Media technology,covert action,and the politics of exposure

Abstract

States wishing to use force in the modern era frequently face strong incentives to exploit secrecy. Successful covert operations can reduce the likelihood of unwanted escalation with powerful rivals and help leaders conceal unpopular actions from domestic and foreign audiences alike. The many benefits of secrecy, however, can only be realized if covert operations remain covert. We argue that access to information and communications technologies (ICTs) is a critical factor that increases the chances that a covert mission will be exposed. As a result, leaders are much less likely to reach for the quiet option when a potential target has dense ICT networks. We illustrate our mechanism through US national security archival vignettes. We test our argument using a dataset of declassified US military and electoral interventions intended to subvert incumbent regimes throughout the Cold War. The core finding, that leaders are less likely to pursue covert action relative to alternative options when the chances of exposure are high, holds across five distinct measures of ICT networks as well as different model specifications and placebo tests. Our findings suggest that Cold War-style covert operations may well be a thing of the past in an age where communication and media technologies have proliferated to the far corners of the globe. We advance debates on communications technologies, covert action, and political violence.

Keywords

covert action information technology media military intervention

Introduction

Why do states seek to deny responsibility for their actions in some cases but not others? A wave of recent scholarship addresses this question by exploring the motivations states have for pursuing covert action and other forms of hidden statecraft to achieve their aims (Baum, 2004; Brown & Marcum, 2011). Democratic interveners intent on toppling fellow democracies, for example, may exploit covert action to avoid angering domestic constituencies (Downes & Lilley, 2010; Poznansky, 2015: 124). Covert action may also reduce the odds of triggering unwanted and unintended escalation (Brown, 2013; Carson, 2016). The potential benefits of plausible deniability, the main benefit of covert action, are many.

Yet, the literature’s emphasis on the incentives for secrecy downplays one of covert action’s greatest liabilities: the possibility of getting caught (Lowenthal, 2015: 231). When factors that increase the chances of exposure during covert operations are present, leaders may be less likely to pursue the quiet option and instead select the next best alternative. Sometimes this means refraining from intervention entirely. In other cases, overt action will be most appealing. Which option leaders choose when the chance of getting caught looms large is determined by many factors, including mission costs and the value of the good.

We identify the factors that increase the chances that a covert operation will be exposed. We argue that the chance of getting caught rises when the target of intervention has widespread access to information and communications technologies (ICTs). Premature exposure is particularly likely when government personnel, reporters, and citizens can communicate with one another easily and citizens are politically aware. Widespread access to ICTs enables entities in the target to witness events unfolding and collect credible evidence that can compromise plausible deniability. Taken together, we predict that covert mission planners are less likely to use covert action against targets with dense ICT networks. This is especially true for interveners that would incur high costs for getting caught conducting covert missions, including heightened escalation risks and reputational damage.

We test our argument by combining two datasets of US-sponsored covert operations in which the goal was to subvert a foreign regime (Downes & O’Rourke, 2016; Levin, 2016). Using a range of statistical tests, we find that the USA is consistently less likely to pursue covert action relative to both no intervention and overt action as the potential target’s level of ICTs increases. Our results are robust to the inclusion of possible confounders, including the target’s regime type, level of development, military prowess, relations with US rivals, and more. We also use placebo tests and selection models to rule out alternative mechanisms.

We focus on the United States because it is among the most capable and frequent covert actors in world politics. This is particularly true for the time period under consideration. It is also a powerful democracy that plays an essential role in maintaining norms and values in the international system. It therefore presents a critical case for our theory. Moreover, out of all the covert interveners throughout history, the USA has a unique record of declassification procedures, investigative journalism, and official histories, all of which help to ameliorate missing data concerns (O’Rourke, 2013: 2).

This test, which is effectively a quantitative case study of US interventions during the Cold War, conforms to our scope conditions. Exposure of secret efforts by the United States to undermine sovereign regimes plausibly entailed non-trivial risks, including escalation with the Soviet Union (Carson, 2016) and delegitimization of the newly installed government (Downes & O’Rourke, 2016: 54; Finnemore, 2003: 128). We therefore expect that US decisionmakers were sensitive to the possibility of getting caught while conducting these kinds of operations. Declassified archival research lends credence to these claims.

