The Domestic Sources of International Reputation

Existing research finds that leaders develop international reputations based on their past behavior on the international stage. We argue that leaders’ domestic choices can also influence their international reputations, perhaps as much as their past foreign policy decisions do. Using formal theory and intuitive argumentation, we develop an overarching framework to predict how much any domestic choice will affect a leader’s international reputation. We theorize that certain domestic choices can inform expectations about future international crisis behavior based on the extent to which (1) the costs at stake are similar to those of an international crisis and (2) the domestic issue is salient relative to foreign policy. We use conjoint experiments and other evidence to show that many domestic choices have significant international reputational effects. There is some evidence that the reputational effect of certain domestic choices may equal that of fighting in a previous international crisis.

I n August 1981, U.S. President Ronald Reagan fired 11,345 air traffic controllers who went on strike. This decision was costly for Reagan because the U.S. public was sympathetic to the controllers and inconvenienced by reduced flight volume (Craig 2020). Although the labor dispute had nothing to do with foreign policy, several observers argued that the president's choice improved his international reputation for resolve. National Security Advisor Richard Allen called it "Reagan's first foreign policy decision," whereas newspaper columnist William Safire said Reagan's choice would give the president a "reputation for strength" that would deter Soviet aggression (McCartin 2011, 329). An aide to Democratic House Speaker Tip O'Neill reported that Soviet officials O'Neill met in Moscow were impressed with Reagan's action (Morris 1999, 448, 792-3).
The possibility that leaders can gain an international reputation for resolve based on their domestic choices has gone almost entirely unacknowledged in the international reputation literature. Scholars increasingly agree that international reputations exist (Crescenzi 2018;Kertzer, Renshon, and Yarhi-Milo 2018;Sartori 2005;Weisiger and Yarhi-Milo 2015) and that they adhere to both states and leaders (Lupton 2020;Renshon, Dafoe, and Huth 2018;Wolford 2007;Wu and Wolford 2018). 1 Yet this research only considers how past foreign policy behavior, usually in the context of international crises, influences foreign observers' expectations about future resolve. We show that past international behavior, although important, is not the sole determinant of a leader's international reputation. 2 We argue that a leader can cultivate a reputation for international resolve through domestic choices that have nothing to do with foreign policy. We identify specific examples of informative choices, such as how a leader responds to protests or whether a leader compromises during domestic policy negotiations. More importantly, we develop an overarching framework that explains the conditions under which any domestic choice can affect a leader's international reputation. We theorize that leaders often face domestic choices that pit a more costly but potentially higher reward option against a safer alternative. In these contexts, the influence of the choice on the leader's international reputation depends on two factors: (1) cost similarity, the extent to which the costs associated with the domestic choice are similar to the costs of war, and (2) salience, the extent to which the leader cares about the domestic choice's payoffs relative to foreign policy.
We introduce these two dimensions into a classic formal model of repeated crises. The model illuminates that domestic choices with sufficiently high cost similarity and salience can have an equal or even greater effect on international reputations for resolve than does the choice to fight in a past international crisis. This is possible because the high salience of some domestic choices can overpower the incentives to misrepresent that typically confound learning in repeated-crisis models. For example, a leader facing an Arab Spring-like protest is too focused on remaining in power to give serious consideration to the strategic implications of his actions for future international crises. This enables the leader's behavior to reveal more information about his true disposition to foreign audiences. Our model predicts that domestic choices with moderate cost similarity and salience can also generate reputational effects, albeit on a smaller scale. However, some domestic choices are too low in cost similarity or salience to have any reputational effect.
Using a conjoint experiment, we provide causal evidence that certain domestic choices can influence expectations about international resolve. In it, we randomly varied descriptions of a leader's past international and domestic choices while in office. We then asked respondents how likely the leader was to stand firm in a future international crisis. Our results confirm our two main predictions: several domestic choices affect international reputation, and choices with higher cost similarity and salience have a larger effect than do choices that score lower on these two dimensions. We find that a leader's choice to violently suppress domestic protests affects her international reputation as much as fighting in a previous international crisis does. We find that a leader's choice to stage a coup and bargain hard over health care reform also affects her international reputation for resolve. Finally, we find that the cumulative reputational effect of these three domestic choices is greater than the effect of fighting in a past international crisis.
We find support for our causal mechanism using a postsurvey questionnaire. We use a supplementary experiment to increase our confidence that our main experiment is robust to alternative design choices. Consistent with our main experiment, the supplementary experiment strongly supports our two main predictions, although it does not show that any single domestic choice is equally as informative as an international crisis. 3 The supplementary study also provides additional support for our focus on cost similarity and salience. Finally, we illustrate our theory using five historical anecdotes.
Our study greatly expands the domain of international reputation research by connecting reputation building to domestic politics. The existing literature's focus on international behavior as the sole method of reputation building is an untested assumption (Crescenzi 2018;Kertzer, Renshon, and Yarhi-Milo 2018;Lupton 2020;Renshon, Dafoe, and Huth 2018;Sartori 2005;Weisiger and Yarhi-Milo 2015). We show this assumption is not valid. This is important because leaders spend the vast majority of their time on domes-tic issues. 4 Thus, a lot of reputation building may have gone unnoticed. We lay the groundwork for further expansion of the international reputation literature by providing a framework for understanding the extent to which any domestic choice can affect a leader's international reputation.
On the flip side, comparativists study many domestic choices largely in isolation from international factors. Our research suggests these may have unrecognized international implications. In addition to demonstrating the reputational effect of repression (Pierskalla 2009), coups (Bell and Sudduth 2017;Powell 2012), and domestic policy negotiations (Canes-Wrone and de Marchi 2002) in our experiment, we use postexperiment questions (see Appendix) to probe the international reputational effects of anti-immigration restrictions (Miller and Peters 2020), discrimination against minorities (Blaydes and Linzer 2012), the nationalization of private industry (Haber and Menaldo 2011), and purges (Magaloni 2008). 5 We find preliminary evidence that all of these domestic choices could have international reputational consequences. Thus, we establish new connections between comparative politics and international relations research.
We also contribute to the literature on the domestic sources of international conflict. Whereas others show that domestic factors constrain leaders or shape their incentives in crises (e.g., Fearon 1994;Putnam 1988;Schultz 1999), our findings suggest that domestic choices made long before a crisis erupts also affect crisis outcomes. Additionally, we contribute to the literature on leaders in conflict (e.g., Colgan 2013; Dafoe and Caughey 2016;Debs and Goemans 2010;Fuhrmann 2020;Horowitz, Stam, and Ellis 2015). Our findings suggest that resolve is viewed at least partially as a leader-level characteristic that persists across domestic and international situations, making individual leaders crucial to analyzing reputations.
Finally, our research has important policy implications. Our findings suggest that leaders trying to enhance their reputations must be mindful of their choices in both the domestic and international spheres. By taking tough positions against domestic opponents, as Reagan did, leaders can enhance their reputations for resolve. Taken to an extreme, this policy implication could have grim consequences for human rights. On the other hand, a more optimistic implication of our research is that states need not necessarily fight international wars to cultivate reputations for resolve. Rather, they have domestic opportunities to show resolve that may be less deadly.
We begin by reviewing the literature on international reputations for resolve. We then lay out our theoretical argument about how domestic choices affect 3 In the supplementary experiment, we find that the effect of protest repression is distinguishable from a past international crisis at 95% confidence, but it is still large and significant with 99.99% confidence. Furthermore, the cumulative effect of domestic choices is larger than past crisis behavior. 4 Lindsey and Hobbs (2015) show that U.S. presidents spend only about one-quarter of their time on foreign policy. 5 A handful of studies connect such events to international economic and institutional outcomes (e.g., Bermeo and Leblang 2015), but they do not relate them to international reputation and conflict. international reputations. Next, we present our main experiment. Finally, we provide additional evidence that increases our confidence in our main claims.

