The idea that democracies have an advantage over autocracies in signaling their intentions is now axiomatic. Audience costs, or the domestic punishment that leaders would incur for backing down from public threats, are thought to increase leaders' ability to convey their preferences credibly during military crises.1
Fearon 1994.
See Fearon 1994; Eyerman and Hart 1996; Partell and Palmer 1999; Gelpi and Griesdorf 2001; and Prins 2003. Schultz 1999 also presents evidence consistent with that hypothesis. Slantchev 2006, in contrast, argues that audience costs are higher in democracies only when press freedom is strongly protected.
The conventional wisdom, however, rests on an underestimate of the vulnerability of leaders in nondemocratic regimes.3
I will use the terms nondemocratic, authoritarian, autocratic, and dictatorial interchangeably, though some scholars attribute more specific meanings to these terms.
Bueno de Mesquita et al. 2003.
Together, these insights about coordination, elite incentives, and visibility have important implications for understanding variation in regimes' abilities to make credible threats and promises. Tests of the effects of regime type on foreign policy must therefore take into account differences between autocracies. I show that existing empirical support for the claim that democracies have a signaling advantage in military disputes results from treating a heterogeneous set of autocracies as undifferentiated. When the group of authoritarian regimes is disaggregated, democracies are not more successful in signaling their resolve than most types of authoritarian regimes. The exceptions are “personalist” regimes and certain types of monarchies, in which the leader has the means to impede elite coordination, as well as new democracies and unstable nondemocracies, where the threat of removal is not observable to outsiders.5
See Geddes 2003. Chehabi and Linz 1996 describe a similar type of regime, which they term “sultanistic.”
I begin with a theoretical discussion of the necessary conditions for generating audience costs. I then argue that autocratic regimes meet these requirements when elites have incentives and ability to coordinate to punish the leader and the potential for punishment is visible to foreign decision makers. Statistical analysis of militarized interstate disputes strongly supports the hypothesis that democracies are not better at generating audience costs than most autocracies.
The Logic of Audience Costs
The audience costs proposition suggests that states can send informative signals about their resolve by making public threats in international crises.6
Because leaders could suffer domestic consequences for making a threat and then not carrying it out, they are able to create potential domestic consequences for backing down. This in turn gives their threats greater credibility.Since the concept of audience costs was first articulated by Fearon, scholars have assumed that democracies have an advantage in generating audience costs, and hence an advantage in signaling resolve.7
Fearon 1994.
See, for example, Guisinger and Smith 2002, 180. Other researchers have taken a more agnostic view, though they still group regimes according to the level of democracy; see Chiozza and Goemans 2004.
See Eyerman and Hart 1996; Partell and Palmer 1999; Gelpi and Griesdorf 2001; and Prins 2003. In addition, Schultz's finding that democracies are less likely to be resisted in international crises can be interpreted as evidence in favor of higher democratic audience costs, though Schultz presents a distinct theoretical mechanism where resolve is revealed through public party competition; see Schultz 1999 and 2001a.
But the possibility that authoritarian regimes exhibit predictable variation in their ability to generate audience costs, and moreover, that democracy is not necessary for generating audience costs, merits further attention. Elections and democratic institutions are only one way in which domestic groups can coordinate to hold leaders accountable. In order to reevaluate prevailing arguments about how audience costs vary across political systems, it is helpful to clarify the logic of audience costs.
A leader's ability to generate domestic political costs is influenced by three central factors. First, audience costs require that a domestic political audience has the means and incentives to coordinate to punish the leader. Second, domestic actors must view backing down after having made a threat as worse than conceding without having made a threat in the first place. Third, outsiders must be able to observe the possibility of domestic sanctions for backing down. Nondemocratic states vary greatly with respect to these three variables.
Domestic Actors Can and Will Coordinate to Sanction the Leader
The first factor influencing audience costs is whether a domestic audience can and will punish the leader for backing down from a threat, the ultimate punishment being removal from office. Fearon does not lay out explicitly when a domestic group qualifies as an audience, though he argues that “kings, rival ministers, opposition politicians, Senate committees, politburos, and, since the mid-nineteenth century, mass publics informed by mass media” have all counted as relevant audiences historically.10
Fearon 1994, 581.
Building on this logic, the working hypothesis has been that leaders are much more vulnerable to domestic punishment in democracies than in nondemocracies, due to the existence of self-enforcing institutions specifically designed to hold leaders accountable. In nondemocracies, in contrast, sanctioning the leader is thought to be a much riskier and costlier endeavor. International relations scholars have therefore tended to assume that autocratic leaders are largely unaccountable to domestic groups.11
As McGillivray and Smith put it, “ousting authoritarian leaders is more costly [than ousting democratic leaders], often requiring social unrest and possibly even civil war”; McGillivray and Smith 2000, 815.
