The Effect of Foreign Pressure on the Collapse of Authoritarian Regimes

Transcription

1 Draft please do not cite without permission The Effect of Foreign Pressure on the Collapse of Authoritarian Regimes Barbara Geddes Department of Political Science UCLA Los Angeles, California Abstract Most observers agree that foreign pressure and intervention of various kinds affect regime transitions, but few systematic tests of these effects have been conducted. This study focuses on the effect of economic sanctions, aid, and military intervention on the breakdown of authoritarian regimes. The analysis confirms the conventional wisdom that foreign pressure or intervention can help bring about regime transitions. It shows that two of the least subtle forms of foreign pressure, costly sanctions and direct military intervention do, on average, destabilize dictatorships. Cheap or symbolic sanctions, however, are associated with increased rather than decreased authoritarian durability. Aid and military intervention to support threatened dictators appear to be ineffective. The study also finds that although foreign pressures can encourage transitions, they do not overwhelm the domestic sources of authoritarian fragility or strength. The standard findings of previous research on the domestic causes of authoritarian breakdown are confirmed even with various foreign influences held constant. Prepared for presentation at the annual meeting of the American Political Science Association, Boston, I would like to thank Daniel Drezner for generously sharing his data on sanctions with me, and Liz Stein for intelligent and conscientious research assistance. 1

2 The Effect of Foreign Pressure on the Collapse of Authoritarian Regimes Most observers, whether presidential advisors or academic theorists, believe that foreign intervention can destabilize governments. The belief that external economic and military pressure can push regimes toward breakdown has motivated sanctions against Iraq, U. S. intervention in Nicaragua during the eighties, and covert C.I.A. operations in many countries over the years. Many observers of the third wave of transitions to democracy, including Huntington himself (1991), have also noted the importance of other kinds of foreign pressures. Most analysts attribute the fall of the communist regimes in Eastern Europe and some of those elsewhere to the collapse of the Soviet Union and, with it, its support for other communist governments. The Carter administration s human rights policy, which led to the abrupt reduction in aid to the Shah of Iran, Somoza in Nicaragua, and several other especially notorious dictators, has been credited with contributing to their falls from power. More recently, scholars have suggested that requirements to hold multiparty elections and increase government transparency, attached as conditions to some loans from international financial institutions, have contributed to the spate of political liberalizations in Africa during the nineties. Despite these widely held beliefs about the effects of foreign pressure and intervention on regime durability, the transitions literature has concentrated on domestic causes of regime change. Most case studies focus mainly on domestic actors, and most large-n studies assess the effects of domestic wealth, growth, institutions, religion, and colonial heritage. One of the few quantitative efforts to test the effects of external pressure, Bratton and van de Walle s 1997 study of African transitions, comes up with somewhat contradictory results. They find that aid dependency in African countries increases the likelihood that the countries will move in a democratic direction, but that explicit conditionality attached to loan agreements has the opposite effect. In short, although most analysts believe and many have asserted in print that foreign pressures of various kinds play an important role in bringing about regime change, little 2

3 systematic research has been done on the subject. What has been done has produced inconclusive results. In this paper, I test systematically the effect of several kinds of external pressure on the durability of dictatorships. I consider the effect of aid flows, which have been credited with helping to maintain our sons of bitches 1 ; economic sanctions, which have sometimes been aimed explicitly at bringing down particular dictators and which, whatever their particular aim, might be expected to increase political instability in the same way that any other economic stress does; direct military interventions seeking to end particular regimes; and threats and interventions, as within the Soviet bloc, aimed at maintaining allied regimes. Note that this study focuses on the breakdown of authoritarian regimes, rather than democratization. I seek to assess the effect of foreign intervention on the durability of dictatorships, regardless of whether their breakdown results in democracy. The analysis below shows that some kinds of foreign interventions do have the expected effects. Direct military intervention, for example, helps to bring down dictatorial regimes, and so do costly sanctions. Other kinds of interventions, however, fail to show systematic effects. Military intervention for the purpose of helping to maintain the regime in power is not, on average, effective. Sanctions that have more symbolic than real economic effects do not destabilize targeted regimes and may, in fact, give them a boost. Although sanctions that exact a high economic cost increase the probability of regime breakdown, as do sanctions that effectively reduce the target country s ability to import, costless or ineffective sanctions seem to have the opposite effect. They reduce the probability of regime failure. This study finds no systematic evidence either that foreign aid helps to maintain dictators in power or that decreases in aid help to bring them down. Neither, surprisingly, does civil war, once other factors have been controlled for. Equally important, the study confirms the importance of domestic causes of authoritarian breakdown even when multiple kinds of foreign pressure and intervention are controlled for. After holding constant military intervention, sanctions, aid flows, civil war, and changes in imports, low growth still destabilizes dictators, and higher levels of wealth still keep them more stable, as has 1 The quip, Somoza may be a son-of-a-bitch, but he s our son-of-a-bitch, has been attributed to 3