THE DETERMINANTS OF INTERNATIONAL REPUTATIONS FOR RESOLVE
Reputation refers to "beliefs about an actor's persistent characteristics or tendencies based on that actor's past behavior" (Jervis, Yarhi-Milo, and Casler 2021). We are interested in how domestic choices affect international reputations for resolve. Resolve generally means willingness to take some action despite costs (Kertzer 2016, 8). In keeping with the international security literature, we define resolve more specifically as willingness to use military force. Schelling (1966) was among the first to write about reputations for resolve. He argued that commitments are interdependent, meaning that backing down in one crisis raises doubt about resolve on all future issues. A variety of recent studies have confirmed that past choices affect the international reputations of states and leaders (Jervis, Yarhi-Milo, and Casler 2021). 6 Virtually all of these studies have focused on how leaders cultivate reputations through foreign policy choices. Most focus on how standing firm in a prior crisis creates a reputation for resolve in future crises (Harvey and Mitton 2016;Huth 1988;Kertzer, Renshon, and Yarhi-Milo 2018;Lupton 2020;Renshon, Dafoe, and Huth 2018;Sartori 2005;Weisiger and Yarhi-Milo 2015). 7 Others have analyzed the reputational effect of keeping alliance commitments (Crescenzi 2018;Gibler 2008;Mattes 2012;Miller 2012;Narang and LeVeck 2019).
The potential for leaders' domestic choices while in office to influence their international reputations for resolve has received little attention. Some find that reputations for resolve in domestic conflict exist in interactions with domestic separatists (Bormann and Savun 2018;Walter 2006), but they do not consider whether these domestic reputations travel to the international level. Wu, Licht, and Wolford (2021) analyze how domestic politics influence leaders' reputational incentives, but they do not consider how domestic choices influence international reputations. Kydd (1997) discusses how domestic ideological moderation and treatment of minorities influences international perceptions, but he is concerned with perceptions of restraint. Similarly, normative democratic peace explanations (e.g., Russett 1993) suggest that peaceful domestic dispute resolution leads to expectations of peaceful international behavior, but this is also more relevant to reassurance than resolve.
In order to understand the formation of reputations for resolve, it is necessary to consider the components of resolve itself. Kertzer (2016) proposes an interactionist theory of resolve that incorporates both situational factors (related to the costs and benefits in a particular situation) and dispositional factors (related to an individual's character). This view accords with game-theoretic work that also models resolve as being both situational and dispositional (Debs 2022). According to this perspective, individual behavior varies somewhat across situations due to different costs and benefits, and yet because an individual's disposition exerts a consistent influence on cost sensitivity, behavior also shows some consistency. Kertzer (2016) shows that four specific dispositional attributes influence individuals' willingness to fight: time horizons, risk tolerance, concern with honor, and willpower. Others have identified casualty tolerance, revenge, concern with perceptions, and madness as dispositional attributes that influence sensitivity to the costs of fighting (Brutger and Kertzer 2018;Dafoe and Caughey 2016;Kertzer and Brutger 2016;Kreps and Maxey 2018;McManus 2019;Stein 2019;Yarhi-Milo 2018).
Because dispositions influence behavior, witnessing behavior allows observers to make inferences about a leader's disposition. Thus, the argument that leaders can develop a reputation for resolve based on prior international crisis behavior is well grounded in the theory that resolve stems, in part, from an individual's disposition. Meanwhile, the situational aspect of resolve implies that the similarity of situations is another source of inference. We take the implications of this conceptualization of resolve one step further, arguing that observers can learn about leaders' dispositions from domestic choices.

DOMESTIC CHOICES AS A SOURCE OF INTERNATIONAL REPUTATION
We theorize that leaders often face domestic choices that resemble international crises in that a more costly but potentially higher reward option (hereafter referred to as standing firm) is pitted against a safer alternative (i.e., backing down). Because leaders' dispositions influence their choices, these situations offer opportunities for foreign observers to learn about a leader's resolve. We theorize that the extent to which a domestic choice influences international beliefs about resolve (if at all) is determined by two attributes of the choice. The first is the similarity between the costs of standing firm domestically and the costs of fighting in an international crisis. The second is whether the domestic choice is sufficiently salient relative to foreign policy.
The first condition required for domestic choices to affect international reputations is that the costs of standing firm domestically must be sufficiently similar to the costs of standing firm (i.e., fighting) in an international crisis. Several scholars have previously noted the importance of similarity. Jervis (1982, 10) notes the importance for reputation formation of both similar 6 For a contrasting view, see Hopf (1994), Mercer (1996), andPress (2005). 7 There is also a minority view that backing down in one crisis could increase expectations of standing firm in the next crisis because observers will reason that the state needs to restore its reputation (Jervis 1982, 12). This may be true occasionally, but the prevailing evidence and game-theoretic logic suggest that this is not commonly the case. situations and similar costs of standing firm. Huth's (1999) literature review emphasizes the importance of similarity by arguing that there is more evidence of reputation formation within dyads and regions than worldwide. Mercer (1996, 17) emphasizes a lack of similarity between situations as a primary reason that reputations do not travel. More recently, Crescenzi (2018, 48) argues that one state will learn more from how another state interacts with a third state when the third state is more similar to itself.
Although situations can be similar in their costs and/or benefits, we focus particularly on cost similarity for three reasons. First, it is common for formal models to incorporate uncertainty about resolve through the cost parameter, so setting up our formal model this way allows for comparability with previous models. Second, many of the main dispositional traits that Kertzer and others associate with resolve-such as casualty tolerance, risk tolerance, and willpower-affect costs more than they do benefits. Third, there are more obvious examples of cost similarity than of benefit similarity. For example, loss of life, risk to a leader's political position, and monetary expenditures are all costs that we could reasonably expect a leader to value similarly at the domestic and international levels. In contrast, it is more difficult to compare the valuation of domestic benefits, such as implementing a policy, with that of international benefits, such as protecting an ally. For these reasons, we focus on cost similarity. However, we believe that our theory can also apply to benefit similarity in principle, and we return to this point in the conclusion.
If the costs in two situations are similar, then the leader's willingness to tolerate costs in one situation will be informative about her cost tolerance in the other. We argue that some domestic choices have costs that are similar enough to the costs of war to influence a leader's international reputation for resolve. For example, the choice to violently repress domestic protests usually results in human casualties, which is also a cost of war. Likewise, the choice to stage a coup puts an individual's safety and power at risk, and initiating a war carries similar risks if it goes badly. As a final example, for a leader involved in domestic policy negotiations, the costs of rejecting compromise might include policy failure, public frustration, or a struggle to remove opposing officials. These costs are less similar to the costs of war but not entirely unrelated, as war can also lead to policy failure and public dissatisfaction.
To be clear, we do not claim that any of these costs are identical to the costs of war. For example, we believe that how a leader perceives the cost of killing civilian protesters is probably not identical to how the leader perceives battle deaths or even civilian casualties resulting from war. However, we argue that these costs are related. Because they both involve human deaths, they both tie into the dispositional trait of casualty tolerance. Thus, observing a leader kill civilian protesters domestically should raise estimates of that leader's casualty tolerance, and this should in turn raise estimates of his willingness to use force abroad. In summary, our assumption is that domestic choices share some common costs with war. We expect that domestic choices with higher cost similarity will have a greater reputational effect.
Cost similarity is sufficient to explain why there is some transference of reputation from the domestic to international level. Yet because international crises are more similar to each other than to domestic choices on average, one might intuit that foreign observers will always learn more from a leader's past crisis behavior than from domestic choices. However, we have not yet accounted for incentives to misrepresent.
Weakly resolved leaders are aware that fighting in a crisis today will enhance their reputations for resolve, which will increase their chances of prevailing in future disputes without fighting (Sechser 2010). This may incentivize even weakly resolved leaders to fight in the current crisis when they expect that future crises are likely (Jervis 1982, 11). This limits the amount that adversaries can learn from a leader's previous international crisis behavior, as they are unsure about whether the leader fought due to genuine resolve or the incentive to appear resolved. However, as other scholars have noted, the stakes in all situations are not equal (Schelling 1966, 56-7). Importantly, the incentive to appear resolved only plays an important role when the present stakes are sufficiently low relative to expected future stakes. When the leader faces a high-stakes choice today, she is more likely to choose based on the current costs and benefits without strategizing about the future. Thus, more salient choices offer more genuine insight into future willingness to fight.
For this reason, salience is the second condition that allows domestic choices to influence international reputations. 8 When a domestic choice is relatively more salient-that is, more important to a leadercompared with foreign policy, the leader will be less likely to act in a way that deliberately exaggerates their resolve and more likely to act in a way that reveals their true disposition. The incentive to misrepresent does not disappear, but it is more likely to be outweighed by concern with present costs and benefits. For example, President Obama's decision to let the federal government shut down in 2013 rather than make budget concessions, as well as President Trump's decision to challenge the 2020 election results, revealed information about these presidents' cost tolerance that was potentially relevant to international resolve. We conjecture that both Obama and Trump weighed the domestic consequences of the choice before them much more than the future consequences for their international reputations. In Jervis's (1970) conception, these domestic choices were indices because it was too costly to manipulate them for purposes of deception. Therefore, these domestic choices arguably revealed a great deal about Obama's and Trump's dispositions.
These examples are not unique. Domestic politics are often quite salient to leaders relative to foreign policy. Domestic politics are an important determinant of a leader's ability to retain power, and domestic achievements are a critical part a leader's legacy. Even in the United States, a superpower with interests around the globe, presidents spend about three-quarters of their time on domestic issues (Lindsey and Hobbs 2015). Our model below suggests that if a domestic choice is salient enough, it can compensate for lower cost similarly to produce a reputational effect equal to that of a foreign policy crisis.
For clarity, we theorize primarily about the reputational effect of individual domestic choices. However, we note that domestic choices are far more common than international crises for nearly all leaders. According to the ICB project, the average country faced only one international crisis every 11 years in the post-World War II era, and even the US faced an average of only one crisis per year (Brecher and Wilkenfeld 2000;Brecher et al. 2021). In contrast, the Phoenix event data coded based on BBC Monitoring's Summary of World Broadcasts shows that the average leader faces a choice of how to respond to a domestic protest every 1.4 years (Althaus et al. 2019). Leaders face lower-stakes domestic choices, such as how to bargain with political opponents, on a near-daily basis. Therefore, we also explore the aggregate effects of domestic choices.