This article makes several contributions. First, we add to the emerging literature on how ICTs influence the production of political violence (Bailard, 2015; Crabtree, Darmofal & Kem, 2015; Shapiro & Weidmann, 2015; Warren, 2015; Yanagizawa-Drott, 2014; Krcmaric, 2017). While most studies focus on how ICTs affect intrastate processes, our focus is on interstate relations. Specifically, we explore how the ICT make-up of one state (the target) influences the decision of an external state (the intervener) to meddle in its affairs. Further, we identify a new role for ICTs in shaping conflict by showing how these technologies can help targets of intervention produce evidence to overcome credibility problems.

This article is also the first systematic analysis of how incentives and constraints shape a leader’s decision to authorize covert missions abroad. Moving beyond the traditional focus on incentives (Carson, 2016; Daugherty, 2004; Treverton, 1987; Downes & O’Rourke, 2016), we shed light on how the probability of exposure weighs on the minds of decisionmakers and therefore influences the decision to use covert action. As a result, we are building in the cost component to the decision-model that underlies existing studies of secret statecraft.

Finally, this article has implications for understanding secrecy in the modern era. On the one hand, the development of novel technologies such as drones and cyber capabilities has made it possible for states to conduct covert operations from increasingly remote locations. On the other hand, the proliferation of smart phones and Internet access make it increasingly difficult for states to achieve plausible deniability, particularly when they need to deploy personnel. Consider Russia’s recent intervention in the Ukraine. Despite Putin’s best efforts at plausible deniability, photos taken on cell phones that were then posted to the Internet generated incontrovertible proof that Russian soldiers were stationed in Eastern Ukraine (Szoldra, 2014). While states may still exploit secrecy wherever plausible deniability is desirable and feasible, Cold War-style covert operations may be a thing of the past.

This article proceeds as follows: the next section defines key terms and outlines the decisionmaking model inherent in the existing literature. Next, we identify the three pathways to premature exposure in the covert sphere, focusing primarily on the role of ICTs. We then describe how and why the prospect of getting caught might influence a leader’s decision to use covert action in the first place. After formulating our hypothesis, we outline our research design, describe our data, and run a series of empirical tests. We conclude by discussing the implications of our findings and detail a series of avenues for future research.

Why select covert action?

The first order of business is to define our terms. By overt intervention, we mean the use of military force wherein the sponsor makes its identity known. We exclude cases where leaders openly threaten force and focus exclusively on interventions designed to achieve objectives through the kinetic employment of force. By contrast, covert actions are ‘special activities conducted in support of national foreign policy objectives abroad which are planned and executed so that the role of the [intervener] is not apparent or acknowledged publicly’ (Daugherty, 2004: 13). All covert operations are tethered together by the same overarching objective: plausible deniability. Thus, when we say that the main objective of covert action is to keep missions secret, we mean that interveners wish to keep their sponsorship of an operation hidden. To be clear, covert operations conceal the sponsor’s participation, which is distinct from the use of secrecy for tactical advantage or surprise, or clandestine operations.

Existing research on covert action shares a decisionmaking model in which states make tough choices about how best to pursue foreign policy goals. In general, states have one of three options available to them. First, they can choose to live with the status quo and do nothing. Second, they can undertake some form of overt action, including diplomacy, economic statecraft, and military force. Finally, they can turn to covert action. The appeal of the so-called ‘quiet option’ is that it promises to deliver foreign policy successes to leaders while simultaneously avoiding the sometimes-messy ramifications of publicity. Covert action may thus allow leaders to have their cake and eat it too. Understanding the conditions under which leaders will find overt action most dangerous and, consequently, plausible deniability most appealing, has been the focus of most covert action scholarship to date.

