A game of domestic imperatives: Audience costs and conflict avoidance

Abstract

I propose an audience costs game with considerations added from selectorate theory. We see that winning coalition and selectorate size have competing effects on conflict choices in an audience costs setting. Large coalition regimes face lower audience costs than non-democracies, making it harder for them to commit to war. But larger selectorates increase the value of office, making conflict escalation more attractive. Coalition effects dominate when interacted with selectorate size. Evidence from 1816–2014 supports the game's implications. The results indicate that both threat initiation and dispute resolution are better predicted by focusing on domestic, leader-specific variables.

Keywords

Audience costs selectorate theory democracy autocracy threats

It takes two to tango and at least two to engage in a conflict or wage war. Hence, war is always a strategic problem. Although not always understood in this way, Thomas Schelling (1960) gave us a first glance at how to think about conflict strategically. Schelling's insights were expanded upon by the pioneering research of Frank Zagare and his mentor, Steven Brams, who helped to establish a firm foothold for studying conflict—and its avoidance– game theoretically. Much of the early research focused on using game theory to explain specific events but in a series of conference presentations, many later published in the journal literature, Brams and Zagare began to explore general principles of conflict, as in their 1981 publication, first presented at Public Choice in 1979, “Double Deception” (Brams and Zagare, 1981). A short while later, Zagare (1983) became one of the earliest contributors to the strategic assessment of deterrence. These early efforts led to a great expansion in the use of game theory to think rigorously about disputes (see, for instance, Niou et al., 1989; Powell, 1990; Wittman, 1979).

Deterrence is understood largely in terms of the ability to incentivize an adversary not to do something that the adversary otherwise would do; that is, to put undesired actions off the equilibrium path. This is widely understood to be achieved by raising costs or raising risks faced by one's adversary. In this analysis, the costs and risks that might alter the probability of war or lesser conflict are considered in their domestic political setting in which government leaders are focused on how foreign policy choices might jeopardize their domestic political survival. As Geoffrey Chaucer wisely observed in “The Tale of Melibee”: “[S]ince there is no man certain if he be worthy that God give him victory …, therefore every man should greatly dread to begin wars.”¹ Chaucer's comment reminds us that while it takes two to make a fight, one can greatly diminish the risk of conflict by making choices to avoid confrontation in the hopes of preventing power from slipping away to a domestic, as well as to a foreign, rival. Put a little more precisely, since it takes two to wage war, a necessary condition for war is that each side must be willing to fight. Hence, if we can identify domestic political conditions that make a government unwilling to fight, then we can identify conditions that preclude war. With that in mind, I explore empirical evidence in a preliminary way based on the game theoretic implications regarding domestic threat avoidance with a simple audience costs game that combines features of selectorate theory and features of the audience costs models by Fearon (1994) and Smith (1998). The evidence will show that a small set of domestic factors are strong discriminators between those who are prepared to wage war and those who are not.

A war or peace game

Figure 1 depicts an “audience cost” view of decisions to engage in a crisis that might culminate in war and that might culminate in leader deposition due to domestic incompetence. The game builds on, and somewhat modifies, the structure of the audience costs games due to Fearon and to Smith. In Fearon's model, audience costs are exogenous and accumulate over the limited life-span of a crisis. In Smith's model, leaders are more directly accountable to constituents, thereby endogenizing audience costs, and these costs are seen as rising with the leader's policy incompetence. Unlike in Fearon's approach, Smith's model includes the possibility that the domestic audience might prefer that their leader back down in a dispute that the leader initiated.

Figure 1.

A simplified audience costs game. 0 ≤ m ≤ 1: m = domestic policy incompetence (capturing the risk of not being reselected due to domestic incompetence). 0 ≤ Ψ ≤ 1: Ψ = audience cost, cost for leader if turned out of office. ka, kb: A and B's respective transaction costs from war. 0 < X < 1: X = value of the status quo internationally for player A (1-X for B). p = probability that A wins if A attacks B (1-p is B's probability of winning if A attacks B). Assumptions: Winning a war or making a successful threat (B Accepts) ensures remaining in office; Losing a war ensures losing office; Backing down after threatening or not threatening implies that the risk of losing office depends only on the leader's domestic record.

In the model proposed here, audience costs are represented as the incumbent's multidimensional expected cost from losing office ( $ψ$ ) which occurs for sure if the incumbent loses a war and occurs probabilistically, depending on the incumbent's past domestic performance (m) in the absence of a war (Bueno de Mesquita and Siverson, 1995; Chiozza and Goemans, 2004). The expected political cost from retrospective domestic policy incompetence is denoted by the product m $ψ$ , meaning the discounted value of losing office conditional on the incumbent's prior domestic policy record of competence (so that the smaller m is, the more competent the incumbent, and hence the lower the expected domestic cost).

The game is designed to capture and expand upon insights from the earlier literature while providing a formalization that is appropriate for use in undergraduate instruction. The model provides new insights into the expected impact of audience costs as a function of regime type. While crisis decision-making is generally analyzed as if choices are taken by states, here, in the spirit of the assumption that politics is primarily about what leaders can do to survive in power, I view player A as the leader of state A and player B as the leader of state B. This will matter when it comes to estimating expected payoffs and the factors that go into shaping leader A's decision to engage in a dispute with the prospect of war or to live with the status quo situation.

An important part of leader A's problem is whether she will survive politically to enjoy the benefits of office or will be deposed either because her domestic performance has been poor (m is large) or because failure in a crisis stimulates her constituents to remove her from office, so that she loses the value of office ( $ψ$ ) for sure following defeat in a war. Following Zagare and Marc Kilgour's (2000) precedent in Perfect Deterrence, the game is solved for subgame perfect Nash equilibrium.

Payoffs at each terminal node are structured to reflect the expected costs, benefits and risks under each possible sequence of actions. If A does not initiate a threat, then the game is assumed to end with the current international status quo prevailing. In that case, A derives her utility from the status quo (X) minus the expected utility of losing office because of past domestic policy failures (m $ψ$ ). To keep things simple, I do not model the domestic circumstances for B so B's payoff from the absence of a threat is simply 1 − X; that is, B's utility for the international status quo.

If A initiates a crisis, that means it threatens B, demanding that B change some policy or policies to be more aligned with A's interests. If A threatens B, then B can accept whatever A has demanded or B can make a counter threat, demanding some change in A's policies. If B accepts whatever demand A made along with A's threat to B, then A gets a utility set equal to 1, meaning A gets whatever she asked for, and B get a utility of 0, meaning B has given A what A demanded. The key feature of these payoffs is that if A threatens and B accepts then A is better off than she was under the status quo.² If B counters A's threat, then A must decide to carry out her threat by attacking B (so that a war occurs) or A can choose to back down, giving B whatever he demanded. In that case A gets 0 on the international front in her dealings with B and might, as well, be ousted from power owing to her prior record of domestic policy performance.³ The key here is that if A can anticipate that B will counter A's threat and that she would then choose to back down rather than fight B, then A is better off not initiating a threat to begin with since $X > 0$ . Thus, the decision to back down is off-the-equilibrium-path so that if it arises in the threat subgame, then the subgame perfect Nash equilibrium of the game is for A not to make a threat. Indeed, near the end of this study, when we investigate the probability that A makes a threat, we will see that the expectation about whether B will counter A's threat strongly influences the likelihood that a threat is or is not made.