However, scholars of comparative politics have long argued that even without democratic institutions, autocratic leaders depend on the support of domestic groups to survive in office.12
The difference is that in authoritarian regimes, these influential groups usually represent fewer societal interests than in democratic regimes. This insight has been integrated into some of the recent international relations literature, though not the literature on audience costs and the ability to convey resolve.13The Costs of Coordination
In understanding audience costs, the democratic-autocratic distinction is only a rough proxy for the ability of domestic groups to sanction leaders for missteps such as backing down. In any political system, punishing a leader can be viewed as a coordination problem between individuals or groups in society.14
Weingast 1997.
The fundamental challenge facing any individual—no matter what the political regime—is how to determine whether the benefits of participating in the removal of the leader outweigh the potential costs. The first source of costs is the individual's expectation that the individual will be punished for moving to oust the incumbent. This depends on whether the individual thinks the ouster will be successful, which depends on whether the individual can learn reliable information about other individuals' intentions.
More specifically, individuals deciding whether to participate in the ouster of a leader face a strategic situation similar to the stag hunt, a classic coordination game.15
See Menaldo 2006, for a similar logic focusing on the importance of constitutions in authoritarian regimes.
The ouster will be successful only if both individuals (or, more realistically, many individuals in a multiplayer game) choose to oust. This situation most strongly resembles a coordination game rather than a prisoner's dilemma because no individual wants to be the “odd one out”; her strategy depends on her expectations about the other players' actions. If everyone else ousts, the individual prefers to oust too, because she can then capture the higher payoff from her preferred outcome of replacing the leader, as well as gaining influence under the new leadership. If the individual cannot trust that enough other players will oust, however, she is better off lying low (catching rabbits). The structure of the game also captures the idea that the payoffs to lying low alone/catching rabbits alone are higher than the payoffs of ousting alone/hunting stag alone.
Given this strategic situation, the outcome depends on the players' beliefs about what the other player(s) will do. In one-shot coordination games in which players care only about their own payoffs, and do not face any additional costs for voicing their preferences, simple communication is usually enough for successful coordination because players have no incentive to lie about their intentions. In such a situation, two people who preferred to oust would simply say so, and would then execute their plan.
In politics, however, coordination is more difficult because individuals may face external incentives to conceal their true preference. Most importantly, individuals may fear retaliation from the incumbent for voicing opposition. This fear will be heightened with increases in the leader's ability to monitor individuals and punish the disloyal. As the incumbent's ability to monitor and punish rises, individuals will find it more preferable to conceal their preference to oust—or, in the words of Kuran, to engage in preference falsification, voicing a public preference that diverges from their private preference.16
Kuran 1991. Kuran's analysis of preference falsification and revolutionary bandwagoning in the context of the Eastern European revolutions is closely related to the logic I develop here. However, my analysis is focused less on the rapidity of coordination than variation in coordination across regime types.
Throughout history, societies have used a number of different methods to limit the leaders' ability to use monitoring and punishment to impede coordination. For example, in democracies, legal protections for freedom of speech and assembly—backed by courts that protect these rights—preclude the incumbent from monitoring and punishing citizens for voicing opposition. Thus, most individuals in democracies live in an equilibrium in which they can express their preferences openly, removing the leader when enough of them agree to oust.
In nondemocratic regimes, on the other hand, the leader and elites collaborate to explicitly preclude participation by the population. They do this, of course, by making it costly for citizens to coordinate, punishing those who criticize the regime (either individually, or by forming political organizations) through imprisonment and other punishment. However, nondemocracies vary greatly in the extent to which regime insiders can coordinate to punish the leader, rendering regime elites an effective audience. While most citizens cannot challenge the leader, elites with key positions in the regime can still oust leaders if they can solve their coordination problem.17
Haber 2006.
The question now becomes to what extent the incumbent can monitor and punish regime insiders for expressing disapproval. First, regimes vary greatly in the extent to which leaders control the intelligence organs, allowing them to monitor elite opposition. For example, while Joseph Stalin, Saddam Hussein, and Augusto Pinochet used their control over intelligence to locate internal dissent, intelligence organs in other regimes are accountable to collective bodies such as juntas or politburos. The ousting of Soviet premier Nikita Khrushchev in 1964 was not carried out until its planners were certain that they had secured the cooperation of the KGB (Soviet secret police) majorities in the Central Committee and Presidium, and officials spanning the territorial party apparatus. It was the relative independence of the KGB from Khrushchev's control, and the fact that Khrushchev did not have his own police forces, that kept him from learning of the plans and punishing the plotters.18
Tompson 1991, 1108.
Regimes also vary in the extent to which the leader can punish the disloyal. While Stalin, because of his control over the secret police, was able to use force to fire, arrest, imprison, and kill officials as he saw fit, the Communist Party after Stalin's death strongly limited the power of the KGB to try and sentence the accused. After 1953, trial and sentencing were carried out by civilian bodies out of Khrushchev's and Leonid Brezhnev's exclusive control.
In addition to preventing elite coordination through monitoring and punishment, a leader may also use positive incentives to either encourage elites to inform on other elites, or to pay off elites so that their payoffs to ousting are less than the payoffs to lying low and receiving promotions and other perks. Bueno de Mesquita and colleagues suggest that this is precisely why leaders of small winning-coalition societies focus on providing private rather than public goods.19
Bueno de Mesquita et al. 2003.