4 been shown in previous research. Military regimes still show more fragility, on average, than other forms of authoritarianism, and single-party regimes continue to demonstrate extraordinary resilience. Previous Research Observers agree that the actions of foreigners can sometimes have decisive effects on the survival of dictators. No one disputes the effect of policy changes in the Soviet Union on the survival of East European communist regimes, and most observers believe that U.S. intervention in Central America helped bring down the Sandinista regime while helping to maintain military rule in El Salvador. A number of Africanists have attributed the halting steps toward democratization in Africa to conditions mandating political liberalization attached by international donor and financial institutions to loan and aid agreements (e.g., Munslow 1993; Young 1993). Scholars have also noted the effect on people in various parts of the authoritarian world of seeing the mass demonstrations in Leipsig and Timisoara on television. All in all, it is hard to believe that foreign influences of one kind or another have not affected the survival of authoritarian governments. Foreign intervention, pressure, and support are discussed in the case study literature about specific transitions in which observers believe it has played an important role, 2 and Richard Snyder (1998) attributes an especially large role in the ouster of sultanistic regimes to foreign influence. Most attempts to draw general theoretical conclusions about the causes of transition, however, have given it little attention. O Donnell and Schmitter expressed what was probably the prevailing wisdom before 1989: [I]t seems fruitless to search for some international factor or context which can reliably compel authoritarian rulers to experiment with liberalization, much less which can predictably cause their regimes to collapse (1986, 18). Haggard and Kaufman (1995) briefly acknowledge the importance of international causes before moving on to ignore them for the rest of their analysis of democratization and economic reform. Quantitative studies of regime change either fail to test for the effects of foreign intervention or fail to find expected effects. Franklin Roosevelt. 2 See, for example, the case studies in Chehabi and Linz (1998). 4

5 Bratton and van de Walle (1997) review many of the claims about international influences on political change in Africa before asserting the priority of domestic sources of political reform. In the quantitative part of their study, Bratton and van de Walle test the claims about loan conditionality and find little support for the conventional wisdom about its effects, though they do find that aid dependence is associated with greater political liberalization. The dependent variable in their study is movement along a political liberalization scale, not regime change, and, as Bratton and van de Walle note, the greatest gains in political liberalization in their sample of African countries during the eighties and early nineties occurred in countries that started lowest. Since the model includes no measure of wealth as a control, it is quite possible that aid dependency serves as a simple proxy for poverty more than as an indicator of foreign influence, and that poverty was correlated with starting at the lowest levels. In other words, the finding that political liberalization is associated with aid dependency probably should not be interpreted as evidence of foreign pressure, and Bratton and van de Walle do not give the result much weight in their conclusions. The Przeworski et al. (2000) study of democratization and development, the current state of the art for quantitative studies of regime change, includes only one indicator of international influence. It is the proportion of all regimes currently democratic, which they treat as an indicator of the general climate of world opinion about democracy. This variable is used as a proxy for the pro-democracy international context of the 1980s and 1990s, which has been cited by many as an important source of pressure and inspiration, but their statistical analysis fails to confirm its importance. In short, most observers believe that international influences affect the likelihood of regime change, but the literature on transitions in Comparative Politics has not shown systematic effects. Scholars in International Relations, meanwhile, have shown that a particular form of international pressure, sanctions, can sometimes destabilize governments (Hufbauer, Schott, and Elliott 1990, vol.1, p. 93). International relations scholars have been primarily interested in the conditions under which sanctions are effective, however, not in their effects on authoritarian regimes. 3 3 See Baldwin (1999) for a useful review and critique of the IR literature on sanctions. 5

6 In the analysis below, I test some of the most common and least subtle arguments about the effects of foreign intervention on the longevity of authoritarian regimes. To do that, I add various indicators of foreign intervention to a domestic model of regime breakdown I have used previously. Tests of the effectiveness of sanctions in the international relations literature implicitly assume that in the absence of sanctions, the targeted policy would remain unchanged. This may be a reasonable assumption to make with regard to some of the policies sanctions seek to affect, but it is clearly not reasonable to assume that dictators will remain in power indefinitely in the absence of sanctions. Dictatorships fall for lots of reasons besides sanctions. If we want to assess the effect of sanctions, we need to use a model that holds those other reasons constant. In earlier work (Geddes 1999) I have found that some kinds of authoritarianism are much more resilient than others in the face of economic crisis and other challenges. This study also confirmed the standard finding that economic performance affects regime survi val. Economic decline hurts all kinds of governments, whether democratic or authoritarian, and regardless of what else is going on. The level of economic development has also been found in numerous studies to affect regime stability. My previous work indicates that authoritarian regimes in richer countries are more stable than those in poorer countries. This finding may seem surprising, but is consistent with the results of Londregon and Poole s (1990 and 1996) analysis of the causes of coups. When authoritarian regimes do break down in richer countries, however, they are likely to be followed by democracies while new dictatorships often follow the ouster of old ones in poorer countries. Variables and Model To assess the effects of external pressures, domestic causes of regime stability need to be included in the model and held constant. I have also included various other control variables in some of the models. The percent of the population that embraces Islam is included in some of the specifications because various studies have shown that Islamic countries are more likely to have authoritarian governments. Dependence on oil and mineral exports was included in some specifications for the same reason: numerous scholars have claimed that the availability of oil revenues helps dictators maintain themselves (e.g., Ross 2001). These regressions are not 6