The payoff structure has been chosen to emphasize the domestic political consequences of making threats and then choosing to carry them out (waging war) or backing down from the threat as this is a central concern in the audience costs literature. Thus, if A attacks then she faces a risky war situation. If A wins, then A retains office and extracts the policy concessions she sought from B but if A loses then A is ousted from office, having conceded to B whatever B's policy demands were. While the certainty of ouster following defeat is a strong condition, the empirical record supports the contention that the probability of leader deposition is, in fact, much higher given defeat in war than it is given victory (Bueno de Mesquita and Siverson, 1995; Chiozza and Goemans, 2004). Of course these payoffs could be readily modified to reflect more limited conditions but these payoffs focus our attention on the key features behind the audience costs literature, namely the risks and costs of making demands and then carrying them out or backing down from them.

Solving the game

Taking the game's sequence of moves into account, we see that A's possible strategies are Threaten, Attack; Threaten, Back Down; Not Threaten, Attack; and Not Threaten, Back Down. B's possible strategies are Counter or Back Down. The game has several possible subgame perfect Nash equilibria that depend, of course, on the precise payoffs. My interest here is limited to an empirical assessment of domestic variables that shape audience costs and their comparative static implications for war and the credibility of threats that follow from this model.⁴ Hence, the empirical investigation focuses on the likelihood of war given that a threat has been made and then on whether a threat in fact is made given the factors that influence the likelihood of war. Before turning to the comparative static results pertaining to the probability of war, however, we should establish the conditions under which war and other outcomes are the subgame perfect equilibrium of the game.

The audience costs game in Figure 1 has Subgame Perfect Nash equilibria that support the Status Quo (No Threat), B's acquiescence to A's threat (Accept) and an attack by A against B (War):

War conditions: war occurs in equilibrium when $p - X - k_{A} \geq ψ (1 - p - m)$ and $1 - p - k_{B} \geq 0$ . It is useful to write these condition in terms of the probability of victory:

\hat{p_{B}} = 1 - k_{B} \geq p \geq \frac{ψ (1 - m) + X + k_{A}}{1 + ψ} = \hat{p_{A}}

(1)

p > \hat{p_{A}} = \frac{ψ (1 - m) + X + k_{A}}{1 + ψ}

and

p \leq \hat{p_{B}} = 1 - k_{B}

. The first condition ensures that A will choose to initiate the crisis over maintaining the status quo (and ensures that A will fight in the final stage) and the second condition ensures that B will choose to resist by countering rather than accepting A's demand. Hence a necessary condition for war is that :

1 - k_{A} - k_{B} - X \geq ψ (k_{B} - m)

(2)

No threat (status quo) conditions: the status quo is chosen by A over backing down if A threatens and B counters. Because $X > 0$ , it follows that $X - m ψ > - m ψ$ . The status quo is chosen over war if A has threatened and B has countered provided that $X \geq p - k_{A} - ψ (1 - p - m)$ .

B Backs Down Conditions: B does not choose to counter A's threat if $0 > 1 - p - k_{B}$ . Hence, if B will back down then A threatens since $1 > X - m ψ$ .

In the absence of domestic considerations, that is, in a unitary state perspective, war is possible if

1 - k_{A} - k_{B} - X \geq 0

. Under this condition domestic politics are irrelevant. It therefore makes sense to focus on conditions when, without domestic consideration, war is unlikely,

1 - k_{A} - k_{B} - x < 0

. This constitutes the hard case when domestic politics will be influential in determining whether war occurs or not. Given this constraint, war requires that

m > k_{B}

and the condition in equation (2) becomes easier to satisfy as m and

ψ

increase; that is, A's domestic policies have been unsuccessful and the leader in state A faces a high loss in value if she is removed from office.

The analysis begins when A has made a threat and B has countered it. Naturally, if B is expected to back down when threatened, then A will threaten for sure ( $1 > X - m ψ$ ). If B is expected to counter, then whether A threatens or not depends on A's preparedness to fight if B has countered. Hence we first explore what happens when B has countered or A expects B to counter.

As the payoffs indicate, if a crisis escalates to the stage at which the leader in A must decide whether to go to war (Attack) or back down, the leader can anticipate that if she chooses war and loses, she will pay a cost imposed by her key constituents; she will be turned out of office by them. If she wins the war, then she is assumed to retain office and to pay no other, lesser, foreign policy audience cost, in keeping with a considerable body of empirical evidence (Bueno de Mesquita and Siverson, 1995; Chiozza and Goemans, 2004).

Of course, war is always a risky choice. The payoff structure indicates that in the event that the leader of A chooses to back down rather than fight, then she pays a cost that is the product of being turned out of office for domestic incompetence and the failure to back up a threat of war once the threat has been made. The idea is that a leader who is sufficiently successful on the domestic policy front pays a smaller price from backing down than one who also has failed on the domestic policy front, in keeping with Smith's (1998) audience costs model. In equilibrium, as the audience cost literature notes, if A expects to back down after making a threat, then A does not threaten to begin with; backing down is an off-the-equilibrium path expectation (recall that X − m $ψ$ $>$ −m $ψ$ given that $X > 0$ ).

When it comes to a choice between backing down or fighting, the decision depends on comparing the expected payoffs under the alternative actions. Of course, if A is in a crisis, she will fight if the expected payoff from fighting is greater than the payoff from backing down and will back down (or not threaten in the first place) otherwise. Obviously, as domestic incompetence (m) increases, backing down becomes less attractive and fighting becomes more attractive as long as $ψ$ , the value A attaches to office, is not 0.

As the value of office, $ψ$ , increases, it alters the attractiveness both of fighting and of backing down. As we know, the game's payoffs indicate that when the choice is to back down or to fight, fighting is chosen over backing down provided that $p > \hat{p_{A}} = \frac{ψ (1 - m) + X + k_{A}}{1 + ψ}$ . Then we want to know whether it is harder or easier to satisfy this threshold condition ( $\hat{p_{A}}$ ) as $ψ$ , the cost the audience can inflict, increases. We see that $\frac{d \hat{p_{A}}}{d ψ} = \frac{1 - m - k}{{(1 + ψ)}^{2}}$ . From this result we realize that as $ψ$ —the utility lost if A is removed from office—increases, the probability of victory that is required to choose to fight rather than back down ( $\hat{p_{A}}$ ) decreases, meaning A will fight with a lower chance of victory; that is, a lower value of p, as $ψ$ increases, making war an easier choice, provided that $m + k < 1$ . Conversely, if $m + k > 1$ then as the cost of losing office through defeat in war increases, the threshold value of p ( $\hat{p}$ ) at which a government chooses to fight rather than back down increases, making it harder to meet the conditions for war.