Formal Presentation
If we accept that domestic choices can influence international reputations, we still have three lingering questions. First, do all domestic choices provide at least some information, or are some choices so different that they are completely uninformative internationally? Second, for domestic choices that provide information, can we identify which choices are the most (or least) informative as a function of cost similarity and salience? Third, can a domestic choice, in theory, be as or more informative than past actions in an international crisis? To answer these questions, we turn to game-theoretic analysis. The full solution and description of the formal model appear in the Appendix. Here, we briefly describe the model and its implications.
Our model, depicted in Table 1, is a variant of the classic repeated-crisis model for analyzing reputation. 9 The challenger (A) is uncertain about the defender's (B's) sensitivity to the costs of fighting in an international crisis in the second period (c 2 ). The challenger decides whether to issue a challenge in the second period after observing B's first-period choice.
In classic models, both the first and second periods are international crises. We innovate by assuming that the first period can represent any setting where B faces a choice between backing down and standing firm, where standing firm yields potentially higher benefits but is more costly and risky. 10 This first period could represent an international crisis, but it could also represent a domestic choice, such as whether to crack down on protests or whether to make concessions on domestic policy. Thus, B in our model represents a leader (or group of elites) who makes both foreign and domestic policy choices.
Our additional innovation is introducing two parameters to capture cost similarity and salience. First, we introduce the parameter α, which is the probability that c 1 equals c 2 , to capture the similarity in the costs at stake in the first and second periods. When α is low, A believes that B's costs for standing firm in the first period are unlikely to be related to B's costs of fighting in the second period. 11 When α equals 1, A believes that B's costs in both periods are certainly identical. 12 Second, we introduce the parameter θ , which is multiplied by B's payoff for standing firm in the first period. This captures the salience of the first-period choice relative to the second. We normalize the salience of an international crisis to 1. When θ is larger than 1, it means that the first period is a domestic choice in which the outcome (including the costs and benefits) is more important to the leader than the outcome of an international crisis is.
We solve for pure Bayesian equilibria and focus on a pure strategy, semiseparating equilibrium. We detail this equilibrium in the Appendix. In it, A makes a second-period challenge if and only if B backs down in the first period. Some types of B are not influenced by A's contingent challenge because their resolve in both periods is high (low). These types stand firm (back down) in both periods. Other types of B would incur a net loss if they stood firm in either period. And yet they stand firm in the first period because they know it will convince A to make No Threat in the second period. Knowing that these types of B exist, A remains uncertain about B's resolve to fight in the second period even after B stands firm in the first period. The equilibrium holds for values of α and θ that are high enough for A to believe that most types who stood firm in the first period will also do so in the second.
We visually represent the model's implications in Figure 1. Each point on the plot represents a different first-period choice. The x-axis plots how first-period choices vary in their cost similarity to an international crisis (α). The y-axis plots the salience of the first-period choice relative to the second-period international crisis (θ). The red dot shows the first-period parameters that represent the repeated-crisis model (θ = 1, α = 1). 13 To answer our first question-are all domestic choices informative?-we search for thresholds for salience (θ) and cost similarity (α) below which players Nature determines B's cost of standing firm in an international crisis (c 2 ). Then, Nature determines whether B's cost for standing firm in the first period is the same or uncorrelated. Information: B observes the true c 1 ,c 2 . A only knows the underlying probabilities.
B is faced with an initial choice, either domestically or in a foreign crisis, to stand firm at a cost or back down. A observes B's choice and can learn from it, but A does not participate.
The foreign policy crisis between A and B.
A decides whether to threaten B or not in an international crisis. The threat need not be a verbal threat and can be any type of military challenge. If A threatens, B can stand firm or back down.
Note: α represents the probability that B's cost for standing firm in the first period is the same as in the second period, and θ represents the relative salience of the two periods. 13 We make the simplifying assumption that all international crises have perfect cost similarity and equal salience. This assumption is consistent with previous game-theoretic and quantitative research (e.g., Sartori 2005; Weisiger and Yarhi-Milo 2015). However, it has been critiqued by Mercer (1996), who argues that international crises are often too dissimilar for reputations to travel among them. We acknowledge that this simplifying assumption is not entirely realistic, but by exaggerating the potential learning from international crises, we create the hardest possible test for showing that domestic choices can be equally informative.
will not play a semiseparating equilibrium. We plot the results using the shaded and unshaded areas in Figure 1. The shaded area represents parameter ranges where the first period choice cannot reveal enough information to support a semiseparating equilibrium. The unshaded region represents parameter ranges where semiseparating equilibria exist. Intuitively, this means that some domestic choices are so dissimilar (or lacking in salience) that they cannot be informative no matter how salient (or similar) they are. We summarize this core result as follows: Result 1: There exists both a minimum salience threshold and a minimum cost similarity threshold. If a domestic choice has a level of either cost similarity or salience that is lower than the respective threshold, it will not affect a leader's international reputation for resolve. If a domestic choice meets both thresholds, then it will affect their international reputation.
Result 1 establishes a theoretical claim that domestic choices can generate international reputations. Yet leaders make dozens of choices daily. Should we expect foreign observers to analyze all of them? The thresholds suggest that the answer is no. Domestic choices must meet two necessary conditions to be influential.
Our second question asks, which domestic events have the largest reputational consequences? We use the model to derive a general expression for the amount of learning that occurs given a specific set of parameters that can support a semiseparating equilibrium. We compute the amount that A updates his beliefs as Q (Equation 1 in Appendix, 3). This measures the difference in A's belief about B's resolve in the second period depending on whether B stood firm or backed down in the first period. When Q is large, it means that observing B stand firm makes A much more confident that B will fight than observing B back down. When Q is small, it means that B's first-period choice has little effect on A's beliefs. The variable Q is used to derive the comparative statics below and is also the quantity that we estimate in our experiment.
The amount A learns from B's actions in the semiseparating equilibrium (Q) is a function of B's probability of success (p), B's value for the issue in dispute (π), cost similarity (α ), and salience (θ ). Although all of these parameters affect the amount A learns in a semiseparating equilibrium, our informal intuition drives us to focus on cost similarity and salience. We compute the partial derivative of Q with respect to both cost similarity and salience in the Appendix. We show that partial derivatives of Q with respect to α and θ are strictly increasing independent of the other parameters. Therefore, these parameters have an independent and monotonic effect on the amount of learning. 14 This means that increasing either the cost similarity or the salience of a domestic choice will increase the amount A learns if B decides to stand firm relative to backing down. 15 We plot the independent effect of salience using the two squares that appear in Figure 1. As explained earlier, all of the first-period choices represented in the unshaded area affect A's belief about B's resolve (Q) as part of a semiseparating equilibrium. However, the curves demonstrate that the precise amount A learns from B's first-period choice depends on α and θ. The two squares exemplify two domestic choices that share the same level of cost similarity. However, one (the shaded square) has higher salience than the other (the unshaded square). Comparing the two squares illustrates that a higher-salience choice contributes more to A's learning than does a lower-salience choice when holding cost similarity and all else constant. We can hold salience constant, vary cost similarity, and get an equivalent finding. This suggests two more results: Result 2: If a domestic choice meets the thresholds for a semiseparating equilibrium to exist, then that domestic choice will have a larger effect on a leader's international reputation for resolve if it is more salient.
Result 3: If a domestic choice meets the thresholds for a semiseparating equilibrium to exist, then that domestic choice will have a larger effect on a leader's international reputation for resolve if its cost similarity is higher.
Note: The plot shows the informational value (Q ) of B's firstperiod choice as a function of α and θ. The case where the first period is an international crisis is marked with a red dot at θ =1, α = 1. The curves show how much learning is possible from various hypothetical first-period domestic choices relative to the international crisis. The shaded regions represent domestic choices where a separating equilibrium cannot be supported. The scales of θ and α are different, so the relative sizes of the shaded regions along each axis are not meaningful.
14 These features allow us to make precise empirical predictions independent of the other parameters, e.g., greater cost similarity increases the reputational effect regardless of other parameters. 15 The Appendix also solves for the partial derivative of Q with respect to p. We show that the effect is non-monotonic. Whether the partial derivative of p is positive or negative depends on θ and other variables. It follows that we cannot make clear predictions about the relationship between the probability of success and reputation.
Finally, we address the third question posed above. We want to know, "Can A learn as much from a domestic choice as from a past international crisis?" The solid red curve in Figure 1 shows the location of all first-period domestic choices that provide exactly the same amount of information as does a first-period international crisis (i.e., they have the same Q). The shape of this curve is intuitive given the independent effects of cost similarity and salience described above. It shows that as the costs of a domestic choice become less similar to the costs of fighting internationally, the domestic choice must be increasingly salient in order to provide as much information as an international crisis does. This emphasizes the importance of including both cost similarity and salience in our theory. Because all or most domestic choices have costs that are less similar to the costs of international crises than the costs of international crises are to each other, focusing only on cost similarity would lead us to conclude that domestic choices have inherently less influence on international reputation. However, by accounting for the trade-off between cost similarity and salience, the figure shows a theoretical range of domestic choices that are equally influential.
We also want to know whether domestic choices can be more informative than an international crisis. Above the red solid line in Figure 1 are cost similarity and salience scores where a domestic choice is more informative than an international crisis (higher Q). Below it, a domestic choice is less informative than an international crisis (lower Q). As examples, we include two other lines showing parameters values that enable a domestic choice to be precisely 20% more (dashed line) and 20% less (dotted line) informative than an international crisis. This suggests, in answer to our third question, that some domestic choices can, in principle, be as or even more informative than an international crisis.
Result 4: It is possible for a leader's domestic choice to have sufficient salience and cost similarity that it will have an equally large or larger effect on the leader's international reputation for resolve as the leader's past international crisis behavior.
To be clear, the model does not tell us what these highly informative choices are or whether they exist in real life.

