State preferences, viable alternatives, and American covert action, 1946-1989

April 18^th, 1948 was the date of Italy’s first parliamentary election after World War II, and was widely thought to represent the completion of Italy’s transition from fascism to democracy. As important of a historical moment as this was for the Italian people, the legacy of this election is more complicated, since this election marked the first major covert operation of the American Central Intelligence Agency (CIA), formed just seven months prior. Policymakers in the United States feared that without an intervention to ensure the Christian Democratic Party, led by Alcide De Gasperi, maintained control of the government, the Popular Democratic Front coalition between the communist and socialist parties would prevail and align more closely with the Soviet Union in the emerging Cold War. The fear of a left bloc victory was not unwarranted. In local elections the previous year, the leftist parties saw increases in their vote share in Rome, Naples, Genoa, Turin, Florence, Palermo, and Sicily.Footnote¹ Early polling in 1948 showed the left as a favorite in the upcoming election.Footnote² In addition to the overt carrots and sticks the U.S. was employing to affect Italian voters (most notably the conditional provision of future Marshall Plan funds), the U.S. embarked on a covert campaign to ensure a Christian Democratic victory. This campaign included secret funding of anti-communist unions, forged documents with salacious details about leftist politicians’ personal lives, pre-written form letters signed by Italian-Americans in the U.S. and mailed to their relatives in Italy, and at least $1 million dollars in covert funding sent to the Christian Democrats.Footnote³ The Christian Democrats would go on to win the 1948 election by a large enough margin to exclude the leftist parties from the government, and would continue to receive cash transfers and other covert campaign support from the United States through the early 1980s.

This may have been one of the first major covert operations undertaken by the United States during the Cold War, but it wouldn’t be the last. In addition to the decades-long activities in Italy, the U.S. also covertly intervened in elections to back pro-American parties in other U.S. allies such as France and Japan. The U.S. would also covertly support extra-constitutional changes of political leadership, backing coups in Iran (1953), Guatemala (1954), Chile (1973), and elsewhere. In fact, over the course of the Cold War, the U.S. was involved in an active covert operation aimed at regime change, regime maintenance, or subversion in 12 per cent of country-years.Footnote⁴ At the peak in 1953, around 20 per cent of countries were experiencing an American covert action campaign, and 15 per cent of states were targeted by the U.S. for covert regime change alone.

Covert action aiming to affect who governs a state (regime change or regime maintenance) is clearly an appealing policy tool, as it allows states to change the policies of other states in the most direct way: by changing the policymakers. Covert subversion, which aims to raise to governance costs of the target state, is also appealing due to its ability to target an adversary’s capacity while managing the risk of escalation. This study shows that the U.S. generally pursued covert regime change, covert regime maintenance, and covert subversion when and where it was rational, given the U.S. objective of aligning as many countries as possible with the liberal international order during the Cold War. Specifically, the U.S. tended to pursue covert regime change in countries that were hostile to the American-led order and in which there was a viable alternative regime the U.S. could back; the U.S. tended to pursue covert regime maintenance in countries that were supportive of the American-led order and in which there was a viable alternative that could threaten that government; the U.S. tended to pursue subversion in countries that were hostile to the American-led order, but in which there was no viable alternative for the U.S. to back in a regime change effort.

Covert action is an important tool in the U.S. foreign policy toolkit, but remains understudied. This study advances the scholarship on covert action by (1) quantitatively testing an argument about when and where the U.S. was likely to pursue covert regime change, regime maintenance, and subversion during the Cold War, (2) creating the first dataset of all known operations of these three types undertaken by the U.S. during the Cold War, and (3) defining and creating a dataset of viable alternative regimes which could be installed via regime change or guarded against via regime maintenance. The remainder of this paper will (1) review the prior literature on covert action and regime change more broadly, (2) lay out a theory of how a combination of state preferences in the target state and the presence of a viable alternative regime in that state drive U.S. covert action decisions, (3) discuss the data used to test that theory, including novel data on covert action and viable alternative regimes, and (4) test the theory and present the results.

This article specifically aims to answer the question ‘What are the causes of covert action?’ It develops a broadly applicable argument, but only tests that argument on the narrower case of American covert action during the Cold War. There are two main contributions of this article to the literature on covert action. First, the theoretical contribution is in expanding on logics of covert regime change and extending these logics to two new domains: covert regime maintenance and covert subversion. While this article is not the first to acknowledge the role of policy disagreements and local actors in driving covert regime change, it is the first to define and measure these concepts in a way that allows for sophisticated large-N empirical testing, as well as the first to use these same variables to explain covert regime maintenance and covert subversion within the same model. This is the second contribution: the use of an original dataset of covert action cases and viable alternative regimes to empirically test arguments about the causes of covert action with large-N statistical models. To date, the causal work on covert action has been primarily focused on qualitative methods such as process tracing.

Regime change and covert action

There is a robust and growing literature on foreign-imposed regime change, primarily focused on the consequences of overt cases. We know that imposed regime change rarely results in democratization and can increase internal instability in the target state, but can also reduce the bellicosity of the target, alter the alliance partners of the target, affect economic activity between the target and imposer, and create externalities for the region surrounding the target.Footnote⁵ There is also a literature focusing on when regime change occurs during interstate wars and interstate crises.Footnote⁶ Focusing on wars or crises is worthwhile when examining overt regime change, since open regime impositions tend to occur at the termination of a war or crises between the imposer and the target, but covert attempts to influence who governs a foreign polity are often covert precisely to avoid escalation to a crisis or war.