In sum, regimes vary in the extent to which the leader has the power—typically backed by force—to monitor and punish the leader's peers. In regimes in which intelligence and security organs are monitored by a collective rather than an individual incumbent, the coordination dilemma between elites can be solved by simple communication. The leader will find it much more difficult to detect criticism and punish elites for voicing criticism.
The Costs of Leader Turnover to the Ruling Group
In the previous section I argued that coordination to punish leaders is possible in both democracies and autocracies, unless the leaders have the monitoring and coercive capacities to punish individuals who oppose them. A closely related proposition is that a leader's domestic vulnerabilities do not translate into accountability unless the fate of the audience is decoupled from the fate of the leader. Otherwise, the payoffs to ousting will be lower than the payoffs to lying low.
In democracies, the credibility of threats against the leader is taken for granted, since the welfare of voters—and some elites—is usually not directly influenced by the identity of the leader (though voters may have preferences over the policies of different leaders). In democracies, the overall strength of the regime has little to do with the identity of the leader: the U.S. federal government does not become less legitimate or powerful when a new president takes office.
Nondemocracies, however, vary greatly in terms of how leadership turnover affects elites' welfare. In some political systems, elites' fates are so closely tied to the fate of the leader—for example, through blood relation or because the elites have no independent base of support or power—that they will view keeping an inferior incumbent as favorable to replacing him with a new leader and risking losing office themselves. Perhaps this is why many dictators, including Saddam Hussein and Kim Il Sung, filled top offices with relatives and other loyal associates. Similarly, many of Ethiopian Emperor Haile Selassie's most trusted ministers were “plebeians” whom he had personally plucked out of the hinterland and raised to high office—supporters who therefore had everything to lose should Selassie fall from power. In all likelihood, this dimension is not separate from the leader's control of monitoring and punishment: more powerful leaders are able to exert greater control over hiring and firing, filling high office with those who depend on them completely for their livelihood.
In contrast, in other authoritarian systems, the leader has much less control over who holds high office. Most high officials do not serve purely at the pleasure of the individual leader, and can expect to continue in politics even if the leader is removed. In the Soviet Union after Stalin, party officials from across the USSR elected the Central Committee, which in turn chose the membership of the Politburo. Accordingly, when Khrushchev was ousted, nearly all top officials retained their positions, as their political careers were not tied personally to Khrushchev. Similarly, in military regimes, military hierarchies rather than personal ties play an important role in promotion to and maintenance of high office. In the Argentine military junta between 1976 and 1983, junta members ousted three separate incumbents. While the regime was able to exert its will on the population indiscriminately, no single leader was ever able to eliminate rivals from the regime and stack it with cronies lacking incentives to oust him.20
Linz and Stepan 1996, 190–94.
Domestic Audiences Disapprove of Backing Down
The second factor influencing audience costs concerns how audiences view leaders who back down from threats. For public threats to be informative through an audience costs mechanism, backing down must be costly for the leader.
There are at least two plausible reasons why domestic audiences might impose audience costs on leaders who back down. The first reason is that bluffing hurts the leader's international reputation, and hence the leader's future ability to bargain effectively; it is therefore in the audience's interest to replace the leader and regain credibility.21
Even actors who actually supported the decision to back down will, ex post, have incentives to remove leaders if they anticipate that this will help the country bargain more effectively in the future. An alternative reason that audiences may disapprove is that a failed bluff conveys information about the leader's competence more generally.22Smith 1998.
Tomz 2007.
For the purposes of predicting variation in audience costs across political systems, then, the question is whether members of domestic audiences in democratic regimes are on average more likely to value credibility or competence than audiences in various types of autocratic regimes. There is no clear theoretical reason that this would be the case. Therefore, the second key precondition for audience costs is likely to be present not only in democracies but also in autocratic regimes.
Outsiders Can Observe the Leader's Insecurity
Finally, the last requirement for sending credible signals via audience costs is that the target state perceives that the leader could face domestic sanctioning. Here, the critical question is whether politics are stable enough for outsiders to determine whether the leader faces an accountability group in practice. In regimes with new democratic institutions such as parliaments or elections designed to hold the leader accountable, it remains unclear whether the leader and domestic groups will play by the official “rules of the game” until the rules have been tested. Similarly, in unstable nondemocratic regimes, observers will have trouble discerning whether the leader shares control of the state apparatus with elites, or rules alone. Thus, leaders of states that have recently undergone institutional change—whether nominally democratic or not—will find it difficult to publicly and credibly jeopardize their political futures.