Empirical Implications We Test
Our model has several nuanced theoretical results. But at root, we make a broad claim that connects the international and domestic levels: the choices that leaders make domestically can influence their international reputations for resolve. Thus, from Result 1, we derive Hypothesis 1: H1: International observers will have higher estimates of a leader's willingness to fight in an international crisis after witnessing the leader stand firm domestically than after witnessing the leader back down domestically.
To be clear, Result 1 does not say that every domestic choice is influential, but rather only those that meet certain thresholds for cost similarity and salience. However, for our theory to be correct, we must be able to validate H1 with at least some domestic choices. If we cannot, it would suggest that either our framework is incorrect or that the thresholds are so high in real life that no domestic choice is informative.
Although we expect all domestic choices that pass the thresholds to be influential, we do not expect them to be equally influential. As Results 2 and 3 tell us, a choice's reputational influence should increase in both its cost similarity and its salience. This suggests our second hypothesis: H2: International observers will update their estimates of a leader's willingness to fight in an international crisis based on a domestic choice to a greater extent when the choice has high cost similarity and salience than when it has low cost similarity and salience.
This hypothesis tests a main implication of our theory because it contrasts choices that our theory predicts have larger and smaller reputational effects. If our experiment showed that choices with greater cost similarity and salience were not more influential, then it would provide evidence against our causal mechanism. To be clear, our theory also suggests that cost similarity and salience have independent effects on the reputational effect of choices. We explore this in our supplementary experiment.
We also push our predictions further by exploring (1) the upper bound of the effect of domestic choices on international reputation and (2) the empirical domain over which our theory applies. Perhaps the most surprising theoretical result from our formal model is Result 4. It tells us that a single domestic choice can be as or more informative than a past international crisis choice if salience and cost similarity are sufficiently high. However, the model does not tell us whether any real-world domestic choices clear this bar. We offer a conjecture that explores the upper bound of reputational effects for a single domestic choice. We focus on the possibility of domestic choices being equally, rather than more, informative because we view this as more likely. Whereas the hypotheses are clearly predicted by the formal model, this conjecture is more speculative: Conjecture 1: A domestic choice will influence international observers' estimates of a leader's willingness to fight in an international crisis equally as much as past international crisis behavior when it has very high cost similarity and salience.
Our formal theory examines the effect of one domestic choice at a time. But in our informal theory we noted that important domestic choices (e.g., how to respond to protests, how to secure support for major policy initiatives) are far more common than international crises and that leaders dedicate most of their time to domestic choices. Therefore, another way to explore the effect of domestic choices is through their cumulative effect. This leads us to a second conjecture: Conjecture 2: The cumulative effect of all informative domestic choices on international observers' estimates of a leader's willingness to fight in an international crisis will be greater than or equal to the effect of behavior in a single past international crisis.
In our survey design, we include four domestic choices and one past military crisis that a leader can build a reputation upon. Given the greater frequency of domestic choices, as discussed above, we believe this provides a reasonable test of our theory. However, we state this as a conjecture and not a hypothesis because opportunities for reputation building will vary across real world settings.
Finally, the thresholds identified in Result 1 suggest that there can be domestic choices that are so low in cost similarity and/or salience that they have no reputational effect. We do not know where these thresholds are located in real life or whether any domestic choices truly fall below them. We offer the following conjecture to establish the range of domestic choices that fall within the scope of our theory: Conjecture 3: Hypothesis 1 will be unsupported for domestic choices with very low cost similarity and salience.
To be clear, where we validate Conjecture 3 is more important than whether we validate it. If we find that domestic choices with moderate cost similarity and salience have no reputational effect, then we can conclude that our theory only applies to a small number of domestic choices. If instead, we find that choices with low cost similarity and salience have a statistically significant reputational effect, then we can conclude that our theory has broad applicability to a large number of domestic choices. Furthermore, by contrasting domestic choices that weakly affect a leader's reputation with choices that do not affect a leader's reputation, we can say that the thresholds for cost similarity and salience lie somewhere between these choices.