7 shown below because neither dependence on oil nor dependence on mineral exports appeared to contribute to regime durability in any of the many specifications tried, and their inclusion reduces the number of observations by about half. Region dummy variables are included in the regressions shown to hold constant idiosyncratic cultural and historical factors that cluster by region. The results are not substantively changed by leaving them out. The data set used here includes information on authoritarian regimes that includes nearly all authoritarian regimes except monarchies, that have existed or come into existence since 1946, and that lasted three years or more in countries of one million or more that had achieved independence before The data set includes more than 160 regimes. I use hazard models for the tests shown below. 5 These models, originally developed for use in the insurance business and medical research, are used to predict the survival of subjects, given certain conditions. In this research, authoritarian regimes are the subjects, and this kind of model allows us to assess the effect of various factors, such as low growth or economic sanctions, on the likelihood of authoritarian survival. Variables The Usual Suspects Growth: Change in GDP per capita, assessed in the prior year. Since regime breakdowns can happen at any time during the year, and the old regime may thus not be responsible for economic performance in the year in which a regime breaks down, it seems safer to use the previous year s growth rate to assess the regime s economic performance. After trying various lag functions, Przeworksi et al. (2000) also use a one-year lag on growth. Level of Development: log of GDP per capita. 6 Logged GDP per capita is used here as in a number of other studies to correct the skew in GDP per capita. Islam: Percent Muslims in the total population Regime Type 4 For more information on the data set, see Geddes In earlier work, logit produced very similar results. 6 GDP per capita and Growth are taken from the Penn World Tables, which cover for most countries. 7

8 I initially classified regimes as military, single-party, or personalist. In the real world, however, many regimes have characteristics of more than one regime type. When regimes had important characteristics of more than one pure regime type, especially when the area specialist literature contained disagreements about the importance of military and party institutions, I put them in hybrid categories. A brief description of the classification criteria is included below. See Geddes (1999) for a full discussion of the coding of these regime types. Military: a regime governed by an officer or retired officer, with the support of the military establishment and some routine mechanism for high level officers to influence policy choice, appointments, and succession. Examples include the Brazilian regime and the Argentine Single-party: a regime in which a dominant party controls most access to political power and government jobs, influences policy, and has functioning local-level organizations. Examples include the Mexican regime under the Partido Revolucionario Institucional (PRI) and the Zambian under the United National Independence Party, as well most of the East European communist regimes. Personalist: a regime in which the leader, who usually came to power as an officer in a military coup or as the leader of a single-party government, has consolidated control over policy, recruitment, and the security apparatus in his own hands, in the process marginalizing other officers' influence and/or reducing the influence and functions of the party. Examples include Idi Amin s dictatorship in Uganda and the Somoza regime in Nicaragua. Military-Personalist: a regime with traits characteristic of both military and personalist, for example, one in which the military remains professionalized and somewhat influential but in which the dictator makes most decisions, including those about promotions and retirements within the military, without consulting them. Examples include the Pinochet government in Chile and the Abacha government in Nigeria. Single-Party Hybrid: a regime with traits characteristic of both single-party and either personalist or military regimes, for example, a regime in which the dominant party plays an important organizing role in society but in which many important decisions are made without consultation by 8