In stable regimes, in contrast, foreign decision makers can typically determine whether the leader rules alone, or is plausibly accountable to parliament, voters, or groups of elites such as politburos and juntas. Similarly, they can see whether the leader conducts purges and repeated firings of high-level officials, or is forced to accept the existence of potential rivals in government. In the Khrushchev-era Soviet Union, for example, Western media ran a series of articles detailing Khrushchev's political insecurity both before and after events such as the Cuban Missile Crisis.24
Published before the Cuban Missile Crisis, “Is Mr. Khrushchev Pressed By Military Clique?” (London Times, 5 September 1961) suggests that Khrushchev was forced to listen to military influences in the elite. After the crisis, “Moscow Rallies Support for Mr. Khrushchev's Policy” (London Times, 6 November 1962) reports that Khrushchev was facing domestic criticism for removing the missiles. “Mr. Khrushchev Reported to Be Facing a Crisis” (London Times, 2 April 1963); “Mr. Khrushchev Regains Some Support” (London Times, 30 April 1963); and “Mr. Khrushchev to Keep His Job” (London Times, 20 May 1963), detail the rise and fall of the Soviet leader's political support—and imply that whether or not he kept his job was not in his own hands.
See “President Videla Is Confirmed for Second Term” (London Times, 4 May 1978, 6); Tony Emerson, “Argentina's Next President May Face Two Crises” (London Times, 6 October 1980, 5); and Patrick Knight, “Viola Replaced in Argentina by Junta Rivals” London Times, 12 December 1981).
The visibility condition described here is quite undemanding: the only requirement is that the opposing state knows that the leader faces a real probability of domestic sanctioning. Recall that the “audience cost” does not arise because domestic audiences disagree with their leaders' policy. Rather, the cost is imposed because leaders either hurt their international credibility or reveal their incompetence. For example, for democracies to have higher audience costs on average does not require that outsiders read public opinion polls about the government's policy statements. Rather, threats by democracies are credible because outsiders observe that domestic groups could punish the leader. Similarly, outsiders do not need information about authoritarian elites' policy preferences as long as they know that elites have the means and incentives to punish the leader if necessary.
This relatively permissive visibility condition contrasts with alternative theories predicting that democracies are better at signaling resolve, such as Schultz's theory about the information conveyed by opposition parties during crisis bargaining.26
Schultz 2001a.
Like the idea that audience costs are higher in democracies, Schultz's theory predicts that on average, threats issued by democratic challengers should be more credible than threats issued by nondemocratic challengers.
Variation in Audience Costs Across Autocratic Regimes
I argue in the section above that there are three prerequisites for generating audience costs: whether domestic actors have the means and desire to coordinate to oust the leader; whether outsiders can observe that an audience can punish the leader; and whether the audience views backing down negatively. More specifically, elites will have greater incentives to coordinate if the leader cannot monitor and punish defection through personal control of intelligence and security organs and does not control political appointments. In turn, foreign observers can infer the possibility of audience costs if the regime is relatively stable, allowing them to observe whether elites have coordinated in the past. Moreover, I argued that nondemocratic audiences have no reason to view backing down more favorably than democratic audiences. The logic implies that a leader's ability to generate audience costs does not depend on features of liberal democratic regimes, such as an inclusive electorate, voting mechanisms, or free speech. As long as the audience knows that the leader made a threat and backed down—and this would be difficult to disguise from regime insiders even in the absence of a free news media—freedom of speech and the political engagement of the masses are not required.
The proper way to predict variation in audience costs, therefore, is to distinguish between regimes according to the remaining conditions noted above: whether domestic elites can and will coordinate to oust the leader, and whether foreign decision makers can observe that the leader faces domestic accountability. This section disaggregates regimes based on whether they fulfill the coordination and visibility criteria.
Geddes's classification of dictatorships is one source of useful data in this regard.28
Geddes's categorization has also fruitfully been used in the crisis literature, in analyses of types of dictatorships and their conflict behaviors. See, for example, Peceny, Beer, and Sanchez-Terry 2002; Peceny and Beer 2003; and Kinne 2005.
Geddes 2003, 50. A different measure that would be useful is the Bueno de Mesquita et al. 2003 measure of the size of the winning coalition, or the relative size of the group whose support is essential to the survival of the incumbent. Powerful leaders may have incentives to reduce the size of the audience they face, filling positions only with loyal supporters. A small winning coalition could therefore be evidence of a more powerful leader. However, Bueno de Mesquita and colleagues explicitly assume that military regimes have particularly small winning coalitions, so their measure would not be a direct test of my hypothesis for all regime types.
Geddes 2003, 48–49.
Fortunately, the rules Geddes used to generate her categorization of autocracies allows one to use the regime type data to test hypotheses about audience costs. Geddes codes regimes by aggregating the answers to three distinct groups of yes/no questions.31
Ibid., 225–27.
Importantly, the coding criteria for both single-party and military regimes indicate that the leader does not usually control appointments or security organs in such regimes. Single-party regimes are “those in which the party had some influence over policy, controlled most access to political power and government jobs, and had functioning local-level organizations.”32
Ibid., 72.
Ibid., 225–27.
Military regimes, according to Geddes, are “governed by an officer or retired officer, with the support of the military establishment and some routine mechanism by which high-level officers could influence policy choice and appointments.”34
Ibid., 72.
Ibid., 226.