One argument in this vein is that covert action appeals to leaders interested in pursuing policies that may be unpopular with domestic audiences (Baum, 2004; Gibbs, 1995). When overt intervention would drag the country into a protracted war, unnecessarily risk the lives of its soldiers, or tarnish the country’s values, the incentives for secrecy are powerful. Similarly, secrecy may be useful for interventions that serve special interests rather than the population at large (Downes & Lilley, 2010). Other accounts focus on international dynamics such as escalation. In this vein, Carson (2016) argues that covert action helps leaders avoid unwanted escalation by reducing pressures on either side to respond.

Figure 1.

Theoretical choice model

Adversaries may even collude in the act of secrecy if it means keeping the lid on escalation relatively closed.

Understanding when covert action will be most appealing is essential to understanding why leaders sometimes go to great lengths to deny responsibility for their actions. Missing from existing accounts, however, is a sustained discussion of constraints. One potential constraint is the possibility that covert action may not succeed. A second is the likelihood that the sponsor’s identity will remain hidden or at least plausibly deniable. In this vein, we do not yet have a good handle on how leaders evaluate whether they can keep a mission secret and, if they cannot, what the implications might be for decisionmaking.

Figure 1 illustrates the three-way decisionmaking model referenced above. The left-most box captures the key drivers of covert action, the primary focus of the existing scholarship. The middle and right-most boxes emphasize the neglected constraints on covert action. In the discussion that follows, we assume that leaders are rational and forward-looking. We expect them to think carefully about the costs and risks of all three intervention options and to choose the one that optimizes their payoffs given the pros and cons of each. Our primary contribution, colored in red and in bold, is to investigate how the threat of exposure influences a leader’s decision to authorize covert missions. Although we do not theorize the likelihood of mission failure and its effect on intervention decisions directly, our empirical models include several factors that influence this probability.

Exposure and the decision for covert action

In our account, a leader’s primary rationale for choosing covert action is the promise of plausible deniability. We conceive of deniability as a continuous variable. At one extreme, the sponsor’s involvement is perfectly hidden from foreign and domestic audiences alike. At the other extreme, there is iron-clad evidence of the sponsor’s involvement. Between these two extremes is a large gray area. Anecdotal evidence reveals that threats to plausible deniability figure centrally in the decision to authorize covert missions. According to Mark Lowenthal (2015: 231), the former Assistant Director of Central Intelligence for Analysis and Production, ‘[p]olicymakers and intelligence officials examine at least two levels of risk before approving a covert action. The first is the risk of exposure.’¹

There are three pathways through which a sponsor’s involvement is exposed without its consent. The first pathway is internal leaks. Governments planning covert missions worry that leaks by dissenters in their ranks will expose secret missions (Hastedt, 2005). The second pathway comes from powerful third-party governments with global intelligence capabilities. Great powers use a range of intelligence assets to identify and understand the foreign involvement, both overt and covert, of key rivals.² The third pathway is the discovery of a covert operation by entities within the target state itself, including reporters, local civilians, and members of the target state’s military and intelligence apparatus. To the extent that these local actors can furnish credible evidence of secret foreign meddling, an intervener’s capacity to plausibly deny its involvement may be greatly diminished.

In the end, decisionmakers may care less about who exposed a covert operation and simply that it happened. As such, they must guard against all potential avenues to unwanted exposure. Fully understanding how the threat of exposure influences decisions for covert action would require exploring each of the three pathways identified above in detail. As an initial cut, we focus our attention on the third.³ There are several reasons for this. First, it is theoretically interesting. There are many substate actors that might expose missions, each with unique capabilities and incentives. Second, we find evidence in declassified documents that planners worry about local actors exposing them. Finally, data pertaining to ICTs and media awareness – factors we think are relevant to exposure – are available for many countries over many years, allowing us to test our argument across many cases.