If we think of economic contraction on a per capita basis as an indicator of a leader's domestic incompetence, then we see that, all else being equal (which it never is), the more a leader's economy contracts ( $m \to 1$ ) the more likely the ruler is to gamble on war. As we know, small coalition leaders (autocrats, monarchs, and junta leaders) are particularly likely to face these conditions (Bueno de Mesquita et al., 2003: chapter 4) and, therefore, are expected to wage war while large coalition, democratic leaders are expected to be less likely to choose war in a crisis situation and, as well, are expected to be less likely to initiate a crisis than non-democrats.

Regime type, audience costs and conflict escalation

The conventional view of audience costs assumes that democracies generate higher audience costs than non-democracies. Hence, democracies are presumed to be less likely to back down, making their threats more credible. Fearon (1994: 577), for instance, states that, “the side with a stronger domestic audience (e.g. a democracy) is always less likely to back down than the side less able to generate audience costs (a nondemocracy).” Thus, in this view, when democracies make threats, rivals are more likely to give the democracy what it has demanded and, if they do not accept a democracy's demands then they are more likely to face war than is true when the party making a threat is non-democratic. As we explore this condition it is important to recognize that the assertion that democracies face higher audience costs than non-democracies is not a deduced consequence of the canonical audience cost model; it is rather a rhetorical claim. As we will see, selectorate logic leads us to question that rhetorical claim and to add nuance to its expectations by distinguishing between the expectation that audience costs are generated and the magnitude of those costs.

Ease of generating audience costs $\neq$ audience costs

The standard account implicitly assumes that the more dissatisfied the audience, the higher the cost borne by the incumbent but, of course, it is possible that the incumbent's assessment of the value of office is low even as the audience credibly threatens to remove the leader from her position. Equating the leader's constituent's ease of generating audience costs with the magnitude of the costs suffered by leaders if they lose office can result in analytical and empirical confusion. Jessica Weeks (2008), for instance, noting this, has raised important challenges to the assumption that it is easier for democracies to generate audience costs than it is for non-democracies. In doing so, she builds on the coalitional logic of the selectorate theory. Weeks reports that the evidence does not bear out the contention that democracies more easily create audience costs than non-democracies. Weeks, however, does not draw out the distinction between generating audience costs and the magnitude of those costs. Why generating audience costs and the magnitude of such costs may be inversely related becomes clear once we probe the connections between a selectorate view of politics with an audience costs perspective.

Confusion between the generation of audience costs and the expected magnitude of those costs arises, in part, because the audience costs literature tends to treat regime type categorically: regimes are democratic or non-democratic. That perspective does not distinguish between the value that office creates for a leader and the ease with which a leader is removed from office. Certainly there is no dispute over the assertion that leaders are more easily removed from office in a democracy than in an autocracy, monarchy, or military junta. But selectorate theory indicates that the value of holding power is strictly decreasing as the coalition on which a leader depends grows larger (Bueno de Mesquita et al., 2003; Bueno de Mesquita and Smith, 2017). That means that it is possible that democratic audiences more easily generate audience costs than non-democratic audiences but that the magnitude of audience costs—the cost of losing office—in democracies is smaller than in non-democracies, making the loss of office more tolerable in large coalition, democratic polities than in smaller coalition regimes or even within categorically democratic governments as the coalition's size increases.

We can think of the value of office, separate from the risk of losing it, on three dimensions. First there is the intrinsic value that any leader attaches to holding power. There seems to be no basis for assuming that the intrinsic value of office is greater for one type of leader, say a democrat, than for another type of leader, say an autocrat. Indeed, it is evident that all types of leaders, even hereditary monarchs, as demonstrated by Sharma (2015), face competition for the right to hold office. This first, intrinsic dimension is probably a characteristic of each individual competing for power rather than a systematic function of regime type. And, unfortunately, it is also unlikely to be measurable. Given that we observe fierce competition for office in all forms of government, a reasonable starting place regarding the intrinsic value of being in power is that it is indistinguishable across regime types.

The second dimension of value in office holding follows from the extent to which an incumbent has authority over government spending decisions and, particularly, the discretion to direct resources to her own uses. Selectorate logic indicates that for coalition members and for subjects not in the coalition, increasing coalition size generally improves their utility from government spending.⁵ In contrast, the value of office diminishes continuously for any leader as coalition size increases while the risk of losing office increases if one governs in a democracy rather than a non-democracy as assumed by Fearon or, in the selectorate context, as the ratio of coalition size to selectorate size (W/S) increases.

That the value of holding office decreases as a government becomes more dependent on a large coalition ( $W \to 1$ ) is most readily understood by thinking about the proportion of government revenue that is available to be spent at an incumbent's discretion after the incumbent has paid her coalition enough so that even the least satisfied member remains loyal to the incumbent rather than defecting to a rival. Selectorate logic establishes that the amount paid to secure coalition loyalty—the incumbency constraint—increases as the coalition (W) increases so that the larger a coalition is the proportionately less discretionary revenue at the incumbent's disposal.⁶ If we think of an incumbent's value from holding office—the thing the audience can take away—as associated with the amount of discretionary revenue (the budget surplus) that remains after keeping the incumbent's critical audience—the coalition—loyal, it is evident that this cost relative to revenue is continuously decreasing as coalition size increases so that the cost is enormous for non-democratic rulers and is only modest for democratic leaders. The former may use that revenue to pursue civic-minded policies or to enrich themselves. The latter are simply much more limited in their ability to direct revenue to themselves or to pet projects.

Furthermore, the lack of discretionary control over the use of revenue is not the only feature diminishing the value of office holding in large coalition, more democratic, regimes. The third dimension of audience costs relates to the fate of deposed leaders depending on the type of regime they led. Certainly, the fate of deposed leaders is quite different as coalition and selectorate size vary; that is, as we shift between different types of “non-democrats” and “democrats.” The latter live to campaign another day or to write their lucrative memoirs. The former, whether they were leaders of a rigged election autocracy (small W and large S) or of a monarchy or junta (small W and small S), often face imprisonment or execution, a much higher cost than their democratic counterparts (Goemans, 2000). Hence, the cost of losing office ( $ψ$ ) must be assessed in terms of: (1) the probability of losing office; and (2) the value lost if removed from office. The first component is expected to be higher for large coalition, large selectorate, democratic leaders but the second component is expected to be lower for large coalition rulers than for autocrats, junta rulers or monarchs.

Specifically, in the selectorate context, as W increases, the value of office decreases but as S increases the value of office, including the size of the budget surplus, increases. Increasing both W and S, and, therefore, their product, WS, means the government is becoming more democratic.⁷ Since

\frac{d^{2} S u r p l u s}{d W d S} < 0

it follows that as W gets larger, the budget surplus from increasing S grows slowly or shrinks. W’s effect dominates S's effect, diminishing the value of office as governments become more democratic. Indeed, the interaction between coalition size and selectorate size is expected to be the main indicator that $ψ$ is diminished in democracies compared with all other types of government. To see why this is so, note that S can be large whether W is large or small, but WS can only be large in a democracy. When both W and S are large, with the effect of W dominating the impact of S, audience costs decrease along with the diminished value of office. This implies that war is less likely when WS is large, making backing down less ominous (−m $ψ \to 0$ as $ψ \to 0$ ) as compared with the independent impact of W or of S.