In one of the most comprehensive study on the origins of forcible regime promotion not limited to war or crisis, Owen looks at regime promotion from 1510 to 2010, and argues that ‘transnational ideological polarization’ leads to bursts of regime changes as ideological state networks engage in competition over which ideology should reign supreme.Footnote⁷ States install regimes similar to their own in foreign countries, because it makes their own regimes more secure and less likely to be replaced by a competing regime type. In a more recent contribution, Willard-Foster highlights a key driver of imposed regime change: the strength of the target state’s domestic opposition.Footnote⁸ A strong domestic opposition is important both because it lowers the costs of imposition for the imposer (who can rely on that opposition for some of the heavy lifting), and because it raises the target state’s costs of making concessions to the imposer. These costs are raised since the opposition could punish the current leader for their concessions, and the lack of concessions to the potential imposer often makes placating them difficult.Footnote⁹

The literature on overt regime maintenance is much smaller, likely owing to the fact that many types of regime maintenance (development aid, military alliances, etc.) are studied as separate phenomenon. One contribution worth specifically noting is by Owen and Poznansky, who argue that for the U.S. to stop maintaining an authoritarian regime abroad, that regime must be facing an exogenous domestic crisis, and American free-market liberal democracy must be the sole credible alternative to replace the outgoing autocratic regime.Footnote¹⁰ The literature of subversion has historically been small as well, but has seen significant recent contributions. Lee shows how hostile neighboring regimes can undermine state sovereignty over some parts of a state’s territory, and Wohlforth argues that great powers are often reluctant to engage in subversion against other great powers due not only to escalation fears, but also concerns about cost-imposition responses from the target state.Footnote¹¹

The literature specifically concerned with covert action intended to install or prop up leaders shows similar findings as the literature on overt regime change: after being targeted for covert action, countries tend to experience a decline in democracy, an increase in imports from the United States, primarily driven by government purchases, and increased likelihood of civil war, future irregular regime changes, and widespread human rights abuses.Footnote¹²

The literature most relevant to this study is that which examines the causes of covert action. Poznansky argues that covertness is preferred when there is no exemption under international law which the acting state can invoke to claim their actions are legal, and Joseph and Poznansky show that covert operations are less likely to be employed in locations with a higher density of information and communication technologies that raise the prospects of the operation being revealed.Footnote¹³ The most significant work on the causes of covert conduct thus far is likely by O’Rourke.Footnote¹⁴ She argues that there are two preconditions for covert regime change: a chronic, irreconcilable divergence in security interests between the U.S. and its potential target, and the presence of a plausible political alternative that shares U.S. policy preferences. These two criteria are similar, but somewhat distinct from the two key concepts in the argument advanced by this study: diverging state preferences (which need not rise to the level of chronic and irreconcilable, or necessarily be directly about security), and the presence of an viable alternative regime (which does not necessarily need to share U.S. preferences).

A three-pronged typology of covert action has been developed by Nutt, who presents a typology of covert checking, covert regime change, and covert weakening.Footnote¹⁵ Nutt’s work differs from the typology advanced below in a few key ways: First, he excludes coordinated propaganda campaigns from his definition of covert action; Second, covert checking is a much narrower idea than regime maintenance, since checking only occurs in moments of what Nutt calls ‘high alignment instability’, while general regime maintenance can occur at any point; Third, I focus on each country in each year as a potential target of regime change, while Nutt focuses only on what he determines to be cases where covert action was genuinely reasonable; and Forth, the argument Nutt uses to explain when different types of covert action are chosen is fundamentally different than the argument presented here. Nutt focuses on the role of alignment instability, where I focus simply on alignment itself combined with the presence of the viable alternative. How the potential future change in alignment could be incorporated into the typology used by this study is left for future work.

A final point that should be noted about covert action is how frequently it is employed during elections in the target country, such as the example of the 1948 Italian elections which opened this piece. As one scholar of the American intelligence and counter-intelligence sphere put it: ‘At times the Covert Action Staff at CIA Headquarters has resembled nothing less than a group of political campaign consultants, producing slick materials for favored foreign candidates’.Footnote¹⁶ Recent research has concluded that the U.S. intervened, either overtly or covertly, in 81 competitive elections between 1946 and 2000, that the interventions resulted in 3 per cent increase in the popular vote for the supported candidate on average, and that such interventions where most likely when American interests were threatened by one plausible election victor but not another, and the non-threatening candidate or party was willing to cooperate with the intervening power.Footnote¹⁷ The degree of covert election interference by the United States highlights an important fact: democracies are often targets of covert action, so much so that these targets have been labeled the ‘hidden victims of the democratic peace’.Footnote¹⁸ Democracies make easy targets for covert action due to their open societies and use of elections, which can function as pre-scheduled opportunities for regime change or regime maintenance. There are some scholars who have worked to reconcile the existence of democracy-on-democracy covert action with the democratic peace theory.Footnote¹⁹ However, the fact of the historical record is clear: there is no effective mechanism deterring democracies from pursuing aggressive covert operations against over democracies.

Covert action, state preferences, and viable alternatives

This section will define the three core concepts (covert action, state preferences, and viable alternative) in the argument, and then present the argument itself.

Covert action has been defined and redefined numerous ways by policymakers and scholars.Footnote²⁰ For the purpose of this study, the original US National Security Council (NSC) definition is useful for both its specificity and completeness:

Note that the emphasized portion of the document specifies the actions that could constitute a covert operation including a range of informational, political, economic, and paramilitary tools. This definition is also useful in that it highlights what it would mean for such an operation to be covert. Specifically, that ‘responsibility for them is not evident … and that if uncovered the US Government can plausibly disclaim any responsibility’. Additionally, this definition conveniently includes a list of activities, such as espionage, which should not be considered ‘covert operations’ even though they are both covert and an operation.

Where the NSC 10/2 definition of covert operations is somewhat lacking is in its description of the goals of covert action. This study uses a novel typology of covert action, which distinguishes operations by their intent: covert regime change is the attempt to alter the effective leader of the state; covert regime maintenance is the attempt to keep the current leader of the state in power; covert subversion is the attempt to internally undermine a state, raising its governance costs and sapping its ability to act effectively on the international stage.