Personalist regimes, in contrast, most closely reflect the conventional wisdom about nondemocracies: there is no domestic audience that can effectively coordinate to sanction the leader. This is for two reasons: the leader has the means to punish internal critics, and the fate of elites is intimately connected to the leader's survival in office, reducing their incentives to punish leaders. For example, among the criteria for personalist regimes are whether the leader “personally control[s] the security apparatus” and whether “access to high office depend[s] on the personal favor of the leader.” Personalist leaders therefore can discipline elites much more harshly than leaders of regimes where power is less concentrated. Not only can these leaders use their control of security organs to arrest, demote, imprison, or even kill critics but they can also use more subtle tactics unavailable to leaders who have less control over political appointments. Pinochet and other dictators often used their control over appointments to frequently rotate elites through different offices so that they did not have an opportunity to build an independent power base.
Moreover, elites in personalist regimes will find it much less appealing than elites in other regimes to remove their leader. The fate of elites is typically tied closely to that of the incumbent. As Bratton and van de Walle note, because regime insiders in personalist states are
Recruited and sustained with material inducements, lacking an independent political base, and thoroughly compromised in the regime's corruption, they are dependent on the survival of the incumbent. Insiders typically have risen through the ranks of political service and, apart from top leaders who may have invested in private capital holdings, derive livelihood principally from state or party offices.36
For elites in personalist regimes, keeping a poor leader in office is more often preferable to ousting the incumbent and risking one's own career; the leader does not face a credible threat of removal.
Finally, foreign decision makers can observe that personalist rulers face no consistent threat of punishment. For example, while followers of Soviet or Argentine politics were aware of rivalries, factions, and leadership turnovers, those interested in North Korean politics had no reason to believe that Kim Il Sung faced any real internal challenges. Instead, they learned that the “Great Leader” was the “unchallenged ruler” of his country, conducted massive purges of elites (once purging seventeen members of the twenty-six-member Politburo), and placed numerous relatives in positions of power.37
“Marshal Kim Tightens Grip on N Korea in Leadership Changes” (The Times, 18 November 1970, 7).
In addition to military, single-party, and personalist regimes, there are two classes of nondemocratic regimes not coded by Geddes meriting discussion: monarchies, and nondemocracies that do not meet Geddes's criterion of having been consolidated for three years.38
In addition, Geddes does not code post-revolution Iran and post-Soviet republics such as Turkmenistan, Tajikistan, Kyrgyzstan, Uzbekistan, and Kazakhstan. These are dropped from the empirical analysis below.
The first group of nondemocracies omitted by Geddes includes monarchies such as Jordan, Morocco, Saudi Arabia, and Iran until 1979. While Geddes does not code monarchies, and therefore does not provide information about whether the leader controls the security apparatus or appointments, scholars have argued that a crucial determinant in whether or not the leader is accountable is whether the monarch rules alone, or with the assistance of the extended ruling family.39
Herb distinguishes between dynastic monarchies—regimes in which “the family forms a ruling institution,” and nondynastic monarchies in which the ruler rules alone.40Herb 1999, 8.
Ibid., 33.
Ibid., 238.
In contrast, nondynastic monarchies tend to more closely resemble personalist regimes. Although family members within nondynastic regimes can expect that one of them will inherit the throne, they are excluded from holding important posts in the regime. Rather, the king can promote loyal followers to high positions, similar to his personalist counterparts. Moreover, leaders of nondynastic regimes such as the Shah of Iran typically have “solid control over the state and its coercive apparatus” that, according to the logic I laid out earlier, should allow them to impede coordination by elites.43
Ibid., 219. While many nondynastic regimes are “constitutional monarchies,” moreover, their parliaments are typically “little more than an arena in which politicians divided up, and fought over, the spoils of rule”; ibid., 211.
Geddes also omits country-years that do not meet her classification of a “regime,” or “sets of formal and informal rules and procedures for selecting national leaders and policies.”44
Geddes 2003, 70.
Marshall and Jaggers 2002.
Similar to the autocratic interregna described above, there are new democracies, or regimes that are democratic according to Polity but have not yet persisted for three years. Since Geddes codes regime type for authoritarian regimes that are ultimately in existence for only three years, one must be careful to treat democracies similarly. Otherwise, the “democracy” category would include a disproportionate number of young or unstable regimes compared to the autocratic categories. This final category is similar to nondemocratic interregna in that while domestic groups may sometimes be able to depose the incumbent, foreigners will find it very difficult to assess whether the new laws reflect the true rules of the game.46
For a related argument, see Mansfield and Snyder 2005.
Finally, there are mixed nondemocracies that fit none of the criteria described above: they are not stable democracies or new democracies, have not experienced regime change in the last three years, and yet their autocratic regime type was not coded by Geddes. This group of regimes includes the post-Soviet states, Iran, South Africa under apartheid, and a number of anocracies—regimes where participation is only partially regulated. While this group represents a diverse set of regimes, there is no reason to think that individual leaders have inordinate capacities to monitor and punish elite criticism in these states. Moreover, since all regimes in this category have experienced regime stability for three years or more, the leader's political insecurity should be visible to outsiders. Mixed regimes should not have a disadvantage in generating audience costs compared to other stable regimes in which elites can coordinate.