The arguments that tie audience costs to coalition size and to selectorate size indicate distinct effects that link democratic governments to war compared with non-democratic governments. The effects of audience costs—through coalition size—and the effects through selectorate size or their interaction—indicate distinct, competing effects of democracy on conflict escalation and the credibility of threats. Leaders who depend on a large coalition have little budget surplus for their discretionary use and so face relatively low costs from losing office; leaders with large selectorates expect more budget surplus and so face higher costs from deposition. The product of the two highlights differences between democracies and the various forms of non-democratic governments and suggests a more nuanced view of the effects of democracy (larger W and larger S) than is argued for in the audience costs literature. These nuanced implications lead to testable hypotheses about the probability of war. Specifically, we expect that when A has begun a crisis (that is, made a threat) and state B has countered state A's threat, then:

1. The larger W is, the lower the probability of war started by A.

2. The larger S is, the higher the probability of war started by A.

3. Because creating audience costs is easier in categorically democratic governments, all else being equal, war should be more likely when a democratic government has made a threat if the rival has countered, indicating a more credible threat from democracies than autocracies.

4. The total effect of W, an indicator variable called Democracy (1 = Democratic and 0 equals not democratic) and the interaction of the two is expected to either be indeterminate or to indicate a significantly reduced risk of war. The same expectations prevail if we examine W, S and their interaction, remembering that the effect of W theoretically dominates the effect of S. That is, as coalition size increases, even a high probability of losing office is offset by a low utility loss, making the total expected effect of $ψ$ smaller than when W is smaller in the wide range of polities for which the categorical value of democracy is 0 (and $0 \leq$ S $\leq 1$ ).

5. The greater the contraction in per capita income (an indicator of m) for citizens in A, the more likely the leader in A is to initiate a war rather than back down.

These testable implications also suggest expectations higher up in the game tree. Specifically, we can ask what is the likelihood that A will initiate a threat to begin with given the expectations about whether B will counter the threat and, if so, whether A will wage war or back down. The equilibrium conditions for making a threat lead to the following hypotheses:

6. The more likely B is to counter A's threat, the less likely A is to make a threat to begin with.

7. The more credible A's threat (that is, the more Democratic A is), the more likely A is to make a threat.

8. But the less credible A's threat (the larger its winning coalition), the less likely it is to make a threat.

9. Although selectorate size may push A toward making a threat, coalition size is expected to dominate selectorate size and so the aggregate effect of coalition size interacted with selectorate size is to reduce the probability that A initiates a threat against B.

Data and variable specification

Kenneth Schultz (2001) has prudently noted that empirical evidence of the effects of audience costs is difficult to find and test for because outcomes rest on selection effects that result in the understatement both of mean audience costs across all states and the difference across regime types. The key to an empirical assessment, then, depends crucially on correcting for selection effects which my empirical strategy will partially do, mindful of the caveats just noted.

I use the Dyadic Militarized Interstate Disputes data (MIDs 5.0, Palmer et al., 2020) from the Correlates of War project to determine for each state A whether it was involved in a militarized dispute with state B. Hence, I construct a binary variable, HostilityA, that equals zero if state A in year t did not threaten B (as indicated by the variable mid5hiacta = 0). If A at least threatened B, then HostilityA = 1. The years over which HostilityA achieves a 1 encompass the period from 1816 to 2014. In all analyses of the likelihood of war, I filter the analysis only to look at cases in which HostilityA = 1 and in which an additional variable, B:Counters = 1, meaning that B responded to A's threat at least with a threat of its own (so that $mid 5 hiactb > 0$ ).

I define the dependent variable, War, in year t as equal to 1 if country A is coded as having begun a war against B (mid5hiacta $\geq$ 20) in either of years t + 1 or t + 2; that is, within a timeframe generally within the scope of the evolution of a dispute from threat initiation to the ultimate decision to back down or fight. Thus, War is forward looking and so is designed to be predictive of how current values of $ψ$ and m in A influence the subsequent, near-term prospects of A's fighting or backing down given that the conflict has unfolded to the point that B has countered A's threat. To ensure that only war initiation is counted and that ongoing wars are not double counted, I create a separate variable called Ongoing that receives a 1 in each year for which a war's final year (endyear in the MIDs data) is greater than its starting year (strytyr) when War = 1. I eliminate all observations for which the indicator Ongoing equals 1 so that War is only evaluated in years when A otherwise is not already at war with B.⁸

Partial correction for selection effects

By including only cases for which HostilitytA = 1, B:Counters = 1, and Ongoing = 0 when analyzing whether the independent variables increase or decrease the subsequent likelihood of war, I partially eliminate selection effects but only partially. Naturally, threats are made only if the threatening state believes engaging in a dispute will yield a greater payoff than not engaging in a dispute. That expectation, however, is not independent of the credibility with which the threatening state can commit to back up its threat by going to war if the rival does not back down. Thus, HostilityA = 1 indicates that the threatening state expects it has a credible threat while B:Counters ensures that we are examining A's choice to attack or back down after B has countered its threat. Hence, after evaluating the risk of war under different audience cost conditions, I then turn to whether A made a threat to begin with. To do so I no longer filter cases on whether B:Counters = 1 but, instead, then treat B:Counters = {0, 1} as an indicator of A's rational expectation about B's response if A decides to threaten B.

In keeping with selectorate theory and audience cost assumptions, I treat $ψ$ as a two-dimensional variable, with one dimension being the selectorate theory's estimate of the value of office and with the other dimension being the ease with which the audience can remove the incumbent from power. Coalition size, W, is used to estimate the utility of office for the leader. The measurement of W is continuous and is based on Bueno de Mesquita and Smith's recent indicator (Bueno de Mesquita and Smith, 2022), which is discussed below. The ease of generating audience costs, that is, the ease with which a leader can be removed from office, is indexed in two ways. I define a country as democratic using a binary variable, Democracy, if the Polity2 score for the country in the relevant year is at least as large as 6 (Vreeland, 2008). As an alternative indicator, I use Bueno de Mesquita and Smith's (2022) recent indicator of selectorate size, S, since when W is large, a large selectorate size means a challenger to the incumbent has many ways to assemble an alternative winning coalition and depose the incumbent. Thus, the value of $ψ$ is estimated as W, Democracy or S, and WDemocracy or WS.