9 the paramount leader. In practice, these are nearly all single-party personalist amalgams because there are very few single-party military combinations in the real world. Examples include the Kim regime in North Korea and the Castro regime in Cuba. Triple Threat: a regime that amalgamates traits associated with all three pure types. Examples include the Egyptian regime that began in 1952 and continues to this day and the Suharto regime in Indonesia. There are very few of these and they are very long-lived. A dummy variable for them is included in the model in order to keep these few unusual regimes from contaminating the results. Each regime type is entered in the model as a dummy variable in order to avoid making assumptions about scaling them. Foreign Intervention Variables Sanctions: scored one if the country was the target of economic sanctions during that year and zero otherwise Length of Sanctions: the number of years sanctions had been in place Cost of Sanctions: the cost of the sanctions to the economy of the target country, as calculated by Hufbauer, Schott, and Elliott. For a description of how the cost was calculated, see Hufbauer, Schott, and Elliott, 1990, appendix A. 7 Aid per capita: total development aid per capita, measured in current US dollars Aid as a percentage of gross national income: total development aid, measured in current US dollars Growth in Imports: Change since the previous year in imports as a percentage of GDP. 8 Growth in imports is used as a proxy for the effectiveness of sanctions since effective sanctions would be expected to reduce the target country s ability to import. Foreign Intervention to Support Regime: scored one if foreign troops are actually stationed in the country, or if a foreign country is supplying extraordinary amounts of military aid along with 7 Sanctions, Length of Sanctions, and Cost of Sanctions are from Drezner s revision and extension of the events data in Hufbauer, Schott, and Elliott (1990). I am very grateful to Daniel Drezner for sharing these data with me. For details about his changes to the Hufbauer, Schott, and Elliott data, see Drezner 1999, chapter 4. 8 Aid and Import figures are from World Development Indicators CD-ROM

10 military advisors to help support the government in power, zero otherwise. (Coded by the author.) Foreign Intervention in Opposition to Regime: scored one if foreign troops are actually fighting against government forces inside the country; or a foreign country is supplying very large amounts of military aid along with advisors to an armed opposition group; or if a foreign country invades, bombs, or shells territory within the official boundaries of the country held by the regime. In the latter situation, the foreign country may not be seeking to overthrow the regime, but I include these situations since observers often think that such foreign incursions can discredit and destabilize governments. (I have not included foreign adventures initiated by the dictatorship that have led to defeats but not to foreign invasion or destruction of the home territory, e.g., the Falklands or the Greek defeat in Cyprus.) Civil War: Scored one if more than 1000 combat-related deaths have occurred in internal conflict, with both the government and an identifiable opposition force each suffering a minimum of five percent of the deaths. Country -years are scored zero otherwise. 9 This is the standard measure used in quantitative work on the causes of civil war. I have compared it to my own independent coding of civil wars, and there were few discrepancies. This variable is included because many of the instances of foreign military intervention occur in the context of civil war. Dependent Variable The dependent variable is regime breakdown, coded one for years when a regime fell and zero otherwise. An authoritarian regime is coded as having broken down if either of the following occurs. Competitive and reasonably free and fair elections are held, the incumbent loses, and the winner is allowed to take power (the date used as the transition year is that of the election). Or alternatively, the ruler is ousted nonconstitutionally and his place taken by an individual or organization that was not part of the prior ruling group Data from Collier and Hoeffler (2001), an expanded and updated version of Singer and Small s data collection. 10 Elkins (2000) has argued that multi-category regime type scales work better than dichotomous ones, and for some purposes they do. Here, however, I do not seek to explain political liberalization or movement in the direction of democracy. Rather, I seek to explain why authoritarian regimes end. A regime is defined as the set of rules and procedures, formal and informal, for choosing leaders and policies. Sometimes particular authoritarian regimes end but 10

11 A Note on the Interaction between Data Sources and Sample Each of the data sets used here covers a slightly different time period, which means that some regressions include slightly different country -years than others, depending on which variables are included. The sanctions data extend through the early 1990s, and data on growth and level of development from the Penn World Tables extends only to That means that a number of recent African transitions are excluded, and it does not seem appropriate to test the effect of IMF conditionality without them. Data from the Penn World Tables go back to 1950, but data based on the World Development Indicators only begin in 1960, so regressions that include Aid and Change in Imports exclude authoritarian country-years during the fifties. Unless we think that the causes of authoritarian breakdown were different in the fifties, this should not be substantively consequential, though it might affect statistical significance. Data on economic performance are unavailable for a number of countries: Afganistan, Albania, Cambodia, Cuba, Iraq after 1988, Laos before 1986, Liberia after 1987, Libya, Myanmar, North Korea, South Vietnam, South Yemen, (North) Vietnam, and (North) Yemen before These countries or country-years are thus excluded from most of the analysis. This set of countries includes several in which sanctions have persisted for very long periods of time without bringing down authoritarian regimes. Most of the countries excluded are also extremely poor, and more than half were governed by single-party regimes. If they were included, they might change some findings, and I discuss the ways results might be affected by their exclusion below. I also show the effect of including them in the analysis (by leaving out the control variables for economic performance). Findings Before turning to the effects of foreign intervention of various kinds, let us consider a model of authoritarian breakdown that includes only domestic causes. The model below was developed are followed by even more repressive and abusive ones. In such a situation, the Freedom House scale and others like it would classify the regime as continuing to be authoritarian rather than as having broken down. Sometimes authoritarian regimes liberalize considerably while maintaining a monopoly of power, that is, essentially the same rules for choosing leaders and policies. In this situation, an incremental scale suggests a regime change that has not really occurred. 11