While the methods behind covert regime change (backing coups or rebel groups, election interference, etc.) and covert regime maintenance (support in counter-insurgency operations, elections interference, etc.) are likely somewhat familiar to readers, subversion is probably less intuitive. The methods are often similar to those utilized when pursuing regime change, but the scope and aim is lesser. For one example, consider the American approach towards much of Eastern Europe during the bulk of the Cold War. The US initially pursued regime change in the Soviet satellite states by covertly backing viable anti-Soviet dissident groups wherever they could be found, but as the Soviet-back governments consolidated control within the region, the U.S. struggled to identify plausible alternatives. The goal shifted away from regime change in NSC 174, which in 1953 articulated the ultimate goal of regime change, but also established the criteria for judging possible covert activities as whether the action was likely to slow down Soviet extraction of human and natural resources from the target. The goal of regime change was completely dropped by NSC 5608, which in 1956 established the criteria as merely that the operation contributes towards weakening Soviet control.Footnote²² After 1956, the U.S. would continue coordinated propaganda campaigns in this region via Radio Free Europe, as well as low-level backing of dissident groups. These actions merely sought to raise the governance costs of the Soviet Union and their client states, and therefore is subversion and not regime change. In this case, the content of U.S. activity does not change, but due to the increased state capacity of the targets and the associated disappearance of viable alternative regimes to be supported, the goal shifts from replacing the leadership in Eastern Europe to merely subverting their authority.Footnote²³

The definition of state preferences in the context of this study is much simpler than the definition and typology of covert action presented above. While state leadership holds preferences over many concerns and policy domains, only one domain is relevant for this study: the state’s position towards the American-led international liberal order. This is the central, defining ideological dimension of international politics since the end of the Second World War, much in the same way that the left-right spectrum is the central ideological dimension of domestic politics in most developed democracies.Footnote²⁴ As Westad notes, most American leaders during the Cold War clearly believed that their friends and enemies abroad were defined primarily by their proximity to the ideological premises the United States sought to champion on the global stage.Footnote²⁵ This meant support for the international liberal order and rejection of rival orders (e.g., anti-Communism).

The final concept to be defined here is the viable alternative regime (VAR). A VAR is an entity which is distinct from the current regime of a state in either its ideology or source of societal support (making it a meaningful alternative), which could plausibly oust the current regime and rule in its place (making it meaningfully viable). VARs can either operate within the current institutions of the state (these include out-of-power political parties and coup-minded military officers) or outside the current institutions of the state (such as an armed rebel group, large non-violent movement, or a politically excluded ethnic group). Each of these groups have distinct sources of viability. For political parties, their viability depends on the presence of free and fair elections for them to compete in, as well as a societal base large enough to make them competitive in the election. For groups of military officers, their viability depends on weak or absent civilian control of the military. For rebel groups, their viability depends on their ability to effectively challenge the state in armed confrontations and an aim to control the political center of the state.Footnote²⁶ For large non-violent social movements, their viability depends on both the size of the movement and the objectives pursued (they must seek leadership change or institutional change). Finally, for politically excluded ethnic groups (or a coalition of politically excluded ethnic groups), their viability depends primarily on their size as a proportion of the population.

Now that the core concepts in the argument have been defined, the argument itself can be presented. Covert regime change should be more prevalent against target states who are aligned against the U.S.-led international order and who have a viable alternative for the U.S. to install. Regime change is valuable as a policy options since it allows the imposer to alter the preferences, not just the incentives, of another state. In an ideal case, regime change can be viewed as an alternative to bargaining, since it removes the differences in policy positions states would be required to bargain over by bringing the target’s preferences in-line with the preferences of the imposer.Footnote²⁷ Covertness is almost always preferred when pursuing regime change, as it lowers the costs of the operation, manages escalation risks, and (when not exposed) does not damage the newly installed client leader’s political legitimacy in the same way being openly helped to power by a foreign state may.Footnote²⁸ Additionally, pursuing regime change through most methods is frowned on by international law and norms. Unless U.S. policymakers can find a legal exemption to justify a regime change operation, they would prefer to conduct the operation in secret than openly flaunt international law.Footnote²⁹

With regards to the alignment against the U.S.-led liberal order, the argument that such an alignment would increase the likelihood of being targeted for regime change is intuitive, as the U.S. would want to replace the regimes of the countries with which it had the highest degree of preference divergence. However, it should be noted that there is a small category of cases which buck this trend: preventative regime changes. These generally occurred when the U.S. had a strong ideological affinity with the leaders in another state, but the U.S. either feared that the leader would provoke backlash (e.g., Trujillo in the Dominican Republic, who the U.S. feared might produce a Castroite backlash), or was not competent enough to remain an ideal U.S. ally (e.g., Diem in South Vietnam, who the U.S. thought was inadequately managing the Vietnamese war effort).Footnote³⁰

The role that VARs play in making covert regime change more likely is two-fold. First, the alternatives dramatically lower the cost of pursuing regime change, since they provide a ready-made actor to install in leadership with a preexisting base of societal support. In such cases, the U.S. would not have to invest in building up a new leadership group from the ground up. Lowering this cost of covert regime change is important, as the U.S. must ultimately choose covert regime change over bargaining (which is much less costly in general, but also produces less concessions for the U.S.). Second, VARs can raise the cost of bargaining for the potential target of regime change. The argument here is partially an extension of Willard-Foster’s argument about how the political opposition can raise bargaining costs: VARs will often use concessions granted to the U.S as a cudgel against the incumbent government in their quest to take the mantel of governance for themselves, creating domestic costs for bargaining.Footnote³¹ This is the dual nature of VARs: they are willing to condemn the incumbent regime for concessions to the United States, and they are willing to give concessions in exchange for American support to help themselves into power. These are political actors capable of holding power on a national scale, and are generally rational in their approach to obtaining that power. The VAR need not even be closer to the U.S. ideologically prior to the operation to be reasonable choice of partner for the U.S., since the viable alternative may be willing to comprise their foreign policy positions heavily in order to obtain power, especially when their opposition to the current government is about an issue orthogonal to the wider Cold War conflict, such as ethnically-based exclusion from state power. We can see this dynamic in play in Angola for example, where UNITA started as a far-left, Maoist organization backed by China and transformed into an anti-Communist organization backed by the United States. The effect of state preferences and the presence of a VAR on the probability of covert regime change can be summarized with hypotheses 1A and 1B:

Hypothesis 1A:

The United States is more likely to pursue covert regime change in potential target states which are ideologically distant from the American-led international liberal order than in potential target states which are ideologically close to the American-led international liberal order.