By folding selectorate logic into an audience cost setting we are reminded of ongoing selection effects. The comparative static results of the game, applying selectorate logic to them, indicate that the preparedness to fight depends on the threshold value that must be passed regarding the probability of victory below which a state will not fight. Imposing selectorate logic on the audience cost framework, we are reminded that the chances of victory must be higher when W is large (Bueno de Mesquita et al., 1999), with W being largest when Democracy = 1, meaning that the value of office is relatively small compared with when A leads a non-democratic government. Conversely, when Democracy = 1, the ease of being removed from office increases, contributing to making more credible A's threat to fight. We also know theoretically that as S increases, the threshold value of p at which A will fight decreases. Thus, there are selection effects for large W political systems that deter them from fighting (and, perhaps, therefore make them reluctant to make a threat) and others that encourage them to fight if their rival does not give in to them. The selection effects pull in two directions, making threats more likely or less likely but making for clearer, albeit weaker predictions once a dispute is underway. As WS increases (meaning A is more democratic), war is likely to be avoided but as S increases, especially when W remains small, then war becomes more likely. Similarly, as W increases war is likely to be avoided but as Democracy increases from 0 to 1, war becomes more credible. As the continuous effects of W within the indicator variable for democracy are conjectured to dominate the effects of categorical regime type (democratic or non-democratic), the net impact of the interaction of W and Democracy is expected to be a decreasing propensity to go to war as W increases.

Models to be tested given a threat is made

I specify two sets of four models to assess the effect of audience costs ( $ψ$ ) and domestic incompetence (m) on the likelihood of war. The first and third are built around logit analyses of W, Democracy or S, and W × Democracy or W × S, as well as a measure of domestic incompetence which I call ShockSize. The second and fourth models are based on regression analyses that otherwise have the same specification. Of course, logit analysis drops cases that are perfectly predicted (whether correctly or incorrectly). These observations remain in the regression assessments. Therefore, the regression analyses apply to a larger sample but at the expense of treating the dichotomous variable, War, as if it were continuous. The third and fourth models add other domestic variables that should be controlled for in keeping with prior research that indicates they influence domestic, internal conflict-proneness. I discuss these below. All four model specifications include country fixed effects (denoted as FE in each table).

The indicator of domestic incompetence, ShockSize, is set to 0 if a country's per capita income increased at least as fast as the global median between t and t + 2. If the economy contracted relative to the global economy then ShockSize takes the absolute value of the contraction as the indicator of incompetence so that greater economic contraction means a larger value for m, the game's indicator of domestic policy incompetence.⁹ Although, of course, no one can know how an economy will perform in the coming two years (absolutely or relative to the global economy), I treat the deviation from global economic performance as a forward-looking indicator of whether a given economy is expected to do poorly relative to global performance. If, instead, we looked backwards at how the economy had done in previous years then we risk the possibility that the relevant leader will already have been deposed because of domestic incompetence.

Coalition size and selectorate size are estimated according to new procedures developed and tested by Bueno de Mesquita and Smith (2022). Their new, continuous indicator of coalition size, derived from several variables in the V-Dem dataset, covers virtually all states each year from 1789 to the present. Analyses of 30 distinct measures of public or private goods allocations, using Vuong tests and Akai Information Criteria, demonstrate that the new coalition indicator significantly outperforms their old indicator, the Polity Democracy–Autocracy indicator, Polity's dichotomous indicator and its tripartite indicator (democracy, anocracy, autocracy), as well as indicators of regime type by Przeworski, Boix, Geddes, Teorell and Lindberg, and others.

As noted, the third and fourth model specifications add controls for other factors known to influence leader decisions over domestic conflict (Bueno de Mesquita and Smith, 2018). The variable Domestic Unrest is coded as 1 if the country faced an attempted coup, civil war, or revolution in the current year. If none of these events occurred, Domestic Unrest is coded as 0. I control for the inexperience of the incumbent, defined as the leader having been in office for 2 or fewer years and also the log of how many years the incumbent has been in office as of the current year. Table 1 shows the bivariate correlations for all variables used in relation to the key audience cost variables, leaving out the correlations among the control variables owing to a lack of space.

Table 1.

Cross-correlation table.

Variables	W	S	Democracy	WS	W × Democracy	ShockSize
W	1.000
S	0.045	1.000
Democracy	0.788	0.246	1.000
WS	0.853	0.514	0.765	1.000
W × Democracy	0.815	0.231	0.987	0.796	1.000
ShockSize	−0.160	0.028	−0.161	−0.127	−0.170	1.000
Inexperienced	0.248	0.004	0.251	0.176	0.228	−0.064
logIncomingTenure	−0.346	0.047	−0.323	−0.251	−0.309	0.126
Domestic_Unrest	−0.085	0.121	−0.127	−0.019	−0.153	0.096

Test results: fight or back down

Tables 2 and 3 display the results of the models designed to predict whether a crisis is expected to escalate to war or not. The first and second model in each table draws our attention to the impact that W, Democracy or selectorate size, and the relevant interaction term, as well as ShockSize have on the likelihood of war. They provide a parsimonious predictive framework that focuses on the essential domestic political variables from the game in Figure 1 ( $ψ$ and m), melded with their empirical analogs from selectorate theory.

Table 2.

Anticipating war when a crisis exists, democracy = 1 if Polity $\geq 6$ .

	(1)	(2)	(3)	(4)
	Logit	OLS	Logit, FE	OLS, FE
Main
W (Coalition size)	−3.269 $^{* * *}$	−0.148 $^{* *}$	−3.963 $^{* * *}$	−0.158 $^{* * *}$
	(0.927)	(0.045)	(1.038)	(0.047)
Democracy	3.346 $^{* *}$	0.139 $^{*}$	4.165 $^{* *}$	0.155 $^{*}$
	(1.295)	(0.068)	(1.445)	(0.071)
W × Democracy	−3.805 $^{*}$	−0.165^†	−4.803 $^{*}$	−0.184^†
	(1.799)	(0.095)	(2.042)	(0.101)
ShockSize	4.793 $^{* * *}$	0.417 $^{* * *}$	4.876 $^{* * *}$	0.437 $^{* * *}$
	(0.903)	(0.068)	(0.911)	(0.069)
Domestic_Unrest			0.348	0.012
			(0.230)	(0.013)
Inexperienced			0.184	0.008
			(0.203)	(0.011)
Years of Experience			0.021	0.001
			(0.014)	(0.001)
Constant		0.129 $^{* * *}$		0.119 $^{* * *}$
		(0.020)		(0.023)

N	2327.000	3284.000	2106.000	3062.000
LROC	0.72		0.74
$Pr (W + Democracy + W Democracy) < 0$	0.0000	0.0000	0.0000	0.0000
$Pr (W + W Democracy) < 0$	0.000	0.0004	0.0000	0.0002
$χ^{2}$ or F	57.15	15.61	66.33	10.17
FE	50	138	47	135

Standard errors in parentheses. OLS, Ordinary least squares.

^†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Table 3.

Anticipating war when a crisis exists considering selectorate size.