12 not to explain fully the fall of dictatorships but rather to demonstrate that some kinds of authoritarianism are more resilient than others to the various challenges all governments face. It is undoubtedly incomplete, but it is at least a good starting place. Domestic Causes of Authoritarian Breakdown The basic argument underlying this model is that the different interests of the high level cadres of different kinds of authoritarian government lead to characteristically different ways of responding to challenges and, in some situations, collapsing or being overthrown. Professionalized military regimes, as I show elsewhere, are more fragile than other forms of authoritarianism because most officers value the unity and efficacy of the military more than they value holding power and, for most, the costs of returning to the barracks are low. Consequently, when faced with splits in the officer corps over succession or policy that threaten military cohesion and efficacy, officers tend to opt for a return to the barracks. In contrast, the cadres of single-party regimes are better off if the party retains its dominant position and if all factions remain adherents. Single-party governments thus tend to try to coopt challengers. They never volunteer to give up power, and they show great resilience in the face of economic and other kinds of crisis. The costs of competitive politics to them are considerable, but not devastating, however, as many can continue successful political careers post-transition. As a result, when they can no longer hang on to power, like the military, they usually negotiate a peaceful extrication. Personalist dictators rely more heavily than do single-party or military rulers on the distribution of individual benefits to secure the loyalty of cadres. Cooptation strategies are thus less available to them since reliance on individual benefits implies the superiority of a minimum winning coalition over broader ones. Furthermore, they usually undermine organizations such as the military or party that might serve as the power base of a rival. Consequently, the cost of regime breakdown to the cadres of personalist regimes tends to be quite high. They lack organized bases for continuing successful political careers after a regime change, and the favors they received while in office are usually perceived by citizens as the fruits of corruption. They are frequently deprived of property, jailed, and even killed during or after transitions. Personalist 12

13 regimes are less likely to negotiate transitions, more likely to renege on deals they make if they are forced to negotiate, and more likely to be violently overthrown than military or single party. 11 They are, however, unlikely to outlive the original ruler, which makes them less durable than single-party regimes. If this argument is correct, we should expect to see that, all else equal, military regimes are more likely to break down and single-party regimes less likely. The Domestic Causes of Breakdown model, shown in Table 1, shows the expected relationships. The figures reported in the first column are hazard ratios, which have a very simple and intuitively appealing interpretation. Numbers above one indicate that the variable associated with them increases the likelihood of breakdown by the amount shown. Here, military regimes are nearly three times (2.83) as likely to break down, all else equal, as personalist regimes (the left out category), and this estimate is statistically significant. Hazard ratios below one indicate that the variable associated with them reduces the likelihood of the outcome. The ratio 0.38 for Single-Party regimes in Table 1 means that singleparty regimes are estimated to be about 40 percent as likely to fall as personalist, all else equal. Military-Personalist and Single-Party Hybrid regimes are on the margin between Military and Personalist or Single-Party and Personalist, and are thus not much different from Personalist regimes. Not surprisingly, the hazard ratios for these regime types are not close to statistical significance. Nor is the Triple Threat regime category, which only contains about six cases. This variable is included in order to hold constant a few very unusual regimes that might otherwise inflate the apparent effect of single-party rule. Increases in Level of Development (log GDP per capita) reduce the probability of breakdown. This may seem surprising, but remember that this model predicts the breakdown of authoritarianism, not democratization. All regimes are more stable when long-term economic performance is good, and Level of Development in essence measures long-term economic performance. A positive growth rate (Growth) makes breakdown extremely unlikely, as would be expected. Perhaps surprisingly, the percent Muslim in the population increases the probability of breakdown and is statistically significant. This finding is not 11 For a much more complete elaboration of this argument, see Geddes (1999). 13