Quantitative Analysis
The previous section provides a theoretical rationale for reexamining the relationship between regime type and audience costs, instead classifying regimes according to the likelihood of elite coordination and whether this is visible to foreign decision makers. Below I present empirical tests of the predictions developed above, namely that democracies, single-party states, military regimes, and dynastic monarchies should all be able to generate audience costs. In contrast, personalist leaders and nondynastic monarchs can impede elite coordination, while nondemocratic interregna and new democracies do not meet the visibility condition. Therefore, personalists, nondynastic monarchs, and leaders of both democratic and nondemocratic regimes that have not persisted for at least three years should be significantly less able to generate audience costs than other regimes.
The strategic nature of crisis behavior presents methodological challenges when testing for the existence of audience costs. As Schultz notes, leaders have incentives to avoid precisely those situations in which one would expect to observe these costs directly.47
Schultz 2001b.
Fearon 1994.
The Militarized Interstate Disputes (MID) data set contains a record of every interstate threat or use of military force since 1816.50
An MID is coded when an initiating state uses or explicitly threatens force against a target state. Targets sometimes respond with a militarized action of their own, while other times they choose to forgo a military response. To capture whether some types of initiators encounter more resistance from their targets than other initiators, Schultz analyzes the variable recip, which has a value of 1 if the target state responded with a militarized action, and 0 if the target state made no militarized response to the challenger's threat or use of force.51 This provides an indication of whether the target was hesitant to escalate the crisis because it thought the threat was genuine. On average, one should expect that initiators with a high ability to generate audience costs should be less likely to face resistance than states with a low ability to generate audience costs.52It bears reemphasis that there are alternative mechanisms through which democracies may be able to generate credible threats. Schultz 1999 and 2001a argues that democracies generate more credible threats because public debate by opposition parties allows the government to signal its resolve more effectively. Both higher audience costs and the existence of public opposition parties imply corresponding lower rates of resistance to threats, though the model developed by Schultz would not be able to explain why single-party or other authoritarian regimes would generate credible threats since public opposition is typically banned.
Table 1 depicts a first cut at the data: targets' rate of reciprocation conditional on whether or not the challenger is a democracy.53
I follow Schultz in restricting the sample to “originator dyads,” where both states were involved in the dispute from its first day. I constructed the data set using EUGene software, version 3.1; see Bennett and Stam 2000.
Changing the democracy threshold to be more inclusive does not substantively affect the results report below.
Target reciprocation rate by regime type of challenger
This table indicates that, in both time periods, democratic challengers are met with lower rates of resistance than challengers of other regime types. In both periods, a chi-square test indicates that one can reject the null hypothesis that there is no relationship between democracy and the rate of reciprocation of target states.
The next step is to analyze rates of target reciprocation after disaggregating autocratic regime types. I first generate variables according to Geddes's coding of single-party, personalist, and military regimes. I also generate categories for dynastic monarchy, nondynastic monarchy, nondemocratic interregna, and new democracy. All remaining regime years are coded as mixed nondemocracy; this category includes all country-years that are not democratic, interregna, or monarchy, but are nevertheless omitted from Geddes's coding.55
“Mixed nondemocracy” include countries that are not considered democratic according to Polity IV but are also neither monarchies nor coded by Geddes. In practice, this category includes the post-Soviet republics, Iran after the fall of the Shah (sixty-seven observations), South Africa under apartheid (twenty observations), Russia after the fall of the Soviet Union (twenty-two observations), and a host of countries that while not considered democratic by Polity IV, were not autocratic enough to merit inclusion in Geddes's study of authoritarian regimes. Of the 1,582 post-1945 MIDs in the sample, 205 involve challengers that were mixed nondemocracies.
Distribution of regime types worldwide by country-year, 1946–99
Table 3 provides a different analysis of the data, depicting the rates of target reciprocation when authoritarian regimes are disaggregated. Recall that targets should resist at lower rates against challengers with higher ability to generate audience costs. Table 3 reveals that although democratic challengers meet lower rates of resistance when compared to all nondemocracies, the supposed democratic advantage disappears when nondemocracies are differentiated from each other. Reciprocation rates of single-party regimes are nearly as low as those of democracies, while the reciprocation rates of mixed nondemocracies are actually the lowest of any regime type. Military regimes and dynastic monarchies follow. Nondynastic monarchies, personalist regimes, new democracies, and nondemocratic interregna feature the highest reciprocation rates, suggesting that when these regimes make challenges, their threats are perceived to be less credible by the target states. Preliminary evidence therefore confirms that there is significant variation in nondemocracies' ability to signal, and it is consistent with the hypothesis that democracies do not have an advantage over nondemocracies in which elites can visibly coordinate.
Target reciprocation rates by regime type of challenger, 1946–99
The next step is to ensure that the relationships suggested in Table 3 are not due to confounding variables such as relative power, military capabilities, geographic proximity, or the issues at stake in the dispute. A binary dependent variable model such as logistic regression allows one to control for variables correlated with regime type that may also affect reciprocation rates. Table 4 reports the results of a logistic regression of target reciprocation on various predictor variables. Along with the regime type variables, I include several measures of the distribution of power within the crisis dyad, including the initiator's share of capabilities and whether each side is a major or minor power; a variable indicating whether the states are contiguous on land or across less than 400 miles of water; whether the two states are part of a formal alliance; a weighted measure of the similarity of the two states' alliances; how closely aligned each state is with the current leader of the international system (to give an indication of each state's evaluation of the “status quo”); and finally, the issues at stake in the dispute.56
All variables generated using EUGene software, version 3.1, Bennett and Stam 2000.