	(1)	(2)	(3)	(4)
	Logit	OLS	Logit, FE	OLS, FE
Main
W (Coalition Size)	−1.654	−0.038	−2.287	−0.044
	(1.569)	(0.090)	(1.887)	(0.111)
S	0.194	0.018	0.005	0.011
	(0.906)	(0.051)	(1.088)	(0.063)
WS	−1.150	−0.085	−0.938	−0.083
	(1.547)	(0.089)	(1.821)	(0.111)
ShockSize	4.876 $^{* * *}$	0.417 $^{* * *}$	4.930 $^{* * *}$	0.437 $^{* * *}$
	(0.878)	(0.065)	(0.881)	(0.066)
Domestic_Unrest			0.308	0.012
			(0.225)	(0.012)
Inexperienced			0.208	0.009
			(0.199)	(0.011)
Years of experience			0.022	0.001 $^{x}$
			(0.014)	(0.001)
Constant		0.092^†		0.084
		(0.048)		(0.061)
N	2478.000	3543.000	2226.000	3283.000
LROC	0.72		0.74
$Pr (W + S + W S) < 0$	0.0091	0.0472	0.0091	0.0739
$Pr (W + W S)$	0.0000	0.0001	0.0000	0.0001
$χ^{2}$ or F	55.78	15.89	61.46	10.12
FE	51	140	48	137

Standard errors in parentheses.

^†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Theory leads us to expect that the larger the size of a government's coalition, W, is, the lower the probability that it chooses war over backing down given that a crisis has begun (i.e. HostilityA = 1), implying as well that A would not have made a threat at an earlier stage in the game. Thus, we expect that the sum of the coefficients for W and its interaction with Democracy or with S in all four models is significantly below zero, indicating war avoidance by A with A's coalition being larger, even within its categorically democratic regime type. That is, W’s effect is either neutralized by or dominates the effect of regime type; regime type does not dominate the continuous, nuanced impact of coalition size. Indeed, the sum of the coefficients for coalition size and its interaction with Democracy or with S is significantly less than 0 in all of the models ( $p < 0.001$ in each model). Additionally, W is significantly less than 0 in all of the tests in which the ease of creating audience costs is indicated by the variable Democracy, indicating that the larger a government's coalition is, regardless of its categorical regime type, the more reluctant it is to engage in war. Substantively, this tells us that leaders in larger-coalition, more democratic, settings act as if they have lower audience costs than non-democratic leaders as hypothesized. Given that their threats are less credible as their coalition gets larger, it seems that they are, in fact, less likely to attack in a crisis (and, therefore, less likely to initiate a threat) than are non-democracies, all else being equal. This seems to be true even though, as anticipated by the audience costs literature, being categorically a democratic government significantly increases the credibility of threats. That is, the coefficients for the democracy indicators are always positive and significantly so, signaling a preparedness to follow through with war after a threat has been made. That credible threat presumably also makes B more likely to back down rather than counter, although that effect is masked here as we filtered out such cases to minimize selection effects.

As hypothesized based on the game's logic, when a leader's domestic competence is in doubt—when ShockSize is large, indicating economic contraction—then it seems that leaders gamble on war as a possible means to resurrect their political fortunes (Downs and Rocke, 1994). The coefficient on ShockSize is positive and significant in all eight models. Likewise, the combined effects of domestic unrest, political inexperience or extensive experience all encourage leaders to gamble on war. Domestic unrest, of course, is an indicator that the incumbent's hold on power is tenuous. Earlier research has also shown that newness in office indicates a perilous time for leaders. The evidence here suggests that the combined effect of these factors encourages leaders to gamble on war as a possible means to resurrect their longer-term political prospects.

What then is the aggregate effect of the two characteristics of democracy that are highlighted here? We know that a larger coalition size diminishes the risk of war when it is—as it must be—accompanied by a large selectorate. But small to mid-sized coalitions, especially when accompanied by a large selectorate, promote war. That is, democracies reduce the risk and non-democracies, especially rigged election autocracies (small W and large S), increase it. Thus we see how the assertion by Fearon can be correct—democracies make a more credible threat than non-democracies—if we focus on just Democracy as a categorical regime type but not if we focus on the individual and interactive effects of W and Democracy (or W and S). The key results here remind us that democracies are not one-dimensional. Their small audience costs from a leader's point of view discourage war (or credible threats of war) and their high risk of domestic deposition encourage war (or make their war threats more credible) when a crisis is underway. Unidimensional understandings of democracy and non-democracy are prone to mislead or confuse our expectations regarding credible commitments and conflict escalation.

Before concluding the discussion of the risk of war when a threat has been made we should pause to examine how the models, driven exclusively by the domestic factors (W, Democracy or S, and economic incompetence, with country fixed effects), do in predicting whether or not war happened within two years of the year of observation. War is said to be anticipated if the predicted value from models 1 or 2 is greater than or equal to the median predicted value. Otherwise, the model is said to predict peace. To capture the impact of the models on actual war occurrence, I compare the probability that a war occurred when the model predicts war to when it predicts no war so that we see the odds ratio given that war was predicted and occurred. I then repeat that process to assess the odds ratio that there was peace when peace was predicted to when war was predicted. Table 4 shows the results for all observations, whether HostilityA = 1 or 0 (N varies between 4593 and 6787 across the tests); that is, before a crisis was known to exist.¹⁰ The table also reports the χ² for the simple crosstab. War is or is not predicted and war in fact did or did not occur.

Table 4.

Predicted odds ratios for war and peace given models 1 and 2.

	$\frac{P r o b (W a r \| P r e d i c t W a r)}{P r o b (W a r \| P r e d . P e a c e}$	$\frac{P r o b (P e a c e \| P r e d i c t P e a c e)}{P r o b (P e a c e \| P r e d . W a r}$	χ ²
Democracy Model 1	2.20	1.15	130.17
Democracy Model 2	4.36	1.15	343.56
Selectorate Model 1	2.28	1.16	132.49
Selectorate Model 2	4.02	1.18	317.52

When war is predicted using the Democracy indicator, it is 220% (N = 4947) more likely to have occurred during the subsequent two years than when Democracy model 1 predicted war would not occur. The odds of war being correctly anticipated are nearly twice as large at 436% (N = 6787) when our focus is on the larger sample in the regression analysis in Democracy model 2. We see very similar results when we focus on the Selectorate specifications in model 1 (odds ratio of war = 2.28, N = 4593) and model 2 (odds ratio = 4.02, N = 6469). When predicting peace, all the models do about equally well. Each adds modestly to a random draw on the odds that peace occurs given that peace was or was not predicted, with the odds ratios varying between 1.15 and 1.18. Peace, of course, is the more common circumstance and these odds ratios tells us that we are more likely to have false negatives—peace when war is predicted—than we are to have false positives—war when peace was predicted –with the predictions all done ex ante, before we know that HostilityA = 1. As the χ² values reinforce, the models do very wells at anticipating war even though they ignore all the standard international politics variables that dominate the study of war. Knowing the domestic circumstances that generate the risk of audience costs and the magnitude of those costs associated with losing office proves to be a good starting place for anticipating when crises are expected to escalate to war and when they are not.

When is a threat made?

Thus far we have conceptualized international crises as arising when the leader of state A threatens the leader of state B and B's leader counters the threat rather than capitulating to it. We have established in that context that breaking audience costs into their two components—the ease with which the audience imposes costs and the magnitude of those costs—helps anticipate when crises become wars or when war is averted. We saw strong empirical support for the hypotheses regarding the constituent dimensions of audience costs and the likelihood of war. With that information in hand we can now work our way up the game tree to A's initial decision whether to make a threat or not.