Hypothesis 1B:

The United States is more likely to pursue covert regime change in potential target states in which there is a viable alternative to the incumbent regime than in potential target states in which there is no viable alternative to the incumbent regime.

Covert regime maintenance should be more prevalent against target states who are aligned towards the U.S.-led international order and who have a viable alternative the U.S. fears could take power. In contrast to regime change, most regime maintenance is done overtly in ways that are rarely identified as ‘regime maintenance’, such as providing economic aid, overt military assistance and alliances, and even basic diplomatic measures such as recognition and the exchanging of ambassadors. When must regime maintenance be pursued covertly? Generally, it is when either (1) the legitimacy of the target regime would be damaged be their dependence on U.S. support to stay in power, or (2) the tools necessary to maintain the regime require an intervention that undermines the sovereignty of target nation and violates norms of self-determination. Such tools include election interference to ensure the incumbent stays in power, covert counter-insurgency support against a popular rebellion, or covert support in cracking down on a social movement or politically excluded ethnic group who are seeking wide-ranging government reforms. For example, in the early 1950‘s the United States provided covert counter-insurgency support to the Philippines in order to help them defeat the communist Hukbalahap.Footnote³² In regards to the key drivers of covert regime maintenance tested below, it is intuitive that the U.S. should want to maintain regimes closely aligned with the American-led order, and that such maintenance would only be necessary when there is a viable alternative capable of challenging that regime, leading to hypotheses 2A and 2B:

Hypothesis 2A:

The United States is more likely to pursue covert regime maintenance in potential target states which are ideologically close to the American-led international liberal order than in potential target states which are ideologically distant from the American-led international liberal order.

Hypothesis 2B:

The United States is more likely to pursue covert regime maintenance in potential target states in which there is a viable alternative to the incumbent regime than in potential target states in which there is no viable alternative to the incumbent regime.

Finally, subversion should be more prevalent against target states who are aligned against the U.S.-led international order, but who lack a viable alternative for the U.S. to partner with. Generally, these cases occur when the U.S. aims to reduce the capacity of the target state, but the U.S. either (1) does not desire to replace the target state’s leadership given the costs of doing so, or (2) cannot identify a truly viable alternative to the current regime, despite working to undermine the target and raise the target’s cost of maintaining control over their territory. An example of the first motive in action would be during the early 1970‘s when the U.S. provided covert support for Iraqi Kurds engaged in armed conflict with the central Iraqi state. The degree of support to the Kurds was carefully calculated to make them a serious threat to Iraq in need of Iraqi military attention, but not so serious as to enable the Kurds to form their own state, which would destabilize Kurdish-dominated regions of both Iran and Turkey.Footnote³³ Examples of the second motive can be found is both the case of Eastern Europe mentioned above, as well as the case of subversion in China during the 1950‘s in which the U.S. provided covert support for multiple groups contesting the dominance of Mao Zedong’s People’s Republic of China, none of which were truly viable contestants for the control of China as a whole.Footnote³⁴ Regardless of which motive is present, subversion is expected to correlate with alignment against the international liberal order and a lack of a VAR:

Hypothesis 3A:

The United States is more likely to pursue subversion in potential target states which are ideologically distant from the American-led international liberal order than in potential target states which are ideologically close to the American-led international liberal order.

Hypothesis 3B:

The United States is more likely to pursue subversion in potential target states in which there is no viable alternative to the incumbent regime than in potential target states in which there is a viable alternative to the incumbent regime.

For all six of the above hypotheses, the alternative hypothesis is simply that the ideological distance and presence of a VAR are unrelated to the likelihood that the United States purses covert regime change, covert regime maintenance, or subversion. The statistical tests conducted below treat the null hypothesis as ideological distance or VAR having an effect of zero on the likelihood of a specific type of covert action occurring (meaning the statistical effect is not meaningfully distinguishable from no effect).

Data and measurement

Covert action

This section will cover the data used to test the above hypotheses, starting with the data on American covert action during the Cold War. Despite the generalizable variables in the hypotheses (ideological similarity and VARs), it may seem puzzling that the data used is specific to the United States during the Cold War. The scope is limited to this actor and period for three reasons. First, the geopolitical context facing the United States in this period is somewhat constant: the U.S. is one of two superpowers engaged in a global ideological struggle. This is not true for less powerful states, or for the United States before or after this period. Second, a combination of American declassification rules, Congressional inquiries, and journalistic coverage have likely revealed a much more complete set of cases for the United States than for more closed countries, specifically the Soviet Union.Footnote³⁵ Three, as other Joseph and Poznansky have shown, covert action is less likely in environments dense with information and communication technologies, which raise the risk of exposure.Footnote³⁶ As the density of these technologies have expanded in recent decades, covert action has been forced to take new forms, making the more recent era of covert action notably different from the past. While the risk of conducting traditional operations has increased, the ability to conduct operations utilizing these technologies, especially influence operations via social media, has increased dramatically, as was clearly on display during the Russian intervention in the 2016 American presidential election.Footnote³⁷

The data considers any independent state as identified by Gleditsch and Ward as a potential target of American covert action for each year from 1946 to 1989.Footnote³⁸ Notably, this excludes covert action taken in colonial possessions prior to their independence, such as the operations to influence the first independent governments of Angola, Guyana, and Mozambique.Footnote³⁹ In order to determine if and what type of covert action was occurring in each country-year, a series of secondary accounts were analyzed, the resulting case list was cross-listed against similar data collection efforts, and occasionally follow up research was conducted with primary materials made available by the United States State Department via the Foreign Relations of the United States (FRUS) series.Footnote⁴⁰ On the bases of these sources, each type of covert operation considered here (regime change, regime maintenance, and subversion) was coded as either occurring or not having occurred in a given country in a given year. In cases of extended campaigns, evidence of distinct actions occurring in each year of the campaign was not necessary if there was no indication that operations ceased between years in which distinct actions could be verified.Footnote⁴¹