Reciprocation by regime type of challenger, 1946–99
Model 1 replicates the conventional finding about audience costs for 1946–99.57
The result also holds for the 1816–1984 period analyzed by Schultz (2001a).
In all models, control variables perform consistently with previous work. Geographically contiguous targets are more likely to reciprocate, and disputes about policy-related issues are less likely to be reciprocated than disputes about other issue types.
I now test whether these results change when nondemocracies are treated not as a homogenous group, but rather as distinct regime types according to the criteria outlined above. Consistent with the approach taken above, I add new dummy variables indicating whether or not the initiator was a personalist, military, single-party, hybrid military/single-party regime, nondemocratic interregnum, new democracy, mixed regime, dynastic monarchy, or nondynastic monarchy.58
Geddes (2003) codes a number of regimes as hybrids between personalist regimes and military or single-party regimes. Consistent with the audience costs logic, it seemed appropriate to treat these hybrids as personalist regimes, though differentiating between hybrid personalist and “pure” personalist regimes does not affect the results.
Tables of further specifications are available on the author's Web site at 〈www.stanford.edu/∼jweeks/research〉.
Recall that based on the logic of audience costs, personalist regimes, nondemocratic interregna, new democracies, and nondynastic monarchies were hypothesized to have a lower ability to generate audience costs than other regime types. The statistical analysis indicates that targets are indeed significantly more likely to resist personalist regimes, new democracies, and nondemocratic interregna. The coefficient on nondynastic monarchies is also positive, though not significant. In contrast, regimes that were hypothesized to more easily generate audience costs, including single-party regimes, military regimes, dynastic monarchies, and mixed nondemocracies, are statistically no more likely to face reciprocation than democracies. The coefficients on single-party regimes, mixed nondemocracies, and dynastic monarchies and are in fact negative in some specifications.60
See additional tables at 〈www.stanford.edu/∼jweeks/research〉, including specifications that include a measure of the size of the winning coalition. A smaller winning coalition might capture the leader's ability to impede coordination by restricting appointments to high office. I find that larger winning coalitions do indeed significantly predict lower reciprocation when it is the only regime type variable included in the analysis. However, when the regime type variables coded for this article are also included in the model, the winning coalition coefficient is no longer significant, and the findings remain consistent with the results presented here.
The p-value for the coefficient on new democracies in bilateral disputes is .06, compared to .02 in the full samples.
It is also useful to assess the substantive effect of regime type on the probability that a target resists. Using Clarify, I estimate the probability that the target state reciprocates conditional on the regime type of the challenger.62
Tomz, Wittenberg, and King 2003.
Predicted probability of reciprocation by regime type of challenger
Democracies, single-party regimes, mixed nondemocracies, and dynastic monarchies face the lowest estimated reciprocation rates, with predicted probabilities of reciprocation of around .27 each. Military regimes follow, with reciprocation probabilities around .37, though the confidence interval contains the point estimates for the first four regimes. Personalist regimes and new democracies, in contrast, face the highest estimated rates of resistance; their threats face a predicted probability of resistance of .48 and .44 respectively. Personalist regimes, therefore, are approximately 1.8 times as likely to face reciprocation as democratic, single-party, and mixed nondemocratic regimes. New democracies are 1.6 times as likely to face reciprocation as stable democracies. Moreover, the confidence intervals of most nondemocratic regime types overlap with the estimated democratic reciprocation rate; only autocratic interregna and personalist regimes are statistically distinguishable from democratic initiators with 95 percent confidence.
Finally, it is worth addressing the robustness of the results, as well as alternative interpretations of the findings. First, I subjected the results to additional specifications to ensure that the results are not being driven by other variables correlated with regime type. For example, I control for measures of economic development to ensure that personalist regimes were not proxying for “weak” regimes, by adding per capita energy consumption to the analysis.63
Peceny, Beer, and Sanchez-Terry 2002.
Another robustness check is to ensure that individual states are not driving the results. In some cases, individual states make up a large proportion of states in their regime category. I exclude high-conflict countries such as the Soviet Union, China, Iraq, Syria, and North Korea, Thailand, the United States, India, and Israel both separately and in various combinations that could potentially affect the results. The results are robust to such exclusions.64
See 〈www.stanford.edu/∼jweeks/research〉 for replication code.
An additional question is whether regime categories simply capture variation in levels of democracy. One possibility is that personalist regimes are merely the “most autocratic” of the autocracies; the conventional wisdom would predict that the most autocratic regimes should be least able to signal. To assess this possibility, I take two approaches. First, I include Polity scores in the logit model to see whether challengers' Polity scores explain variation in reciprocation. This estimation is reported in Model 4 in Table 4. I estimate the model only on the sample of nondemocratic states, with personalist regimes as the base category, since the main question is whether Polity scores explain variation within autocracies.65
Moreover, including the Polity score in an estimation that includes the whole sample leads to problems of collinearity because the democ variable is correlated with Polity scores at .79.