Table 5 replicates the earlier four models that included Democracy, but now with whether A makes a threat as the dependent variable and with whether B counters if a threat is made as an independent variable. Table 6 substitutes selectorate size for the Democracy indicator variable. The newly introduced independent variable that is a 1 if B counters a threat and a 0 otherwise is treated as A's rational expectation of what B would do if A threatened. This is measured using the variable for which we controlled earlier: B:Counters.

Table 5.

Anticipating when a threat is made, Democracy = 1 if Polity $\geq 6$ .

	(1)	(2)	(3)	(4)
	Logit	OLS	Logit, FE	OLS, FE
Main
B Counters	−0.043 $^{* * *}$	−0.006 $^{* * *}$	−0.045 $^{* * *}$	−0.006 $^{* * *}$
	(0.005)	(0.001)	(0.006)	(0.001)
W (Coalition size)	−1.695 $^{* * *}$	−0.274 $^{* * *}$	−1.967 $^{* * *}$	−0.303 $^{* * *}$
	(0.323)	(0.051)	(0.345)	(0.053)
Democracy	−0.932 $^{x}$	−0.161 $^{*}$	−1.161 $^{*}$	−0.192 $^{*}$
	(0.510)	(0.080)	(0.535)	(0.083)
WDemocracy	1.557 $^{*}$	0.271 $^{*}$	1.862 $^{*}$	0.309 $^{* *}$
	(0.690)	(0.108)	(0.735)	(0.115)
ShockSize	−0.690	−0.120	−0.537	−0.092
	(0.481)	(0.080)	(0.487)	(0.081)
Domestic_Unrest			0.184^†	0.027^†
			(0.105)	(0.015)
Inexperienced			−0.008	−0.005
			(0.090)	(0.013)
Years of Experience			−0.021 $^{* * *}$	−0.004 $^{* * *}$
			(0.006)	(0.001)
Constant		0.985 $^{* * *}$		1.025 $^{* * *}$
		(0.025)		(0.028)
N	5762.000	5921.000	5544.000	5652.000
LROC	0.71		0.72
$Pr (W + Democracy + W Democracy < 0$	0.0000	0.0000	0.0000	0.0000
$Pr (W + W Democracy) < 0$	0.8272	0.9765	0.8761	0.9535
$χ^{2}$ or F	93.14	18.55	112.02	13.99
FE	133	154	132	151

Standard errors in parentheses.

^†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Table 6.

Anticipating when a threat is made considering selectorate size.

	(1)	(2)	(3)	(4)
	Logit	OLS	Logit, FE	OLS, FE
Main
B Counters	−0.045 $^{* * *}$	−0.006 $^{* * *}$	−0.047 $^{* * *}$	−0.007 $^{* * *}$
	(0.005)	(0.001)	(0.005)	(0.001)
W (Coalition Size)	−1.295^†	−0.192^†	−1.598^†	−0.247^†
	(0.781)	(0.107)	(0.912)	(0.129)
S	−0.271	−0.033	−0.369	−0.059
	(0.461)	(0.064)	(0.531)	(0.077)
WS	0.285	0.030	0.404	0.062
	(0.790)	(0.108)	(0.914)	(0.130)
ShockSize	−1.007 $^{*}$	−0.194 $^{*}$	−0.920 $^{*}$	−0.175 $^{*}$
	(0.438)	(0.076)	(0.442)	(0.077)
Domestic_Unrest			0.129	0.017
			(0.097)	(0.015)
Inexperienced			−0.003	−0.001
			(0.084)	(0.013)
Years of experience			−0.018 $^{* *}$	−0.003 $^{* *}$
			(0.006)	(0.001)
Constant		0.974 $^{* * *}$		1.031 $^{* * *}$
		(0.061)		(0.075)
N	6280.000	6429.000	6010.000	6133.000
LROC	0.72		0.72
$Pr (W + S + W S) < 0$	0.0051	0.0022	0.0034	0.0016
$Pr (W + W S) < 0$	0.0000	0.0000	0.0000	0.0000
$χ^{2}$ or F	102.88	20.33	116.77	14.49
FE	134	157	132	154

Standard errors in parentheses.

^†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

The results reported in the two tables make clear that if B is expected to counter, meaning that A will have to choose between war and backing down, then, as hypothesized, A is very significantly less likely to make a threat to begin with. Here we see some of the important selection effects that Schultz (2001) has warned us to look out for in studying audience costs empirically. Crises, that is, the initiation of threats that could lead to war, are unlikely if the rival is expected to counter, resisting A's threat rather than capitulating to it. Additionally, the tables demonstrate that the probability of a threat being made is greatly diminished if coalition size is large, the state is a democracy or has a large selectorate, and if the interaction of these variables is large. However, the results in the two tables part ways when we distinguish between the effects of W × Democracy and W × S.

The larger the coalition given that the state is a democracy $(Polity 2 \geq 6$ ), the more likely it is that a threat will be made. This is consistent with the notion that the more democratic a democracy is (the larger its winning coalition), the more credible it expects its threat to be. However, that effect is partially contravened by the fact that when a state is categorically democratic (Democracy = 1), it is significantly less likely to initiate a threat. The net effect is that the three audience cost variables collectively significantly reduce the probability that a threat is made but the effects of coalition size and its interaction with Democracy are insignificant, highlighting the indeterminacy discussed in the statement of hypotheses earlier. Conversely, when we abandon a categorical view of regimes and, instead, focus on the aggregate effects of coalition size and selectorate size, the effects are uniformly highly significant and support the inference that large coalition, large selectorate regimes are especially unlikely to initiate threats. Again we see that what follows from the audience cost perspective depends importantly on whether we distinguish between the categorical ease with which the audience can impose costs or the magnitude of those costs from the perspective of the incumbent leader of the state.

Preliminary, exploratory conclusions

Prior investigations of audience costs make two assumptions that seem to obfuscate, rather than clarify, how domestic institutions shape audience costs and leader choices. First, much of the literature assumes that democratic audiences are better able to generate audience costs than are non-democratic audiences but then conflates the ease of generating audience costs with the (expected) magnitude of such costs. Thus, much of the literature ignores the possibility that how high the cost is to an incumbent leader may be inversely related to the ease with which the cost can be generated by the incumbent's key constituents.

The selectorate perspective indicates that citizen welfare is increasing as coalition size increases; that is, as a government becomes more democratic and accountable, the citizen-audience becomes better off. Yet the selectorate perspective also demonstrates that the incumbent's welfare is decreasing as coalition size increases. If a government becomes more democratic and its citizens suffer losses from failed policies (whether domestic or foreign), then those losses may lead them to threaten or successfully replace the incumbent, generating an expected or realized audience cost for their leader. However, the incumbent in a large coalition setting has little discretionary control over revenue and is often able to compete again another day. Both of these considerations make the immediate value of office small compared with its value for a non-democrat who, by the selectorate logic, has an incumbency advantage compared with democratic leaders. Small coalition leaders, especially when the selectorate is large, have substantial discretionary resources to lose if ousted from power and, in addition, face higher personal survival costs if ousted from office.