How targets ruin plausible deniability

There are three local actors in the target state that can expose a mission and damage the sponsor’s plausible deniability. One is the target government. Most states are watchful for foreign meddling. Exposing covert missions can generate increases in domestic support akin to a rally around the flag effect, thereby shoring up a regime’s hold on power. Governments train their security forces to monitor foreigners and identify suspicious behavior. Relative to the population at large, however, government personnel are few in number and less likely to stumble across covert planners. Moreover, given the government’s incentive to cook up conspiracy theories, the amount of evidence they must generate to shake plausible deniability is high. According to Godson, ‘Try as they might to expose secret assistance from the United States [in postwar Europe], on the whole the KGB and Communist parties found it difficult to produce smoking guns. Party and front publications were vociferous in their denunciations of CIA aid, but this was written off as simple Communist propaganda’ (Godson, 1995: 44).

A second set of actors are the citizens of the target state. Unlike governments, citizens are broadly dispersed across the country and are the most likely to stumble across covert operators preparing for missions, training, or moving to and from the mission site. Further, citizens have weak incentives to lie about foreign meddling and therefore provide more credible testimony. Yet civilians are poorly trained to identifying suspicious activity as foreign meddling. At a bare minimum, they must recognize that a covert operation is underway, link that action to a sponsor, collect tangible evidence of the unauthorized action, and broadcast that evidence to the world. Even when they notice suspicious activity, they may be unlikely to collect evidence and more likely to attribute it to local criminal activity. Thus, although civilians are more numerous than government personnel, they lack sufficient training and awareness to recognize and report on covert operations around them.

Finally, enterprising reporters in the target state can expose covert operations. Not only are their incentives to misrepresent lower than the target government’s, but they are also much better trained than civilians to attribute sponsorship for a covert operation and to follow up with further investigation.⁴ Declassified documents suggest that covert planners worry about reporters exposing them. In a memo dated 31 August 1967 to President Johnson, Special Assistant Walt Rostow described the various risks associated with secretly sponsoring pilots in the Congo: ‘Like all covert operations, this one runs the risk of exposure […] It is conceivable that an enterprising reporter (of whom there are very few in the Congo) could blow the cover’ (Rustow, 1967: emphasis added). However, reporters are typically not present in large enough quantities to easily expose operations on their own. Even in countries with many reporters, they tend to concentrate in cities rather than the countryside and thus may not observe training and preparation for covert operations.

Each of these local actors faces limitations when it comes to exposing covert operations. But the limitations for each are different. The risk of exposure is amplified if these three actors can communicate effectively to combine their strengths. This is not an easy task. Recent work suggests that the capacity to transmit information locally and nationally varies dramatically (Larson & Lewis, 2017). We argue that dense ICT networks reduce the coordination barriers that these groups face, thereby increasing the likelihood that a covert operation would be exposed. There are two reasons that ICTs reduce the barriers for coordination that correspond with the I (information) and C (communication) in ICTs.

First, dense ICT networks allow local actors to communicate quickly and inexpensively across large distances. Therefore, ICTs help information spread more rapidly across countries in ways that would be impossible if suspicious activity were simply reported by word of mouth. Regarding covert action in particular, access to technologies that reduce the time and expense of the diffusion of information should increase the chance that evidence of foreign involvement will be transmitted and, as a result, plausible deniability destroyed.

Consider how ICT networks affect the capacity of reporters to expose covert missions. As noted, reporters have the skills to verify covert operations that most citizens lack but are often sparsely distributed across a given country. When reporters can communicate with local citizens and media outlets, they are more likely to be notified of suspicious activity in a timely fashion. Therefore, they are more likely to arrive on scene and verify it quickly. The speed with which reporters conduct investigations into covert operations is a critical factor in their capacity to expose one. Since covert operators work fast and cover their tracks, reporters must work faster to catch them before evidence disappears.

Communications technology should also enable target governments to communicate with, or eavesdrop on, reporters and citizens to identify suspicious covert activity and blow the lid off an operation. The ability to tap into local networks is an invaluable asset in a government’s intelligence collection capabilities (Daugherty, 2004: 10). The more access governments have to citizens and local reporters, the better off they will be when it comes to detecting the presence of foreign meddling within their borders.