MAIN EXPERIMENT
We test our hypotheses and conjectures using a conjoint experiment. In our experiment, respondents were told about a hypothetical leader's past domestic and international choices and then asked to evaluate how likely the leader is to use force in an imminent international crisis. The range of domestic choices that might influence a leader's international reputation is vast, and we do not attempt to identify all relevant choices here. Rather, we seek to validate our hypotheses and conjectures with some potentially relevant choices and establish approximate upper and lower bounds for the effect of domestic choices on international reputations. To do this, we focus on two choices that we expect to be high in both cost similarity and salience and two choices that we expect to be lower on both dimensions. To contrast our theory with prior research, we also include information about the leader's choice in a past international crisis. Table 2 shows the four domestic choices we focus on and summarizes how we classify their cost similarity and salience. We argue that both coming to power in a coup and violent repression of protesters score high on both dimensions. We classify a coup attempt as involving costs similar to those of fighting in an international crisis because both choices put the leader's safety and political position at risk. We classify coups as highly salient because the potential benefits (national leadership and power) and costs (death or imprisonment) are considerable. Prior research shows that leaders who enter office irregularly are likely to be more ambitious, risk tolerant, and casualty tolerant (Colgan 2013;Horowitz, Stam, and Ellis 2015;Kim 2018). We also classify using force against domestic protesters as having costs similar to those of war because both involve human casualties. We classify how leaders respond to protests as highly salient because protests threaten a leader's hold on power. Prior research shows that protests can lead to regime change and have other important effects (for a summary, see Hale 2013). We expect these two choices to establish the approximate upper bound for how much domestic choices can influence international reputations. We conjecture that these choices could affect international reputations as much as past crisis behavior does.
At the other extreme, we classify giving a speech at a university experiencing scandal as low on both dimensions. We view it as low in cost similarity because this choice does not involve political negotiations or violence. Similarly, we classify the salience as low because the consequences of giving a moderately controversial speech are likely to be minimal. This is an example of a The Domestic Sources of International Reputation domestic choice that we conjecture falls in the shaded area of Figure 1, where cost similarity and/or salience is too low for the choice to influence the leader's reputation.
In the middle, we argue that standing firm in negotiations about a flagship domestic policy scores moderately on our two dimensions. It involves costs similar to those of conflict because both situations involve tense, multiparty negotiations that can lead to policy failure. Failure involves public and elite disapproval and associated audience costs. However, the cost similarity is only moderate because domestic policy bargaining does not involve violence. We classify the salience as moderate as well because leaders often care deeply about flagship domestic policies, but they are unlikely to care about policy as much as about gaining or retaining power or threats to their personal safety. Therefore, we expect the domestic policy negotiation treatment to have some reputational effect but smaller than that for the coup and protest treatments.
As described in our preregistration documentation (available at https://osf.io/uwdjh), we will infer very strong support for Hypothesis 1 if staging a coup, protest repression, and standing firm in domestic policy negotiations each have a positive and statistically significant effect on expectations of a leader's willingness to fight. We will infer strong support for Hypothesis 2 if staging a coup and protest repression each have a significantly larger reputational effect than both standing firm in domestic policy negotiations and giving a controversial speech.
We will infer strong support for Conjecture 1 if any of the domestic choices in Table 2 have a reputational effect that is statistically indistinguishable from fighting in a past international crisis, and we will infer strong support for Conjecture 2 if the three domestic choices that we expect to be independently informative (i.e., staging a coup, protest repression, and standing firm in domestic policy negotiations) jointly have a reputational effect that is statistically indistinguishable from or significantly greater than that of fighting in a past international crisis. 16 Finally, if we find a significant effect of standing firm in domestic policy negotiations but not of the speech, we will conclude that the minimum cost similarity and salience thresholds for a domestic choice to have informational value lie somewhere in between these two choices. This would lead us to infer strong support for Conjecture 3.
We used a fully randomized ratings-based conjoint design (Hainmueller, Hopkins, and Yamamoto 2014). We gave subjects five tasks. For each task, respondents were shown a profile of a leader's past domestic and international choices and asked to rate the likelihood that the leader would stand firm in a hypothetical international crisis. Respondents viewed leader profiles detailing all four domestic choices discussed above as well as information on whether or not the leader fought in a past international crisis. Unlike many conjoint designs in political science, we showed respondents one profile per task instead of two. We used a single profile because states have one leader at a time, and opponents must assess the resolve of that leader in a crisis. 17 Prior work has used single-profile designs to investigate attitudes toward immigrants (Hainmueller, Hangartner, and Yamamoto 2015) and when the public understands violent acts as terrorism (Huff and Kertzer 2018).
For each conjoint task, respondents viewed four domestic choices and one international choice simultaneously. This captures the "noisiness" of the domestic sphere in which domestic choices are more common than international crises. Furthermore, as information about each choice is independently randomized, the experiment is realistic in the sense that most subjects saw a combination of choices that provided conflicting evidence about resolve. Thus, subjects had to decide which choices to weight more in their evaluation. If there is a limit to how much weight observers are willing to give to domestic choices as a whole, then showing multiple domestic choices simultaneously may also create bias against finding support for our theory. Thus, we present a hard and realistic test for our theory. 18 We conducted our study in June 2021 (N = 1,878). 19 We recruited respondents through Lucid, an online survey platform. 20 Prior work has shown treatment effects using Lucid similar to those that occur with other commonly used online platforms (Coppock and McClellan 2019). To address concerns about inattentiveness among online respondents (Aronow et al. 2020), we retained only respondents who passed a pretreatment attention check. 21 As in many survey experiments in international relations, we asked members of the public to stand in for difficult-to-obtain samples of elites (e.g., Cebul, Dafoe, and Monteiro 2021;Kertzer 2016;Lupton 2020;Yarhi-Milo 2018). Despite potential concerns about external validity, studies that have directly compared results in elite and public survey samples have generally not found much difference Yarhi-Milo, Kertzer, 16 In our preanalysis plan, we include tests of both Conjecture 1 and Conjecture 2, but we treat them as alternate interpretations of the same conjecture. Here, we split them out for greater clarity. This also means that Conjecture 2 in the preanalysis plan has been relabeled as Conjecture 3 here. 17 Two profiles are appropriate when two-way comparisons are plausible, e.g., voting for a candidate. 18 Our experiment compares the effect of choices to stand firm or back down. We do not compare these choices with the absence of information. This keeps our setup comparable to those of previous experiments about international reputation. Also, all leaders face some domestic choices, meaning that the absence of domestic information would be unrealistic. 19 We fielded a similar pilot study in August 2020. The pilot results are shown in the Appendix. 20 See the Appendix for data on the demographic composition of our sample. 21 Subjects were asked to identify the issue in dispute in our pretreatment vignette and presented with multiple choice options. This is similar to the approach recommended by Kane, Velez, and Barabas (2020). and Renshon 2018). In the most systematic analysis of this issue, Kertzer (2022) argues that differences in traits between elite and public samples do not necessarily imply that their behavioral responses to experimental treatments will differ. Kertzer (2022, 7-8) then compares 162 treatment effects across 48 paired elite and public experiments and finds that the direction and magnitudes of the effects are statistically indistinguishable 88% of the time. Kertzer further finds that the most problematic experiments, which account for many of the differences, are representation experiments in which elites and members of the public try to assess each other's behavior or beliefs. Because our experiment does not fall into this category, Kertzer's results suggest that our use of a public sample is unlikely to be problematic. To further assure the external validity results, we include some realworld anecdotes later in the manuscript. 22