The literature generally treats democracy and non-democracy as either dichotomous or as unidimensional. The selectorate perspective, in contrast, treats all types of regimes as falling, minimally, within a two-dimensional institutional space made up of the size of the selectorate and the size of the winning coalition. From that perspective, rigged-election autocracies, monarchies and military juntas have quite distinct institutional structures and so cannot all be lumped together as “non-democracies.” While they are alike in generating high private rewards for members of the winning coalition, they differ markedly in the level of loyalty they generate to the incumbent. Monarchies, juntas and large coalition regimes all generate weak loyalty, making it easier for their key constituents to generate audience costs. This means that the risk of ouster is more comparable, all else being equal, in monarchies, juntas and democracies than in autocracies but the expected cost of ouster is quite different. This makes lumping all non-democracies together in terms of audience costs problematic.

The empirical results reported here indicate that we can improve our anticipation of crises and of their escalation by paying closer attention to the two dimensions of governance highlighted in the analysis. At least some of the indeterminacy in past analyses of crisis initiation and escalation can be cleared up by distinguishing between the independent effects of W, S, and WS on conflict choices. Given the empirical results here, combined with the merger of a simple audience cost perspective with the selectorate perspective, this preliminary investigation suggests that future crisis analyses grounded in audience cost concerns need to distinguish between the ease of generating costs, the magnitude of those costs, and the risk of actual ouster from office. Following that path, rather than merely attending to “regime type” and standard international power variables, can improve our understanding of the origins and escalation—or termination—of disputes.

Frank Zagare helped pave the way for recognizing that conflict choices are anticipatory and, therefore, involve looking ahead—working our way back up the game tree—to guide current choices. That lesson remains as powerful today as it was when first articulated decades ago. What has changed, thanks to Frank and others, is that we now know much more about how to do such analysis and how to translate it into useful policy guidance. Thank you Frank!

Supplemental Material

sj-tex-1-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-tex-1-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-tex-2-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-tex-2-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-tex-3-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-tex-3-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-tex-4-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-tex-4-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-tex-5-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-tex-5-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-pptx-6-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-pptx-6-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Supplemental Material

sj-zip-7-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental material, sj-zip-7-cmp-10.1177_07388942231153598 for A game of domestic imperatives: Audience costs and conflict avoidance by Bruce Bueno de Mesquita in Conflict Management and Peace Science

Footnotes

Acknowledgement

I am thankful for the helpful comments and discussion of this paper by the attendees at the October 2021 conference honoring Frank Zagare. And in that context, I must especially thank Frank Zagare and Vesna Danilovic for their comments, their support, and for assembling such an excellent and enjoyable conference. I am also extremely grateful for the helpful comments and criticisms offered by Alastair Smith and by the reviewers and editor of Conflict Management and Peace Science. I am, of course, responsible for all the many remaining flaws.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Bruce Bueno de Mesquita

Supplemental material

Supplemental material for this article is available online.

Notes

References

Brams

Zagare

(1981) Double deception: Two against one in three-person games. Theory and Decision 13(1): 81–91.

Bueno de Mesquita

Siverson

(1995) War and the survival of political leaders: A comparative study of regime types and political accountability. American Political Science Review 89: 841–855.

Bueno de Mesquita

Smith

(2017) Political succession: A model of coups, revolution, purges and everyday politics. Journal of Conflict Resolution 2017: 707–743.

Bueno de Mesquita

Smith

(2018) Political loyalty and leader health. Quarterly Journal of Political Science 13: 1–29.

Bueno de Mesquita

Smith

(2022) A new indicator of coalition size: Tests against standard regime-type indicators. Social Science Quarterly 103: 1–15.

Bueno de Mesquita

Morrow

Siverson

, et al. (1999) An institutional explanation of the democratic peace. American Political Science Review 93: 791–807.

Bueno de Mesquita

Smith

Siverson

, et al. (2003) The Logic of Political Survival. Cambridge, MA: MIT Press.

Carnegie

(2021) Military Resources and International War. Peter Lang.

Chiozza

Goemans

(2004) International conflict and the tenure of leaders: Is war still ex post inefficient. American Journal of Political Science 48: 604–619.

10.

Downs

Rocke

(1994) Conflict, agency and gambling for resurection: The principal–agent problem goes to war. American Journal of Poplitical Science 38(2): 362–380.

11.

Fearon

(1994) Domestic political audiences and the escalation of international disputes. American Political Science Review 88: 577–592.

12.

Goemans Henk

(2000) War and Punishment: The Causes of War Termination and the First World War. Princeton: Princeton University Press.

13.

Niou

Ordeshook

Rose

(1989) The Balance of Power: Stability in International Systems. New York: Cambridge Unirsity Press.

14.

Palmer

McManus

Dorazio

, et al. (2020) The MID5 Dataset, 2011–2014: Procedures, Coding Rules, and Description. Working paper.

15.

Powell

(1990) Nuclear Deterrence Theory: The Search for Credibility. New York: Cambridge University Press.

16.

Powell

(2006) War as a commitment problem. International Organization 60(1): 169–203.

17.

Schelling

(1960) The Strategy of Conflict. Cambridge, MA: Harvard University Press.

18.

Schultz

(2001) Looking for audience costs. Journal of Conflict Resolution 45: 32–60.

19.

Sharma

(2015) Kinship, property, and authority: European territorial consolidation reconsidered. Politics and Society 43(2): 151–180.

20.

Smith

(1998) International crises and domestic politics. American Political Science Review 92: 623–638.

21.

Vreeland

(2008) The effect of political regime on civil war: Unpacking anocracy. Journal of Conflict Resolution 52(3): 401–425.

22.

Weeks

(2008) Autocratic audience costs:Regime type and signaling resolve. International Organization 62(Winter): 35–64.

23.

Wittman

(1979) How a war ends: A rational model approach. Journal of Conflict Resolution 23(4): 743–763.

24.

Zagare

(1983) Toward a reconciliation of game theory and the theory of mutual deterrence, delivered at the Annual Meeting of the International Studies Association, Mexico City, 9 April.

25.

Zagare

Marc Kilgour

(2000) Perfect Deterrence. Cambridge Studies in International Relations, Vol. 72. New York: Cambridge University Press.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

3.49 MB

0.00 MB

0.08 MB

0.07 MB

0.00 MB

A game of domestic imperatives: Audience costs and conflict avoidance

Abstract

Keywords

A war or peace game

Solving the game

Regime type, audience costs and conflict escalation

Ease of generating audience costs ≠ audience costs

Data and variable specification

Partial correction for selection effects

Models to be tested given a threat is made

Test results: fight or back down

When is a threat made?

Preliminary, exploratory conclusions

Supplemental Material

sj-tex-1-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-tex-2-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-tex-3-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-tex-4-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-tex-5-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-pptx-6-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Supplemental Material

sj-zip-7-cmp-10.1177_07388942231153598 - Supplemental material for A game of domestic imperatives: Audience costs and conflict avoidance

Footnotes

Acknowledgement

Funding

ORCID iD

Supplemental material

Notes

References

Supplementary Material

Ease of generating audience costs $\neq$ audience costs