14 really inconsistent with prior work showing an association between Islam and authoritarianism, however, because authoritarian collapses in Muslim countries are usually followed by renewed authoritarianism, not stable democracy. The regions are included here to control for idiosyncratic historical and cultural characteristics that might affect regime durability. Sanctions Governments use economic sanctions, most often in the form of cut-offs of finance, aid, and trade, to try to change the behavior of other states. Sanctions have been used to try to persuade governments to reduce human rights violations and to deter them from developing nuclear capacity, among other goals. They are also sometimes used with the intention of bringing down particular political leaders. Even if that is not their explicit purpose, sanctions would be expected to affect regime stability by undermining economic performance. Table 2 shows the variety of reasons that sanctions are imposed. A little less than half of the sanctions episodes coded by Hufbauer, Schott, and Elliott (1990) and Drezner (1999) are actually included in the analysis here. The sanctions episodes excluded include those that occurred before 1951, those imposed on countries for which no economic data are available, and those imposed on democracies. In general, the goals of sanctions in the included sample were similar to those in the full sample, with one notable exception. Sanctions for human rights violations were twice as frequent in the included sample, as might be expected in a sample composed only of authoritarian regimes. Note that in both samples, the percent of sanctions aimed explicitly at destabilizing regimes is only about ten percent. Somewhat surprisingly, the full sample contains a somewhat larger proportion of sanctions aimed at destabilization. This is because of a number of U.S. initiated sanctions episodes aimed at destabilizing elected governments such as those of Allende in Chile and Goulart in Brazil. How economically damaging sanctions are varies a lot. Some sanctions are much more costly to target countries than others. Table 3 shows the sanctions episodes with the highest cost 12 in this data set. Note that in about 40 percent of the costliest episodes, the sanctions 12 As calculated by Hufbauer, Schott, and Elliott (1990, Appendix A) and revised by Drezner (1999). 14

15 aimed to destabilize the government of the target country, while destabilization was the goal in only about 10 percent of the full sample of sanctions episodes. Evidently, when destabilization is the goal, policy makers impose more costly sanctions. Some sanctions are not meant to be very costly, however. Instead, they aim to influence the behavior of allies rather than cause serious damage. Others are intended to cause damage but may not do so because the sanctioned country can find other markets for its products, other suppliers for the imports it needs, other investors, and other sources of aid. In these situations the sanctions, though intended to be costly, are not effective. Where sanctions are both costly and effective, however, we might expect them to help bring down dictators, though some well-known cases (e.g., Saddam Hussein, Fidel Castro) of their failure to do so may cast doubt on their usefulness. Hufbauer, Schott, and Elliott (1990, p. 59) code 62 percent of sanctions aimed explicitly at destabilizing the target government as successful. As noted above, however, the targeted governments might have fallen for reasons other than sanctions so other causes of regime breakdown need to be included in the analysis in order to judge the effect of sanctions. Sanctions can be measured in a number of ways. The variable labeled Sanctions in Table 4 is a dummy, that is, country-years were coded one if subjected to sanctions and zero otherwise. Model 1 shows the effect of simply having sanctions applied. Authoritarian governments subjected to sanctions are, on average, only about 60 percent as likely to fall as those that are not sanctioned. This result suggests that sanctions actually help dictators to stay in power, but the hazard ratio is not quite statistically significant, so the result has to be considered somewhat tentative. The data set contains two other measures of sanctions, one is the length of time they have been in place and the other is an estimate of how costly they are. We might think that length would matter because sanctions might become harder to bear over time, and we would expect that more costly sanctions would be more likely to destabilize any kind of government. Model 2 in Table 4 shows the expected effects. As sanctions become more costly, the probability of regime breakdown rises. As sanctions go on longer, they become somewhat more challenging, but the 15

16 effect of lengh is less strong and not quite statistically significant. When Cost and Length of sanctions are controlled for, sanctions per se further reduce the likelihood of regime collapse, and this finding is now statistically significant. The substantive interpretation of these findings is that cheap, symbolic sanctions actually contribute to authoritarian survival, perhaps by causing the same kind of rally-round-the-flag effect that occurs in democracies that face foreign challenges. Costly sanctions do, however, tend to destabilize dictators. Model 3 shows the effect of a couple of different indicators of the effectiveness of sanctions while contolling for their cost. It has been suggested that aid from other sources can reduce the effectiveness of sanctions. Here Development Aid per capita is used as an indicator of the availability of aid. The hazard ratio shown here suggests that development aid increases the likelihood of regime breakdown rather than contributing to survival. Getting one dollar of aid per capita rather than none increases the probability of overthrow by one percent. Since development aid per capita ranged between less than a dollar and a little over two hundred dollars in the countries included in this data set, large amounts of aid might have an important destabilizing effect if we believe this result. We might suspect that this apparent effect is spurious, however, because the desire to support threatened regimes might motivate aid givers. Many of the largest recipients of aid do indeed fit the profile of threatened regimes, but either because they were democratic or because they were monarchies, they are not part of this data set. Within this data set, the largest recipients of aid were Botswana , Central African Republic 1971 and , Congo (Zaire) 1987, Gabon , Guinea-Bissau , Mauritania , Mozambique , Nicaragua 1990, Senegal , Somalia , and Syria One of these, the Somali, was nearing collapse, but aid declined during the last three years before the breakdown. Aid to the governments in Nicaragua and Guinea-Bissau during transition years went to the governments that replaced them, not to the threatened old regimes. Most of these recipients were, however, very stable. Those in Botswana, Congo, Gabon, Senegal, and Syria were some of the longest-lived and stablest non-communist authoritarian regimes in the world. So aid does not seem to flow generously to governments on the verge of overthrow. 16