Survey Instrument
We presented respondents with a scenario about a dispute between the US and the fictional country Arcadia. The experiment had two phases. First, respondents were given one page of information about Arcadia's domestic and foreign policy history. Respondents were told that Arcadia was governed by a civilian nondemocratic ruler, yet it had a strong court system and an influential parliament. 23 Respondents were also told that Arcadia had historically been the largest military and economic power in its region but that its foreign policy ambitions had fluctuated. The detailed vignette reduces the chance of heterogeneous treatment effects (Dafoe, Zhang, and Caughey 2018) and creates realism that helps the subjects take the survey seriously.
All subjects saw the same information in the first phase of the survey. The first phase concluded by providing information about a potential foreign dispute between the US and Arcadia in the "Topaz Sea." Respondents were told Arcadia and the US disagree about the status of the Topaz Sea, an important maritime trade route located near Arcadia's coast. Most of the Topaz Sea clearly falls into international waters. However, from time to time over the last century, Arcadia has acted as if it held sovereign control over the Topaz Sea. For example, the Arcadian Navy has inspected vessels sailing through it. One Arcadian map showed all of the Topaz Sea as part of Arcadian waters. However, the map was reprinted following international backlash.
To ensure that the Topaz Sea remains neutral, the United States is considering a naval blockade that prevents any naval vessel from patrolling the waters. Some worry that Arcadia will respond by sending its Navy to permanently occupy the Sea. This could lead to a direct military conflict with the United States.
In the second phase, subjects were presented with information about the Arcadian leader's past choices in office and asked to predict his international crisis behavior. For each of five conjoint tasks, subjects were shown a leadership profile that varied randomly and independently along five dimensions (or attributes) that we summarize in Table 3. The first four dimensions in Table 3 correspond to the four domestic choices in Table 2. The final dimension relates to the leader's past choice of whether to fight in an international crisis. Below each leader profile, subjects were asked, "Would this leader send the Arcadian Navy to the Topaz Sea if the US imposed a naval blockade?" Subjects responded on a six-point scale from "very unlikely" to "very likely." The order in which the leader choices were presented varied randomly from profile to profile to address possible ordering effects. The survey instrument is available at the American Political Science Review dataverse (Goldfien, Joseph, and McManus 2022).
The treatments (or levels) are designed to represent "standing firm" or "backing down" when the leader faced a choice. Therefore, the treatments correspond to the choice that the defender (B) faces in the first period of our formal model. The subjects in our experiment step into the role of the challenger (A) in the second period of our model by evaluating the defender's likelihood of fighting based on the firstperiod choice. Therefore, the difference in beliefs between the paired treatment conditions in Table 3 corresponds to the theoretical quantity of interest, Q, from our formal model. We exploit random assignment over the sample as a whole to compute the average marginal component effects, or AMCEs (Hainmueller, Hopkins, and Yamamoto 2014), for every treatment and interpret each AMCE as the true Q. For example, the ACME for the response to protests measures the average difference in beliefs about Arcadia's resolve to fight given that a subject observed the Arcadian leader order police to violently repress protesters rather than stand down.

Results
We estimate the AMCEs for each leader choice using a preregistered ordinary least squares model with robust standard errors clustered at the subject level. We estimate the AMCEs using two versions of our dependent variable: (1) the original six-point scale of expected resolve and (2) a dichotomized version of that variable which codes a leader as "likely to use force" if the subject said the profiled leader was "somewhat likely," "likely," or "very likely" to respond to a U.S. blockade with force and "unlikely to use force" otherwise. An advantage of the dichotomized dependent variable is that it puts the AMCEs on a probability scale. The results using these two dependent variables are substantively similar, and we report results for both. As indicated in our preregistration document, we focus on the dichotomized variable for ease of exposition. 22 In addition, the Appendix shows that the results of our main study are robust in subsamples of our respondents with the most elite characteristics. 23 We used a nondemocratic country to make the treatments more plausible. Autocracies also provide a good test case for our theory because autocratic leaders have greater control of foreign policy.
The left-hand panel of Figure 2 displays the AMCE coefficients and 95% confidence intervals from a model using our six-point dependent variable. The right-hand panel displays the coefficients using the dichotomized dependent variable. These coefficients were estimated in models without demographic controls. These results, and similar models with demographic controls, are presented in tabular form in the Appendix. The AMCEs represent the average causal effect of the stand firm choice for each attribute compared with the back down base category while accounting for the effects of the other attributes by averaging over them.
From these plots, we infer strong support for both hypotheses. Hypothesis 1 proposed that observers will have higher estimates of a leader's willingness to fight in an international crisis after witnessing the leader stand firm domestically than after witnessing the leader back down domestically. Figure 2 shows that this is true for each of the three domestic choices that we classified in Table 2 as being moderately or highly salient and similar to an international crisis. According to the righthand panel of Figure 2, standing firm in a domestic policy negotiation increased the probability that a leader would be judged likely to use force by 7 percentage points. When a leader came to power by violently overthrowing their predecessor, it increased the probability that the leader would be judged likely to use force by 11 percentage points. A leader's response to domestic protests had the most striking effect. Using force against demonstrators increased the probability that a leader would be considered likely to use force against the United States by 24 percentage points. Furthermore, even giving a controversial speechdeliberately included as a low cost similarity, low salience domestic choice-had a small but statistically significant effect using our binary dependent variable. These results strongly support H1 and illustrate that a broad range of domestic choices can influence a leader's reputation for resolve.
Hypothesis 2 proposed that highly salient and similar choices will have a larger effect than will choices that are lower along these two dimensions. This led us to expect that coming to power in a coup and using force against protesters would each have a larger effect than either standing firm in domestic policy negotiations or giving a controversial speech. This is indeed what we find. The coefficients for coups and protest repression are larger than for the other domestic choices. Indeed, in percentage-point terms, repression of protesters had three times the effect of bargaining hard domestically and 10 times the effect of giving a controversial speech. Coming to power in a coup had 1.5 times the effect of bargaining hard domestically and five times the effect of giving a controversial speech. Preregistered linear hypothesis tests, presented in the Appendix, confirm that these differences are statistically significant. We also note a statistically significant difference in the effect sizes of standing firm on domestic policy (moderate cost similarity and salience) and giving a speech (low cost similarity and salience). Although we did not preregister this comparison, it supports our expectation that marginal increases in cost similarity and salience will marginally increase the informational value of domestic choices.
Our first two conjectures examine the relative effect of domestic and international choices. We infer strong support for Conjecture 1: a single domestic choice can have the same (i.e., statistically indistinguishable) Entering office via a coup The current leader of Arcadia came to power after killing the previous leader in a coup.
The current leader of Arcadia was appointed following the natural death of the previous leader.

Using force against domestic protesters
A few years ago, nation-wide protests broke out across the country. The leader instructed police to use violent repression -2,000 civilians died.
A few years ago, nation-wide protests broke out across the country. The leader instructed police to stand down-no civilians died. Standing firm in negotiations over a flagship domestic policy This leader's major domestic policy initiative was a health reform plan. However, rival officials refused to implement it unless concessions were made. This leader refused to make concessions.
This leader's major domestic policy initiative was a health reform plan. However, rival officials refused to implement it unless concessions were made. This leader made concessions. Proceeding with a controversial speech After a bribery scandal arose at Arcadia University, the leader proceeded with a planned graduation speech there.
After a bribery scandal arose at Arcadia University, the leader canceled a planned graduation speech there.

International choice
Stand firm treatment Back down treatment

Fighting in a previous international crisis
When Arcadia's northern neighbor challenged Arcadian control over the Greywall province, this leader did not make territorial concessions and sent the military.
When Arcadia's northern neighbor challenged Arcadian control over the Greywall province, this leader made territorial concessions and did not send the military.
Note: Italics represent varied text. Treatments are copied verbatim from the survey instrument.
reputational effect as a past international crisis choice does. Figure 2 shows that violently repressing protesters is just as informative as fighting in a past crisis. Both behaviors had a similar effect on respondent beliefs that the leader would stand firm in the present crisis. Using force against protesters increased the probability a leader would be judged resolved by 24 percentage points, and using force in a previous international crisis increased it by 26 percentage points. Importantly, these effects are statistically indistinguishable. Our ability to statistically distinguish the effect sizes of all of our domestic choices from one another shows that the experiment was well powered. It is illuminating that we could identify a significant difference between the protest treatment and all other domestic treatments but not between the protest and past international crisis treatments. Therefore, we can be confident that the protest and past crisis treatments had very similar effects on international reputation.
Using a preregistered linear hypothesis test presented in the Appendix, we also find strong support for Conjecture 2: the cumulative effect of the protest response, domestic bargaining behavior, and ascent to power is significantly larger than that of past international crisis behavior. As noted in our informal theory, leaders face more domestic choices than they do international crises. Therefore, this cumulative effect gives us some plausible insight into the importance of domestic choices in the aggregate.
Our final conjecture examines the empirical domain over which our domestic theory of reputation applies. Conjecture 3 states that Hypothesis 1 will be unsupported for domestic choices with low cost similarity and salience. Under our preregistered design, which focused on the binary dependent variable, we do not find support for it. The effect of giving a moderately controversial speech on expectations about a leader's resolve is small but statistically distinguishable from Note: Following a pretreatment vignette that described a fictional country, Arcadia, and a militarized crisis between it and the United States, subjects were presented with randomized information about the past domestic and international choices of several possible leaders of Arcadia. For each leader, subjects indicated on a six-point scale how likely they believed the leader was to use force in Arcadia's dispute with the United States. The left-hand panel presents the average marginal component effect (AMCE) of each leader choice using the original six-point scale, whereas the right-hand panel presents the AMCEs using a dichotomized version of the variable indicating whether or not the subject believed the leader was at least "somewhat likely" to use force.
zero. 24 However, we note that the effect is not statistically significant when using our six-point dependent variable; we find support for Conjecture 3 using this specification. Mixed results for the speech treatment (and significant results for all else) provide evidence that giving a modestly controversial speech sits just at or below our cost similarity and salience thresholds. In the context of our other results, this finding suggests that a wide variety of real-life domestic choices can influence a leader's international reputation. Therefore, the domain of our theory is quite large. Indeed, in certain contexts seemingly unimportant leadership choices-such as whether to go through with a speech-seem to influence international reputations on the margins.

