions on a wide range of policy issues for all major parties in over sixty countries. Linking these data sources, we can assign measures of hawkishness or dovishness to leaders and cabinet ministers based on their party affiliations.

I start by constructing an index of party-election-level hawkishness from four individual measures coded in the Manifesto data: “per101: Foreign special relationships: positive” plus “per104: Military: positive” minus “per105: Military: negative” minus “per106: Peace.” For commensurability with the theoretical model, we want to consider these hawkishness values in relative terms, in the particular political context the parties operate within. Thus, within each election, I arrange parties by their hawkishness values; label the vote-share-weighted median party a Centrist party; and label parties above and below the median Hawk and Dove parties, respectively.⁸⁷ Individual leaders and cabinet ministers are then assigned their parties’ Hawk/Dove/Centrist coding from the most recent election (up to ten years in the past). Full details on the coding procedure are presented in Appendix 10.2.

With these codings, the first two columns of Table 3 break down appointments of defense ministers by the leader's party image. (Similar analyses of ministers of foreign affairs are reported in Table A7 in the appendix.) The first two rows show, unsurprisingly, that Hawk leaders are more likely to appoint Hawk ministers than Dove ministers, and vice-versa for Dove leaders. But we also see an important asymmetry across parties: Dove leaders are more than twice as likely to appoint Hawk defense ministers than the reverse (24 percent versus 9 percent); more than three times as likely as Hawk leaders to appoint independent, non-partisan defense ministers (19 percent versus 6 percent); and substantially less likely to appoint co-partisans (41 percent versus 64 percent). These findings are consistent with the model's core predictions that leaders of either party may seek to enhance deterrence by soliciting biased advice from a hawkish advisor; but Dove leaders see greater benefit in selecting politically independent appointees to serve an internal fire-alarm function, enhancing their own accountability to their electorate.

TABLE 3. Partisanship of leaders and ministers of defense

Notes: Country-year observations, across 58 countries from 1963 to 2021. Centrist leaders omitted, but reported in Table A6 in the appendix. Numbers denote the percentage of a given appointment type within a column. For example, the minister of defense is from a Hawk party in 24% of all country-years with a Dove leader (and 20% of country-years with a Dove leader with an upcoming election).

The next four columns consider temporal heterogeneity in appointment strategies. Result 6 stated that leaders of either party will appoint politically independent agents if and only if electoral incentives are sufficiently small. One intuitive proxy for the strength of electoral incentives is the proximity of an upcoming election in which the incumbent leader is eligible for re-election. Columns 3 and 4 split the sample of Dove leaders on the basis of whether the leader is facing re-election within the next two years, while columns 5 and 6 do the same for Hawk leaders. A consistent pattern emerges: leaders are more likely to select co-partisan appointees, and less likely to select independent or cross-partisan appointees, when re-election concerns are more salient. The logic underlying this pattern, according to my theory, is not that leaders are afraid of independent agents revealing unfavorable information to the public in the run-up to an election; rather, it is that leaders anticipate that they will not receive full “credit” in the eyes of the voter for their moderate policy choices, if voters perceive that those choices were strongly influenced by the presence of an independent appointee. When surrounded by loyalists, a leader's governing behavior provides a clearer signal of his true policy preferences.

These cross-national descriptive patterns are largely consistent with the model's implications. Further research is needed, however, to determine the extent to which the theoretical mechanisms posited in the model are operative across the diverse contexts represented in the sample. In particular, it should be recognized that portfolio allocations in parliamentary coalition governments are the result of bargaining processes that are far more complex than the unilateral appointment decision modeled here—and conversely, that parliamentary cabinets are characterized by a more straightforward and credible delegation of policy authority as compared to presidential appointments.⁸⁸

The present theory may still help explain why portfolio allocations with hawkish parties in charge of defense ministries would provide an efficient solution to the coalition government formation problem, and it can provide insights into how such allocations affect foreign policy and domestic politics. Notably, the qualitative patterns that emerge in the full sample (Table 3) are also observed when disaggregated by presidential and parliamentary systems (Table A8 in the appendix.)

Discussion

This study set out to explain how leaders select their top-level foreign policy appointees. It developed a theoretical model that incorporates the most pressing considerations that factor into the leader's decision—the quality of policy advice he receives, the voter's assessment of his foreign policy performance, and the impacts on foreign actors’ behavior—to understand how the leader evaluates trade-offs in his appointment strategy, and it has provided cross-national evidence of appointment patterns consistent with the model's predictions.

Like any theoretical model, the present analysis invokes some simplifying assumptions; future work should seek to revise and generalize some of these. Most notable, perhaps, is the present model's focus on a single appointee; in reality, of course, leaders appoint and consult with larger teams of foreign policy advisors, and the dynamics and disagreements among these multiple advisors have been a subject of sustained theoretical interest.⁸⁹ We might also question the model's focus on an international game of deterrence. While deterrence has long been a central pillar of US national security strategy, many foreign policy issues are better represented by other basic games at the international level (“prisoner's dilemma” or “stag hunt,” for example, rather than “chicken”).⁹⁰ Future work can consider how these different international games might map onto the portfolios of different foreign policy positions, and how that heterogeneity can explain leaders’ appointment strategies across those different positions.

Finally, we can consider other implications of the model that future work can seek to test empirically. On the one hand, the model's implications regarding citizen welfare (that is, appointees’ impacts on policy responsiveness and electoral selection) involve theoretical quantities which, though normatively and politically important, are fundamentally unobservable and thus untestable. The implications regarding deterrence would seem to lend themselves more naturally to systematic empirical evaluation: Result 3 provides straightforward predictions relating appointee bias and independence to deterrence against foreign aggression, which could be tested using conventional conflict data sets. On further consideration, however, some major difficulties arise with this approach. Results 5 and 6 tell us that the likelihood of a leader's selecting a deterrence-enhancing (or deterrence-undermining) appointee is a direct function of the cost they suffer from deterrence failure. If, for instance, a leader's (unobservable) perception of a higher international threat environment affects both the leader's choice of a hawkish appointee and the likelihood of a challenge, then any observed correlation (either in the same or the opposite direction as predicted in Result 3) could be entirely spurious. Future work might seek to develop research designs that leverage domestically originating sources of variation in appointments to identify their effects on foreign policy outcomes.

An alternative approach for finding evidence of appointees’ impacts on deterrence could involve qualitative examination of the internal deliberations and decision-making processes of foreign governments, and whether and how they incorporate considerations of an appointee's attributes and influence into their assessment of her government's behavior. Some preliminary observations suggest that such an approach could be fruitful. Archival records and journalistic accounts, for instance, indicate that foreign governments are keenly attuned to the particular experiences, priorities, capabilities, and ideological leanings of individual high-level appointees in the United States; that they have fairly sophisticated understandings of the internal workings of the US foreign policy process, and the points of conflict between and within the different agencies involved; and that they form expectations of future US foreign policy behavior on the basis of that knowledge.⁹¹ Finding evidence linking these assessments to the success or failure of deterrence is an important task for future research.