17 Aid per capita is not included in subsequent regressions because its inclusion in models with more variables reduces the number of observations and causes some of the region dummy variables, often including the Middle East, to drop out. Dropping different region dummies in different regressions makes the results bounce around in unpredictable and probably meaningless ways, so I have not included these regression results in the tables. I have run about 25 different specifications with aid included, however, sometimes per capita, sometimes as a percent of gross national income, and sometimes as change in aid per capita. In none of the specifications does it appear to reduce the likelihood of authoritarian breakdown. Its hazard ratio is remarkably stable across specifications, varying from 1.00 (no effect) to 1.01 for aid per capita, with a P value that range from.03 to about.75, that is, from conventionally significant to not close. I conclude from these results that although the finding that aid destabilizes might be spurious (or non-existent), these data provide no support for the claim that aid actually maintains dictators in power. Growth in Imports is used here as an indicator of the effectiveness of sanctions. If sanctions are effective, target countries should not be able to import as much as they had prior to sanctions, but if they find alternative markets and suppliers, their imports should not be affected. Imports are obviously affected by things other than sanctions and can be considered a general measure of economic performance. Growth is controlled for in these models, however, so I think it reasonable to treat Growth in Imports as reflecting in part the effects of sanctions. Model 3 shows that Growth in Imports has the expected effect. The change from no change in imports to a one percent increase reduces the likelihood of breakdown slightly. When the effects of aid and growth in imports are controlled for, the effect of the cost of sanctions is somewhat reduced, but cost still contributes to the destabilization of dictators. In Model 4, all three measures of sanctions are included along with Growth in Imports as a measure of effectiveness. When all are included, all are statistically significant, and the effects of all are at least as strong as in the earlier models. These results suggest either that costless sanctions are inadvertently helpful to dictators, or that they are symbolic political gestures aimed 17

18 by sender governments at placating a domestic audience rather than actually damaging the target. Costly and effective sanctions, however, can undermine authoritarian regimes. Note that the domestic sources of authoritarian survival remain important when variables measuring sanctions and aid are controlled for. Military regimes are more fragile in all model specifications. If anything, the effect of being a military regime on the probability of breakdown is even stronger when external influences are controlled for. The hazard ratio for single-party regimes also remains about the same, indicating that they are about forty percent as likely to fall as personalist regimes, all else equal. In Model 4, the hazard ratio for single-party regimes does not achieve conventional statistical significance because there are only a small number of singleparty regimes so estimates of their effects bounce around some, especially as the sample size is reduced, but the estimate of the ratio is similar. The effects of growth and level of development remain about the same, and the effect of Islamic faith is the same in Model 4 as in the domestic model of authoritarian breakdown (Table 1). In short, costly sanctions affect authoritarian survival, but they do not overwhelm the domestic causes of regime fragility. Foreign Military Intervention Analysis of the effects of foreign military intervention shows both expected and unexpected results. Actual military intervention against dictatorships increases the likelihood of breakdown, but military intervention to support authoritarian regimes appears to have no effect. The second result seems surprising and might result from the way other countries decide which regimes to support with troops, supplies, and military advisors. With the partial exception of the communist regimes in Eastern Europe, countries do not usually supply other countries with extensive military assistance unless their governments are threatened. The total data set includes 31 regimes that were supported for some or all of their time in power by foreign military involvement and 35 that were challenged some or all of the time by foreign military intervention. In nearly a third of the total number of regimes that experienced foreign involvement, foreign troops fought both for and against incumbent governments at the same time. Table 5 shows a list of the regimes and years of foreign military involvement. Usually involvement by foreigners both for and against a government occurred in the context of civil war. I have included civil war as a control variable in 18