Postconjoint Questions: Mechanisms and Related Debates
Our information-based theory reveals a new connection between leaders, domestic politics, and crisis behavior. To further explore the mechanism behind our results and to contrast it with other strands of literature in international relations and comparative politics, we posed postconjoint questions to subjects. Some subjects were asked open-ended questions regarding the influence of the Arcadian leader's domestic choices on their beliefs about the likelihood that the leader would use force. Many subjects expressed, in their own words, an inferential logic consistent with our theory. For example, one subject wrote that "if the [Arcadian] leader is not willing to make concessions, will not try to use nonviolent methods to deal with protesters and negotiate fairly with other countries, they are likely to engage in conflict." Another wrote that "if the Arcadian leader made positive decisions for its people, it would be less likely to send military force against the U.S." Additional examples can be found in the Appendix.
Other subjects were asked to rate their agreement with distinct theoretical logics that could connect coming to power in a coup or using force against protesters with international resolve (see Appendix). One line of research suggests that domestic institutions both constrain and incentivize leaders to make specific choices during crises (e.g., Bueno de Mesquita et al. 1999;Putnam 1988). In fact, scholars have identified a connection between domestic constraints and international crisis behavior in regimes at risk of coups and protests (Belkin and Schofer 2005;Ciorciari and Weiss 2016;Talmadge 2015). Although several features of our design address these issues (e.g., we told respondents that Arcadia was politically stable), it could be the case that information about how the leader came to power or responded to protests implied information about domestic politics and that this influenced subject responses. To address this, we asked subjects about these different mechanisms. We find that about three quarters of respondents expressed agreement with the logic of our mechanism: coming to power in a coup or using force against protesters revealed information about the leader's disposition. 25 Another line of research suggests that what a leader did before entering office shapes their foreign policy, including willingness to use force (e.g., Fuhrmann 2020; Horowitz, Stam, and Ellis 2015;Horowitz et al. 2018). If this is true, then biographical information should affect assessments of resolve. The Appendix reports the results of another postconjoint question in which we asked subjects to evaluate the informational value of other domestic choices made by a leader in office as well as aspects of a leader's pretenure biography. We find that a leader's choice to execute his sister over a personal dispute raises estimates of resolve more than does previous military experience. We also find that a leader's choice to end antidiscrimination policies or restrict immigration raises estimates of resolve more than a leader's prior business experience does. This is not to say that prior experiences do not matter but rather that domestic choices while the leader is in office could have a greater reputational effect.
Overall, the findings from the postconjoint questions suggest that responses in the conjoint reflected our proposed dispositional mechanism. These findings also suggest that a broad set of domestic actions-not just those used in the conjoint-could have a reputational effect.

ADDITIONAL EVIDENCE
The main experiment provides strong evidence in favor of our theory. To demonstrate that these results are robust, generalizable, and plausibly influential in a policy context, we present additional evidence from a supplementary survey experiment and from important real-world cases.

Supplementary Experiment
To ensure the robustness of our main results to alternative design choices and to further interrogate our causal mechanisms, we fielded a supplementary study (N = 486) in January 2022. The experimental design of the supplementary study closely follows that of the main study but includes five important changes.
A summary of the changes is as follows. First, rather than use a fictional country, Arcadia, we described a scenario involving China in the near future. For the crisis facing the future Chinese leader, we changed the issue in dispute from the Topaz Sea to the South China Sea. For the leader's past crisis behavior, a border dispute over Greywall became a Himalayan border dispute with India. Second, we replaced the coup treatment from the main study with a treatment about the leader's response to domestic terrorism. Following a domestic terrorist attack, the leader could (1) meet terrorist demands to release prisoners, (2) order a police counterterrorism campaign while emphasizing a commitment to abide by the law, or (3) commit a clearly unethical human rights violation by ordering police to burn down a village believed to harbor terrorists. The terrorism response treatment offers insight into another domestic action that could influence leaders' international reputations and, by including ethical and unethical uses of force, may help to disentangle the degree to which assessments of leader morality influence leader reputations. It also allows us to vary cost similarity while holding salience mostly constant. Any choice of how to respond to a large domestic terrorist attack is highly salient, but the indiscriminately violent response is more similar to war.
Third, we added additional detail to the domestic policy bargaining treatment to indicate the outcome of the policy initiative when the leader refused to compromise with political rivals. Fourth, we included additional detail about China's probability of victory in different international disputes to hold power constant across crises and account for possible heterogeneous treatment effects. Finally, we used an alternative postconjoint questionnaire, asking subjects directly to rate the cost similarity and salience of the leaders' choices. This allows us to measure subject beliefs about each choice's cost similarity and salience and compare those beliefs with the average marginal component effects of those choices.
Results. We performed the same analysis for our supplementary study as we did for our main experiment. Figure 3 displays the AMCE coefficients and confidence intervals from the supplementary study, estimated in models without demographic controls. 26 The core results are consistent with our theory and the results in our main experiment. Most critically, we find strong support for Hypotheses 1 and 2. Three distinct domestic choices-the response to protests, the response to domestic terrorism, and domestic policy bargaining behavior-had a statistically significant effect on the leader's reputation for resolve. Comparing the size of the effects, choices high in cost similarity and salience (the protest and terrorism responses) have larger reputational effects than do choices that are lower in these dimensions (domestic bargaining and a controversial speech).
We also find strong support for two out of three conjectures. We find only limited support for Conjecture 1. The difference in the reputational effect of standing firm in an international crisis and the most informative domestic choice-launching an unethical domestic counterterror campaign-was distinguishable at the 95% confidence level using our binary outcome variable and borderline distinguishable using our raw, six-point dependent variable (p = 0.07). Nonetheless, the effect of domestic actions was substantial. Cracking down on protests and the unethical counterterrorism campaign each had about 80% of the reputational effect of standing firm in a past international crisis using our six-point dependent variable and two-thirds the effect using our binary dependent variable. The limited support for Conjecture 1 in the supplementary experiment contrasts with strong support for it in the main experiment (and our pilot). Taking these findings together, we conclude that the context of a domestic choice influences the size of its effect relative to that of a past international crisis. For example, we conjecture that the size of the protests, the scope of repression, the importance of the issue under protest, and the number of prior domestic protest episodes will all affect whether a leader's protest repression choice matters as much as past international crisis behavior. In some, but not all, contexts the reputational effect of protest repression can be as large as past international crisis behavior.
We find strong support for Conjecture 2: the cumulative effect of three informative domestic actions (protest repression, unethical counterterrorism, and standing firm on domestic policy) has a significantly larger effect on the Chinese leader's reputation for resolve than does China's behavior during a past international crisis. We also find support for Conjecture 3. Consistent with our expectations and the results from our main experiment, we find that the choice to give a modestly controversial speech did not have a reputational effect distinguishable from zero.
The newly introduced domestic terrorism response treatment demonstrates the existence of an additional domestic choice that can influence a leader's international reputation. It also helps to illustrate the independent effects of cost similarity and salience and to disentangle different dispositional attributes that connect a leader's domestic choices to their international reputation for resolve. We find a significant reputational effect when a leader stood firm against terrorist demands using targeted, lawful force. However, we find an even stronger effect when the leader ordered indiscriminate, unethical, and unlawful violence. This illustrates how the amount of learning increases with cost similarity, even while salience is held mostly constant. It also suggests that in addition to previously recognized traits such as willpower and concern about honor, a lack of ethics is another dispositional trait that contributes to estimates of resolve to use military force.
The Appendix shows the results of the supplemental study's postconjoint questions, which asked subjects to score different choices along our two dimensions: cost similarity and salience. There are several notable results. First, the average subject scored the cost similarity and salience of each event in a way that is broadly consistent with the coding of cost similarity and salience we used to design our experiments. This increases our confidence in the deductive reasoning we used to make effect-size predictions for the main experiment. Moreover, the average marginal component effects from the supplementary experiment closely match the order of the cost similarity and salience scores, further 26 The results with and without controls are presented in tabular form in the Appendix.
increasing our confidence in the support for Hypothesis 2. Among domestic actions, the protest and domestic terrorism responses were rated highest in terms of salience and cost similarity and had the largest reputational effects. The choice to proceed with or cancel a potentially controversial speech scored very low on both dimensions and had no effect. Domestic bargaining behavior was in the middle for salience and cost similarity and had a moderate effect on reputation for resolve. Second, we find that a leader's response to a domestic terror attack is more salient than a response to an international crisis. This fits with our argument that domestic choices can be more salient than international crises. In addition, a leader's response to protests is nearly as salient as crisis behavior and statistically indistinguishable at the 95% confidence level.
Third, we find that the cost similarity and salience scores for individual respondents are not strongly correlated (i.e., respondents who say the protest response is highly salient do not necessarily say that it is highly similar). Together with the results of the terrorism treatment, this increases our confidence that cost similarity and salience are independent parameters that do not covary perfectly. The extent of covariance is probably influenced by subtle nuances that surround each domestic choice. Although we do not provide this context in the experiment, individual subjects may bring their own assumptions to it, causing this variation. Overall, the supplementary conjoint results demonstrate the general robustness of our main results to alternative designs and settings. Furthermore, subject ratings of the cost similarity and salience of domestic actions in the supplementary study provide strong validation for our theoretical approach and interpretation of the empirical results.  Note: Following a pretreatment vignette that described a future hypothetical crisis between China and the United States, subjects were presented with randomized information about the past domestic and international choices of several possible leaders of China. For each leader, subjects indicated on a six-point scale how likely they believed the leader was to use force in China's dispute with the United States. The left-hand panel presents the average marginal component effect (AMCE) of each behavior using the original six-point scale, whereas the right-hand panel presents the AMCEs using a dichotomized version of the variable indicating whether or not the subject believed the leader was at least "somewhat likely" to use force.