Here it is worth emphasizing why this data is useful despite the fact that some successfully concealed operations may be missing from the data. A criticism that can be levied at any attempt to form a list of covert operations is that such operations are by their very nature concealed and that anyone constructing such a list can never be sure they have the full list of cases, let alone an random selection of those cases. The operations easiest to detect are those that failed to remain covert at the time, or that the U.S. was willing to declassify years later. This can be responded to in two ways. First, while it is true that the very nature of covert action means that a definitive list will likely never be compiled, the importance of the subject matter demands the compiling of as complete a list as possible. Political scientists and international relations scholars have widely avoided studying covert action, and the intelligence domain more generally, to the extent it is warranted due to concerns about data unavailability and biased case selection.Footnote⁴² The dataset compiled for this study shows that during 12.4 per cent of country-years between 1946–1989 an identifiable American covert operation was occurring. This wide presence, combined with what the prior literature argues about the impact of these operations, justifies covert action being studied by social scientists with the same commitment that other foreign policy tools, such as war, sanctions, or alliances, are studied. Second, as already mentioned above, a combination of American declassification rules, Congressional inquires, and press coverage have revealed many cases of U.S. covert action, often decades after the events took place. If this list is not complete, it is probably not a random sample. However, there is little reason to believe that the fundamental drivers of covert action would be different for the population of cases which have been revealed versus those which have not.Footnote⁴³

It is also worth emphasizing at this point how this collection of covert action data differs from three prior efforts by Berger et al., Levin, and O’Rourke.Footnote⁴⁴ Berger et al. build a dataset of CIA interventions which ‘installed or propped up’ a foreign leader. This is measured by a single variable which does not conceive of regime change and regime maintenance as two separate types of activity: their indicator takes the value of 1 if either occur, and 0 otherwise, with no way of disaggregating the two type of operation. Subversion is not coded. Additionally, Berger et al. only include successful cases (and not operations which failed to produce the desired effect) in this variable. There is a second variable coded for failed operations, but it is incomplete and is missing many cases identified by this study. Levin builds a dataset of both overt and covert election interference. This dataset is well constructed and heavily overlaps with the universe of covert electoral interference cases identified in the data used here, but does not include covert operations which seek to influence a target state in any way other than by affecting the outcome of national-level elections. Finally, O’Rourke has assembled what is probably the best researched and well-sourced list of American attempts at covert regime change during the Cold War. She does not, however, code covert regime maintenance or subversion. Additionally, the list of cases compiled by O’Rourke include multiple instances in which the target of regime change is not an independent state, but instead a regional unit associated with a state that existed before or after the period in question.Footnote⁴⁵ This study only examines cases of covert action targeted towards a state which is sovereign at the time of the operation. Furthermore, while O’Rourke only considers the phenomenon of covert regime change, she utilized a much broader sense of the term than was defined above. For example, she codes the example of subversion against China mentioned in the previous section as an attempt at regime change, and she considers covert electoral intervention of any kind (even when it is support the incumbent) as a regime change operation. This study considers a partisan electoral intervention on the behalf of the incumbent as regime maintenance, and only considers it regime change when a non-incumbent political party is supported.Footnote⁴⁶

With the data-generating procedure, the data’s value, and uniqueness established, we can now look at the variation in American covert action across space and time. Figure 1 shows the proportion of countries targeted by the U.S. for covert regime change, covert regime maintenance, or subversion for each year of the Cold War.Footnote⁴⁷ As many previous scholars of covert regime change have noted, the first half the 1950‘s was the heyday of American covert action to remove foreign leaders from power. This included the consistent, but somewhat restrained, attempts as regime change through much of Eastern Europe, as well as some of the most well known cases such as Operation Ajax (culminating in the 1953 coup in Iran) and Operation PBSUCCESS (culminating in the 1954 coup in Guatemala). The count of regime change operations drops heavily in 1957 as U.S. policymakers resign themselves fully to Communist entrenchment in Eastern Europe, and transition from regime change operations to subversion operations in this entire theater.Footnote⁴⁸ Regime change would continue to be prevalent through the 1960‘s, dip in the 1970‘s, and rebound in the 1980‘s as the idea of rolling back Communism found new life in the Reagan Administration.

Figure 1. American covert action across time.

Display full size

The proportion of covert regime maintenance operations consistently builds over the first decade of the Cold War as the U.S. begins propping up allies across the globe, and then experiences two peaks closely following the peaks in covert regime change in the 1950‘s and 1960‘s, reflecting covert regime maintenance efforts intended to keep the covertly installed leaders in power. From that point on, covert regime maintenance slowly petters out until it is nearly non-existent by the end of the Cold War. This occurs because covert regime maintenance is slowly phased out by overt regime maintenance, as the U.S. shifts to an emphasis on overt aid and open democracy promotion, and also becomes more willing to openly arm and train the militaries of countries which are closely aligned with the U.S.-led order.

Finally, subversion experiences a large spike in 1957 as it becomes the dominant strategy throughout Eastern Europe, and then steadily declines for the rest of the Cold War. Much of this decline is simply due to Eastern Europe being a smaller proportion of total states as decolonization occurs. There is a small, but sustained increase in subversion during the 1970s, during the same period that covert regime change operations are at their lowest.

Figure 2 shows the variation in the three types of covert operation across region. As regions deemed less important in the overall strategic picture of the Cold War, Africa and Oceania are targeted for the least amount of covert action. The Americas see a significant amount of both regime change and regime maintenance, with the latter often following the former. Asia sees the highest rate of covert regime maintenance operations, largely due to the extended operations the U.S. undertook to prop up and maintain friendly regimes in Japan, Taiwan, the Philippines, and South Vietnam. Europe sees the overall highest rates of regime change operations and subversion, likely due to three factors: Europe’s position as the most strategically valuable region during the Cold War, Eastern Europe being under constant attempts at regime change (through 1956) or subversion (most of the Cold War after 1956), and Western Europe being made up primarily of democracies, whose elections serve as pre-scheduled opportunities for regime change or regime maintenance. Finally, the Middle East and North Africa see less covert activity than Europe, Asia, or the Americas, but more than either Africa or Oceania.

Figure 2. American covert action across region.