Average polity scores in sample by regime type, 1946–99
A related concern is that the personalist variable is actually picking up measurement error in the relative military capabilities data. Available measures of capabilities are well known to be imperfect. Since personalist regimes are often secretive, it could be that estimates of personalist regimes' military capabilities are biased upward in comparison to other regime categories. Consistent with that line of reasoning, Peceny, Beer, and Sanchez-Terry, following Quinlivan, argue that personalist leaders are particularly wary of military coups, and that “coup-proofing” the regime requires leaders to weaken the military apparatus.66
The inability of personalist states to induce their targets to acquiesce might therefore be because personalist regimes tend to be weaker, rather than less able to signal. But this would be a problem only if capabilities share were an important predictor of reciprocation rates. In fact, initiator capabilities share is not a significant predictor of reciprocation when the models are estimated on a sample that excludes personalist regimes (thus, on a sample of states with potentially less biased capabilities measures).67In fact, the finding that the initiator's share of capabilities does not predict reciprocation rates supports the prediction of Fearon's 1994 model.
Finally, the high reciprocation rate to personalist challengers could indicate that personalist regimes have lower “values for war,” not higher audience costs.68
Quinlivan 1999.
Schultz 2001a.
Andreski 1980.
Conclusion
The conventional theory that democracies are systematically superior to nondemocracies in generating audience costs underestimates the difficulties most autocrats face in maintaining power. The literature on audience costs has taken a narrow view of accountability—one that focuses primarily on electoral procedures for removing leaders. This article argues that most nondemocracies do in fact meet the basic requirements for generating politically significant audience costs. Only when the leader can use monitoring and punishment to prevent elite coordination, or when foreign decision makers cannot observe the possibility of such coordination, can states not generate audience costs. My statistical analysis of behavior in MIDs shows that threats by democracies are not significantly more credible than threats by most autocratic regimes. Personalist regimes, new democracies, and unstable nondemocracies, on the other hand, are much more likely to face resistance from the targets of their threats.
The evidence in this article not only supports the existence of autocratic audience costs but also casts doubt on alternative theories about the effects of domestic politics on international relations. Schultz has argued that public, legitimate, and institutionalized party competition helps states credibly reveal resolve in crises.71
Schultz 2001a.
The findings also suggest that much remains to be learned about differences in domestic politics across nondemocracies, and how such variation affects international relations. If, as I argue in this article, some autocratic elites can truly hold leaders accountable, one might expect nondemocracies to vary in the extent to which they punish leaders for other foreign policy failures.72
Weeks 2007.
The analysis is not only of theoretical importance but also suggests practical implications for policymakers. According to many scholars, uncertainty about other states' intentions significantly increases the likelihood of international conflict.73
Under anarchy, states that cannot discern another state's intentions will tend to assume the worst, amassing arms and potentially creating a “security dilemma” that increases distrust and makes conflict more likely.74Jervis 1978.
Dyads in which both states have high audience costs, therefore, are often hypothesized to be more peaceful in their international relations than other dyads. The conventional wisdom about the relationship between democracy and audience costs supports the view that democratization increases peace, or that “The best hope for peace in our world is the expansion of freedom in all the world.”75
President George W. Bush, Second Inaugural Address, 20 January 2005. Text available at 〈http://www.whitehouse.gov/inaugural/〉.
The overall implication of this analysis is that scholars and policymakers misplace attention when they focus on electoral competition and other elements of democracy, rather than theorizing about less normatively appealing forms of political rivalry. Autocratic leaders, while they may exert enormous control over their subjects, are not usually immune from domestic threats to their tenure. The standard dichotomy, therefore, masks important variation between types of authoritarian regimes and may lead to mistaken inferences about the effects of democratic domestic institutions on foreign relations. This analysis shows that by analyzing differences between nondemocratic regimes, one is led to question previous assumptions about the relationship between democracy and international relations. Perhaps more importantly, analyzing the effects of domestic politics in nondemocracies offers a fresh avenue for gaining insights into international behavior.
Appendix
Regime Type Categories
Monarchies
Since Geddes does not code monarchies, I use the Polity IV xropen variable for “Openness of Executive Recruitment” and define monarchies as country-years where xropen < 4 & xropen > 0. This identifies all regimes with hereditary succession. I code dynastic vs. nondynastic monarchies according to Herb.76
Herb 1999, 8.
Control variable definitions
Pure Versus Hybrid Personalist Regimes
Geddes codes regimes as “hybrids” when they exhibit characteristics of more than one regime type. The logic I develop in this article focuses on whether the leader controls appointments and the intelligence and security apparatus. In my analysis, I therefore place hybrid regimes with a “personalist” component in the personalist category because leaders of these regimes, like pure personalist leaders, are more likely than other regimes to control these aspects of the state. Treating hybrid personalist regimes and purely personalist regimes as separate categories does not affect the central findings.