Anecdotal Evidence
To illustrate that the empirical pattern we identified travels to real-world settings, we provide several anecdotes of cases where leaders' domestic choices influenced their international reputations. First, as noted earlier, there is evidence that Soviet officials were impressed with Reagan's firing of the air traffic controllers (Morris 1999, 448, 792-3). Belief that this choice enhanced Reagan's international reputation became so widespread among Republicans that Wisconsin Governor Scott Walker cited his own union showdown as a foreign policy credential 34 years later (Rucker 2015). Second, Jerrold Post, a psychologist who spent over 20 years profiling foreign leaders for the CIA, testified before Congress in 1990 that Saddam Hussein's ascent to power via a coup revealed information about his international resolve. Reflecting on Saddam's swift execution of a coconspirator in the coup, Post said that the "act was a paradigm for the manner in which Saddam has rewarded loyalty and adhered to commitments throughout his career. He has a flexible conscience: commitments and loyalty are matters of circumstance, and circumstances change. If an individual, or a nation, is perceived as an imminent threat, no matter how loyal in the past, that individual or nation will be eliminated violently without a backward glance" (Post 1991, 281, emphasis added). It is illuminating that Post focused on Saddam's domestic choices given that Saddam had recently fought a brutal international conflict with Iran.
There are also more modern examples. In 2017, President Trump cited "starving and killing his own people" as evidence that Kim Jong-un was a madman (Stevens 2018), and National Security Advisor H. R. McMaster similarly cited North Korea's "unspeakable brutality against its own people" as evidence that Kim was undeterrable (Kaplan 2017 under a nuclear overhang, will lead to massive bloodshed far-reaching consequences for the world." In this quotation, Khan uses a mix of domestic and international choices by Modi to argue that India is highly resolved and even aggressive. Like the Reagan anecdote, this illustrates how our theory applies to democratic as well as autocratic leaders. Finally, there are examples where domestic choices suggesting lower cost tolerance led to lower estimates of resolve. When Nikita Khrushchev began the process of de-Stalinization in the Soviet Union, U.S. officials saw this as evidence that Soviet foreign policy would be less resolved. The U.S. embassy in Moscow assessed that the Soviet leadership still sought to compete with the US, but under a "more sensible set of ground rules," in which the Soviets recognized "that modern war, i.e. destruction of Soviet resources and base of power, would advance neither their national interests nor Marxism" (U.S. Department of State 1989). Therefore, whereas Stalin was seen as willing to go to nuclear war in a crisis, Khrushchev was not.

CONCLUSION
We have argued that leaders' international reputations for resolve are influenced by domestic choices that have little to do with foreign policy. We identify two dimensions that determine how much domestic choices influence international reputations: the degree to which the costs involved are similar to the costs of war and the level of salience relative to foreign policy. When the costs are very similar and the salience is high, our model predicts that domestic choices have the potential to be equally or more influential in determing a leader's international reputation for resolve than past international crisis behavior is.
Our main and supplementary experiments support our main predictions. First, we find that a variety of domestic choices-including repressing protests, seizing power in a coup, bargaining hard over domestic legislation, ordering domestic counterterror operations, and possibly even making a controversial speech-affect leaders' international reputations. Second, we find that choices with higher cost similarity and salience have a greater effect. Our main experiment suggests that at least one domestic choice-violently repressing domestic protests-can be equal in influence to an international crisis, although our supplementary experiment results indicate that this is not always the case. Although we do not find empirical evidence that any single domestic choice is more informative than international crisis behavior, it remains a theoretical possibility that some domestic choices might be. Furthermore, we find in both experiments that the cumulative effect of domestic choices has more influence on international reputation than a past international crisis does. Given the greater frequency with which leaders make domestic choices, it is plausible that domestic choices have a greater overall influence on international reputations than do foreign policy choices. Indeed, leaders who have recently entered office and not yet experienced any international crises may have an international reputation that is based solely on their domestic choices.
Our findings have important theoretical and policy implications. We contribute to the reputation literature by providing new insight on the sources of reputation. We also provide a new perspective on the relationship between domestic politics and international conflict and the important role of individual leaders in international relations. Overall, our research suggests that the development of international reputations is more complicated than previously recognized, as a leader might be able to obtain an international reputation for resolve without actually fighting on the international stage. Conversely, a leader who has stood firm in past international crises may nonetheless find their reputation weakened by less-resolved domestic choices. This carries the policy implication that leaders should consider the effect of their domestic choices on their international credibility.
Although our experiments focused on autocratic countries, the basic logic of our theory does not depend on regime type, and some of the domestic choices with significant reputational effects-such as standing firm on domestic policy or an ethical response to domestic terrorism-commonly occur in democracies. Moreover, although democratic leaders are generally less likely to do things like repress protests, it would probably have a large reputational effect if they did. The anecdotes about Reagan and Modi illustrate the reputational effect of domestic choices by democratic leaders. Future research could explore the reputational effect of additional domestic choices that are more commonly made by democratic leaders.
There are several other possible directions for future research. Future work could create a more comprehensive typology of different domestic choices that are more or less likely to influence international reputations. It could also analyze whether there are systematic variations in how different observers interpret domestic behavior. In addition, although we have focused on cost similarity as a condition for reputation formation, the logic of our formal model suggests that there could be a similar role for benefit similarity. Certain dispositional traits, such as concern with honor, may affect the perceived benefits of standing firm both domestically and internationally. Future research could explore this more. Finally, instead of only focusing on reputations for resolve, future work could look at the effect of domestic choices on other types of behavioral expectations.

SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit http://doi.org/10.1017/S0003055422000855.

DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi. org/10.7910/DVN/U5IESG.