Display full size

Viable alternative regimes

In addition to the covert action data described above, the second piece of novel data introduced in this study is the data on viable alternative regimes (VARs). When the concept of the VAR was defined above, the five types of VARs measured in this data were enumerated: armed rebel groups, non-violent social movements, politically excluded ethnic groups, out-of-power political parties, and sufficiently independent military officers.Footnote⁴⁹ The presence and viability of each of these types of groups, for every country-year, can be determined from data sources compiled by previous research projects.

The UCDP/PRIO Armed Conflict Dataset is used to determine the presence and viability of rebel groups.Footnote⁵⁰ For a group to be deemed a viable alternative to the central government of a state, it must (1) meet the minimum UCDP threshold of being involved in conflict resulting in at least 25 battle-deaths per year, and (2) be challenging the state they are in conflict with for control of the political center. The first criteria, which measures the capacity of the group to either inflict casualties on the state or absorb those loses within their ranks, proxies for the group’s general capacity to govern and exercise control. The second criteria ensures that the group actually aims to be a VAR, and not merely establish viable control over a portion of the country.

In order to determine the presence of a non-violent social movement, this study relies on the NAVCO 2.0 data.Footnote⁵¹ For a campaign to be deemed a viable alternative, it must (1) consist of over 100,000 people, and (2) actively be campaigning for either regime change or significant institutional reform. The first criteria ensures that the group is large enough to be considered a significant social movement, and the second ensures that the campaign has ambitious enough goals to motivate a base of societal support that could back its rise to governance.

A politically excluded ethnic group, or a coalition of politically excluded ethnic groups, could also form a VAR. The Ethnic Power Relations Core Dataset is used to determine the presence of such a group or coalition.Footnote⁵² For a politically excluded ethnic coalition to exist in a given country-year, (1) one or more politically relevant ethnic groups in the state must be either effectively powerless at the national level, actively discriminated against by the state, or have chosen to self-exclude themselves from participation in the state, and (2) the total population of all groups meeting the first criteria must exceed 20 per cent of the state’s total population.

In order to determine if a viable alternative exists in an out-of-power political party, the VAR data relies directly on the list of competitive elections most commonly used in the emerging election interference literature: the PEIG dataset.Footnote⁵³ This data source is used in part to ensure the results below are as comparable with the findings in the wider electoral inference literature as possible. The PEIG itself relies on the Database of Political Institution’s executive electoral competitiveness index, with a minor modification.Footnote⁵⁴ For an executive election to be considered competitive, it must receive a 7/7 on the competitive index, and the leading party or candidate must not receive more than 75 per cent of the vote. This study considers one or more opposition parties to be a viable alternative in any election deemed competitive by these criteria.

Finally, the last category of viable alternatives are military officers capable of removing and replacing the current leadership via a coup d’état. While a large enough group of officers would have to material force necessary to take control of the government in most states, the factor preventing them from doing so is the norm of civilian control of the military. A dynamic measure of civil control is constructed by Kenwick, and this measure is the basis for how this study separates countries in which a group of military officers are a viable alternative to the incumbent regime from countries in which that is not the case.Footnote⁵⁵ In order to create a cutoff between viable and non-viable alternative military governments, Kenwick’s dynamic civilian control metric was cross-referenced with a list of all global coup attempts.Footnote⁵⁶ After determining the median level of civil control in country-years that experienced a coup attempt and the median level in country-years that did not, the midpoint between these two values was used as the cutoff. These types are VAR and their measurement are summarized in Table 1.

Table 1. VAR categories and measurement.

Download CSVDisplay Table

Figure 3 shows the variance of each of the five types of viable alternative regimes over time. While there are small peaks and valleys in the prevalence of each type of viable alternative over the course of the Cold War, the consistency across time is what most stands out. Military officers are consistently the most present, being viable in 30 per cent to 40 per cent of states in any given year. Politically excluded ethnic groups are the second most common VAR, present in around 20 per cent of countries during the first decade of the Cold War, and around 25 per cent of countries once the major wave of decolonization has begun. Viable alternative parties contesting competitive elections are usually present in between 10 per cent to 20 per cent of countries in a given year, although the presence of such groups does trend downward overtime as the introduction of newly decolonized states to the state system reduces the proportion of democracies in the system. Finally, violent and non-violent movements contesting the state are the least common viable alternatives for the first half of the Cold War, at below 10 per cent of countries, but an increase in the presence of these types of groups over the 1970s and especially the 1980s brings their presence in line with the presence of alternative political parties.

Figure 3. Viable alternative regimes across time.

Display full size

Figure 4 shows the distribution of the five types of VAR across region. While Figure 3 already displayed how viable military officers were by far the most common viable alternative, here it can be seen that this phenomenon is primarily driven by Africa, the Americas, Asia, and the MENA region, each of which seen a viable military alternative in more than 40 per cent of country-years. Europe only has a moderate presence of viable military officers, and they are entirely absent from the Oceania region. Africa has an extremely high rate of politically excluded ethnic groups, much higher than any other region, and is the only region in which that is the most common type of VAR. There are also sizable counts of these groups in Asia, MENA, and the Americas. Viable political parties are most represented in Oceania (due to the low count of countries, many of which are stable democracies), Europe and the Americas. Finally, viable rebel groups tend to exist in the same four regions as viable militaries, and non-violent social movements were much more common in Asia than any other region, in large part due to extended campaigns in Afghanistan, China, Indonesia, the Philippines, and South Vietnam.

Figure 4. Viable alternative regimes across region.

Display full size

State preferences and other measures

The final core concept which needs to be operationalized in order to test the above hypotheses is a measure of support for, or alignment with, the American-led international order. This study utilizes estimates of state preferences generated from United Nations General Assembly Voting Data for this purpose.Footnote⁵⁷ These estimates are constructed as ideal points along a spectrum of support for the U.S.-led liberal international order, can correctly identify significant foreign policy shifts, are less sensitive to changes in the UN agenda than previous measures, and have a high level of face validity.Footnote⁵⁸ Higher values on the scale indicate a closer alignment with the international liberal order.