19 some of the regressions, but it never demonstrates a significant effect on regime survival. This may seem surprising, but it should be kept in mind that although civil wars can lead to the overthrow of governments, they can also strengthen the hands of authoritarian rulers who claim the need for strong measures in order to defeat insurgents. Model 1 in Table 6 shows the effect of foreign military intervention while excluding the various indicators of sanctions. Direct military opposition increases the probability of regime breakdown by a factor of about three and is statistically significant at conventional levels. Foreign military support for dictators, however, seems to have no effect, and neither does civil war. We might expect that military intervention and sanctions would be correlated and that therefore the inclusion of sanctions in the model would reduce the apparent effect of military intervention. Model 2, which adds the sanctions variables to the specification, however, shows little change in the effect of either military opposition or support. Model 2 includes the same sample of country-years as do Models 3 and 4 in Table 4. In this sample, as in the one included in Model 1, foreign military intervention against an incumbent regime increases the likelihood of breakdown by a factor of nearly three, but military support has no effect. Nor does civil war. With military intervention controlled for, cheap sanctions still increase regime resilience, but costly and effective sanctions still contribute to collapse. The effect of regime type remains strong, as do level of development and growth. In short, foreign military intervention increases the likelihood that dictatorships will fall but does not overwhelm the other causes of authoritarian breakdown. Model 3 in Table 6 omits the economic variables in order to be able to include some of the cases for which economic data are unavailable. Since several of these countries (e.g., Cuba, North Korea, and Libya) have endured sanctions for a very long time, we might suspect that their inclusion would change results. A comparison of the hazard ratios across the columns of Table 6, however, shows that the basic findings remain the same. Opposition military intervention still destabilizes regimes, and military support still appears ineffective. Civil war may increase the likelihood of regime breakdown in the larger sample, though the hazard ratio for civil war does not achieve conventional levels of statistical significance. The hazard ratios for all other variables are approximately the same, and those that were statistically significant in Models 1 and 2 remain 19

20 statistically significant in Model 3. The single-party regime type is statistically significant in Models 1 and 3 because of the increased sample size. In short, the inclusion of what might be called the rogue states does not change the basic conclusions either about the effects of sanctions and military intervention or about the domestic sources of authoritarian stability. A Note on Selection Bias Some scholars in the international relations field have criticized the literature on the effects of sanctions on the grounds that selection bias renders findings suspect (Morgan and Schweback 1997; Pape 1997). Two different selection bias problems afflict most studies. First, arguments about the effects of sanctions are typically tested on samples of cases in which sanctions were applied, in other words, the cases are selected on the basis of their scores on the independent variable (sanctions). This is a problem if policy makers have chosen to apply sanctions where they expected them to work. If that is true and policy makers are competent, then selection into the included category of the independent variable would be correlated with the effectiveness of sanctions. Consequently, statistical conclusions about the their effectiveness may overestimate it because some of the causal power attributed to sanctions was really caused by factors that led policy makers to think sanctions would work. Second, within a sample of sanctioned countries, it is not possible to assess the effectiveness of sanctions relative to other foreign policy instruments or doing nothing. To do that, one would need to compare the sanctioned countries to a sample of the full universe of countries that might have been sanctioned. For sanctions in general, this universe would be difficult if not impossible to identify. The tests undertaken here do not suffer from the second problem. In the analysis above, the universe includes almost all authoritarian regimes, not just those that suffered sanctions. Country-years in which sanctions were in operation are compared to country -years in which they were not. Thus the simplest form of selection bias does not mar the results above. Nevertheless, it still seems likely that policy makers tend to impose sanctions on those countries where they think they will work. If so, then any apparent effectiveness shown in analyses of sanctions events will overstate the general usefulness of sanctions (though not, of course, their usefulness within the set of targets chosen by sensible policy makers). In other 20

21 words, if analysts find that sanctions achieve their ends 40 percent of the time in countries targeted by policy makers, we should not conclude that if sanctions were used to deal with all foreign policy disagreements their success rate would be the same. The same argument could be made about foreign military intervention. Governments may well choose to send troops on foreign adventures when they think they can achieve their ends that way, so an attempt to assess the effectiveness of military intervention as a foreign policy instrument is likely to overestimate it. How serious is the selection bias problem likely to be, and how should we interpret the results above in light of it? To assess this, it seems useful to compare the selection of an interventionist foreign policy with a standard exemplar of selection bias. The example often used in introductory statistics classes involves Head Start. In the example, some parents in poor neighborhoods choose to enroll their children in Head Start; the children are tested a few years later, and those who went to Head Start are doing better in school than those who did not. We do not know whether to believe these results, however, because we cannot tell if the difference we observe was caused by Head Start or by having the kind of concerned and competent parents who would enroll their children in Head Start. Thus the selection of children into Head Start may cause the observed difference rather than the treatment they receive in Head Start. In this example, it is easy to imagine that parents make a big difference in how well children do and that the apparent correlation between attending Head Start and doing better later is spurious. Now let us compare this example with the factors that would be likely to affect policy decisions about military intervention. If policy makers choose to intervene militarily in places where they think they can win, we would expect them to intervene in small, weak countries where the contest will be uneven, and in countries in which the regime is already fragile because of poor economic performance or widespread discontent. The selection bias in this situation is analogous to that of doctors selecting treatments they think will work for particular patients, not that of more responsible parents selecting higher quality experiences for their children regardless of the type of child. When doctors select treatments, we should not conclude that the treatment with a high success rate among the patients selected to receive it would be equally successful for 21