Additionally, multiple control variables are used in the analysis below. In order to account for the military capacity of the potential target of covert action, the Dispute Outcome Expectations (DOE) score is used.Footnote⁵⁹ Using the same six base components as the Composite Indicator of National Capability (CINC) originally presented in Singer, Bremer, and Stuckey.Footnote⁶⁰ Carroll and Kenkel produce a much more sophisticated measure of state power. Using machine learning techniques, they determine the combination and weighting of the six components that best predicts which country would prevail in a Militarized Interstate Dispute (MID) between a given dyad in a given year.Footnote⁶¹ The DOE score successfully predicts who wins MID’s by an order of magnitude better than the CINC score and is also presented in a more intuitive unit: where CINC is the raw average of six ratios, DOE is presented as the probability of victory for each state in a dyad-year (and the probability of a stalemate). The probabilities of American and target victory (with stalemate as a comparison category) should be included in a model of covert action, as how the U.S. is expected to fair in a conflict with the target country (should the operation be revealed and provoke retaliation) is important to the cost calculation of American policymakers when they are decided whether or not to pursue covert action in a given case. Here, we should expect that the U.S. substitutes subversion for full-scale regime change when confronted by a target who is more likely to hold their own in an escalated conflict with the U.S.

The level of democracy in the potential target, the amount goods from the potential target imported to the United States, and the degree of security hierarchy between the United States and the potential target, are included as confounding variables that represent alternative measures of state preferences.Footnote⁶² However, according to the above argument and prior literature, none of these variables should be seen as good measures of state preferences in a way that predicts covert action types. As previously noted, democracies are frequently targets of U.S. covert action. Despite ample literature on economic imperialism, there is little evidence that U.S. access to raw materials or produced goods affect the probability of undertaking a covert action. Similarly, the two elements of Lake’s security hierarchy measure (count of unique alliances and number of U.S. military personal stationed in a country) are likely to capture the investment of the U.S. in a country, but without the political direction of the regime in that country, it cannot differentiate between cases in which the high degree of salience to the U.S. should promote anti-regime action (regime change or subversion) or pro-regime action (regime maintenance). In fact, shared alliance membership with a potential intervener has been shown to increase the likelihood of overt regime change, making it a poor measure in the context of the above argument.Footnote⁶³

Two additional variables are included to account for a country’s susceptibility to covert action, specifically, how likely an operation undertaken in that country is to be exposed. The first of these variables is the Media Density Index (MDI) which measures the penetration of television, radio, and newspapers for each country-year.Footnote⁶⁴ The density of these communication technologies is thought to be a proxy for the probability that an operation is revealed, as argued by Joseph and Poznansky, since more people have access to the tools to learn about American activities, and the presence of these outlets necessitates the existence of a larger domestic press within the country.Footnote⁶⁵ Investigative entities outside of government are often necessary to reveal operations, as targets of covert conduct will often collude with the perpetrator of the conduct to keep the actions from the public.Footnote⁶⁶ The second variable meant to proxy for susceptibility for covert action is the real GDP per capita of the target.Footnote⁶⁷ Wealthier societies may be more or less resistant to coordinated attempts to effect their politics from abroad.

The final variable is simply a lag of covert action in the previous period, since covert campaigns usually persist over multiple years and the existence of an operation in one year is expected to predict the existence of an operation in the next year.Footnote⁶⁸

Analysis

Results of a single, straightforward model are presented in Table 2. This model is a multinomial logit regression of the occurrence of American covert regime change, covert regime maintenance, and subversion in each country-year during the Cold War (with no covert action as the reference category). This allows each year to be a new decision to pursue any type of covert action or no action. To account for the fact that covert action is more likely if it had occurred in the previous year, and lag of any type of covert action is included across models. The results support all hypotheses outlined above. We can reject the null hypotheses that political alignment and VARs have no effect on the occurrence of covert action. Covert regime change is most likely when the target state is ideologically distance from the American-led order, and when that state has a viable alternative present. Covert regime maintenance is most likely when the target state is ideologically close to the American-led order, and when that state has a viable alternative present. And finally, subversion is most likely when the target state is ideologically distance from the U.S., but there is no viable alternative.

Each of these findings is intuitive and expected. This does not, however, make them trivial. These findings show that the U.S. covert action policy was systematic and strategically reasonable during the Cold War. While every case is unique, each case did not emerge from fully idiosyncratic processes. The CIA was no ‘rogue elephant’ pursing covert adventurism wherever it could get away with it. The American interventions of the Cold War were calculated to maximize the benefits to the U.S. in the global conflict against the Soviet Union.Footnote⁶⁹ Additionally, the expected results in this analysis serve to bolster the face validity of the covert action data and viable alternative regime data introduced in this study. If something surprising or unexpected was found, that could give readers reason to question the validity and coding criteria used to generate the data. This is especially true for the findings concerning covert regime change, since the logic driving the above hypotheses is partially an extension of established scholarship.

While the model in Table 2 supports the hypotheses with regard to covert action occurrence, many readers may be more interested in a model of covert campaign onset (when a covert action begins). To that end, the start year of each campaign is also modeled and presented in Table 3. The results are substantively similar, showing that a state’s position toward the liberal international order and the presence of a viable alternative are significant indicators of both covert action occurrence (start and continuation years), and of the initiation of a new campaign against a given target.

How do the other variables included in the model fair? The DOE score variables, which proxy for the material power relationship between the United States and a potential target, also shows intuitive results: As the U.S. is more likely to prevail in a hypothetical militarized dispute with the target, the U.S. is also more likely to employ covert regime change or covert regime maintenance against that target. This makes sense, as a larger probability of U.S. victory in a hypothetical conflict represents a lower cost of escalation if the covert operation should go awry. Similarly, the U.S. is mostly likely to utilize subversion when the odds of a U.S. victory are lower, which squares with the idea that subversion likely entails less escalation risk than other types of operations. As the odds of a U.S. defeat grow higher, all three types of covert operation become less likely.

More democratic countries, as measured by Polity, are less likely to be targets of subversion, but are more likely to be targets of regime maintenance. The regime change picture is more nuanced, as democracies are more likely to see the onset of a covert campaign, but less likely to see occurrence. This is because democratic elections are functionally pre-scheduled opportunities for regime change, and electoral interference is one of the cheapest and lowest risk ways of pursuing regime change or regime maintenance. Given this, there is little reason to pursue subversion against a democratic target, when pursing total regime change is relatively easy. The fact that democracies see more regime change operations start, but less total years of operations implies multiple things: that regime change is so successful in democracies (via elections) that they quickly cease to be true democracies if targeted, and that regime change operations in democracies are isolated attempts leading up to an election, and not extended campaigns over time.

Contrary to expectations, the Media Density Index is not associated with a decreased likelihood of regime change or regime maintenance covert action, and is actually associated with an increased probability of subversion occurrence. One conjecture could be that such outlets not only provide opportunities to reveal an operation, but tools to be utilized by convert influence campaigns, especially electoral interference campaigns.

As expected, Lake’s security hierarchy measure is associated with a greater likelihood of being targeted for American covert action. While this may seem unintuitive to some readers, the politics of alliance systems and especially hierarchy maintenance give a theoretical foundation to expect senior alliance partners, like the United States, to impose their will on junior partners embedded in hierarchical networks managed by the senior partner.Footnote⁷⁰

The two economic variables, American imports and GDP per capita, do not produce meaningful effects aside from countries with greater GDP per capita being targeted less from regime maintenance (likely because wealthy nations can maintain their own regimes). This can be seen are a refutation of one claim of the economic imperialism argument: the U.S. does not target states from which it extracts the most economic value. This does not, however, refute a key claim of the argument: that after being targeted, the U.S. does receive economic benefits from the target state having pro-American leadership.Footnote⁷¹ Finally, covert action in the previous year is, as expected, strongly correlated with cover action in the present year across all operation types.

In addition to the primary models above, multiple robustness tests were performed to assess how sensitive the results are to variable measurement. The first set of robustness tests separated electoral covert action from non-electoral covert action. To this end, separate models were fitted of covert action in election years, and non-electoral covert action. These models (presented in Tables 1 and 2 in the Appendix) show the same effects of a potential target’s position toward the U.S.-led order and the presence of a viable alternative in the target. Additionally, to test the sensitively of the model results to the viable alternative measurement, five additional models were fitted, each excluding one type of alternative from the measure (military, rebels, ethnic groups, non-violent campaigns, and political parties). The results of these models are presented in the Appendix with Tables 3. Across these models, this paper’s argument makes 30 predictions about the direction and significance of coefficients, and 28/30 of these hypothesis are supported. In the two exceptions, it appears that in order for the presence of a viable alternative to be empirically associated with an increased likelihood of regime maintenance, both politically excluded ethnic groups and out-of-power political parties must be included in the viable alternative measure. This could be taken to mean that these two groups are those the U.S. most sought to keep out of power via regime maintenance, but more research in needed to confirm this conjecture.

While it is reassuring that the observable implications of the above argument are supported in the model results presented in Tables 2 and 3 as well as the robustness tests in the presented in the Appendix, how well do the full models actually predict the three types of covert action? Since covert action in the previous year strongly predicts covert action in the current year, covert action occurrence is substantially easier to predict (due to the easy cases of predicting the reoccurrence of an action that occurred last year) than covert campaign onset. For this reason, the focus is on campaign onset below.Footnote⁷²

Figure 5 is a series of separation plots showing how well each type of operation can be predicted within the Cold War sample.Footnote⁷³ Each vertical line in the figure represents a country-year. Those colored darkly represent a year in which the given operation type started and those covered lightly represent a year in which the given operation type did not start. These lines are ordered left to right by the model’s predicted probability of a given operation type occurring, with the highest probabilities on the right. A ‘perfect’ model would sort all dark lines to the right and light lines to the left. As can be seen in the three figures, for each type of operation, the large majority of country-years in which an operation of the given type is started fall on the right side of the figures, indicating that the model gave those country-years a high probability of seeing the onset of that type of operation.

Figure 5. Prediction displayed via separation plots.

Display full size

But how much do the key explanatory variables contribute to the model? Unfortunately, as Figure 6 shows us, the majority of heavy lifting is being done but other variables. In this case, the full onset model from Table 2 is compared to a baseline model that lacks the ideal point and VAR variables. The Receiver Operator Characteristic (ROC) plots in Figure 6 show the relationship between the true positive rate and the false positive rate as the threshold to predict the occurrence of each category of covert action is varied. Here we can see that covert regime change and covert regime maintenance become meaningfully easier to predict with the inclusion of the two key variables, but subversion does not.Footnote⁷⁴

Figure 6. ROC plots comparing full to baseline models.

Display full size

Readers should be reminded that the above analysis is an observational study, and not an experiment. Given than state preferences towards the U.S.-led order and the presence of a VAR were not randomly assigned, we must be aware of multiple possible threats to inference in such analysis. The above model results should not be taken as definitive proof of the argument advanced in this study, but they can be interpreted as evidence consistent with the argument being true. In order to increase confidence in the argument, further research utilizing various methodologies is needed.

This article has contributed to the study of covert action by introducing novel data on American covert action during the Cold War, as well as new data on the presence of viable alternative regimes (VARs) throughout the Cold War. It had described the distribution of both these phenomenon across both space and time, and tested an intuitive theory about where the U.S. chose to employ covert action during the Cold War. The results showed that the U.S. was likely to employ covert regime change against potential targets aligned against the U.S.-led order in which there was a viable alternative regime for the U.S. to partner with. The U.S. was likely to engage in covert regime maintenance against potential targets aligned with U.S.-led order in which there was a viable alternative regime the U.S. desired to keep out of the power. Finally, subversion was most probably against states who stood against the U.S. order, but in which there was no viable opposition for the U.S. to partner with. While the empirical results for covert regime change provide additional support to existing arguments in the field, the findings for covert regime maintenance and subversion are entirely original, and show that logics based on similar variables guided American decisions about when and where to employ all three of these types of covert behavior during the Cold War period.