Second, citizens with more access to political information via news outlets are more aware of foreign threats (Warren, 2014). Greater awareness of national and global events, which comes from consuming national and international media (Potter & Baum, 2014), increases citizens’ capacity to recognize suspicious activity and consider foreign intervention as the cause. Politically unaware populations may view suspicious activity as low-level domestic criminal activities and choose not to report it. Those that understand that foreign actors may attempt to subvert them are more likely to see something and say something.

The effect of access to media on the odds of exposure is amplified by state messaging and propaganda. As Warren (2014) shows using cross-national data of ICTs, citizens in states with widespread access to media are more easily influenced by government messages.⁵ Governments use media and propaganda to incite fear in their citizens that foreign actors are out to get them. Iran, for example, regularly reports that the United States meddles in their affairs. As a result, Iranians are more likely to be on the look-out for foreign activities.

How exposure affects decisionmaking

Recall that the main advantage of plausible deniability is that leaders can avoid triggering escalation with powerful adversaries or incurring domestic and normative costs associated with overt actions to achieve the same ends. Which set of concerns are most pressing likely varies from state to state and across time. States that subscribe to liberal democratic norms may find plausible deniability attractive to the extent that they face unusually large domestic or reputational costs for flagrant meddling abroad (O’Rourke, 2013). Conversely, interventions on the doorstep of a powerful adversary may be prime candidates for plausible deniability (Carson, 2016). During the Cold War, the ‘tension between perceived ideological threats (justifying intervention) and self-determination norms (undermining its result) shaped much of the intervention behavior of the superpowers and other intervening states’, including the frequent use of covert action (Finnemore, 2003: 128). In this vein, consider how decisionmakers described the risks of intervention against Salvador Allende of Chile in 1970. Henry Kissinger told President Nixon: ‘We are strongly on record in support of self-determination and respect for free election[s]; you are firmly on record for non-intervention in the internal affairs of this hemisphere and of accepting nations “as they are.” It would therefore be very costly for us to act in ways that appear to violate those principles’ (quoted in Downes & Lilley, 2010: 291).

So long as the sponsor’s identity remains secret, relevant costs can be avoided. But this advantage is squandered once the sponsor’s hand is shown. Therefore, as plausible deniability becomes more difficult, we argue that leaders look to other options. When the costs and risks of exposure are high relative to the objective being sought, decisionmakers may choose to refrain from intervention altogether. In other cases, when the policy objective is valuable enough but plausible deniability is elusive, leaders may choose to intervene overtly.

One might wonder why, if exposed covert operations carry similar costs and risks as overt operations, a state bent on intervention would not simply attempt covert action anyway and pray that the mission remains secret even if the odds of exposure are high. If there is a minuscule chance they will not get exposed, and the costs incurred after exposure are the same for visible interventions, isn’t covert action better?

We argue that going covert entails three additional disadvantages above and beyond overt action that can make the latter more appealing when the chance of exposure is high. First, there are concealment costs. It takes interveners time, effort, and additional planning to cover their tracks. Furthermore, well-trained US forces and advanced equipment are easily recognizable. Often it is simply infeasible to use US troops and equipment and retain plausible deniability, forcing mission planners to choose between higher risks of exposure and relying on less reliable local actors (O’Rourke, 2013; Treverton, 1987). If mission planners did not care about exposure, this would not be a concern.

A second reason why interveners might prefer overt to covert action when the chances of exposure are high is that acting secretly eliminates any opportunity to frame the issue and justify the importance of the mission to the public. Leaders who strongly believe that an unpopular mission is necessary can plead their case openly. Even if the mission is unpopular, leaders can typically alleviate some of the blow-back through these public appeals (Berinsky, 2007; Gelpi, Feaver & Reifler, 2009). This opportunity is lost if a decisionmaker selects covert action and the story is broken by a news outlet instead.

Finally, leaders may incur costs for appearing dishonest, which can happen when undertaking covert missions that circumvent public scrutiny. Some evidence suggests that dishonesty induces a psychological response in individuals, who may then act to punish their leaders (Greene et al., 2001). However, rational theory also shows that a leader’s reputation as an honest actor is valuable (Sartori, 2005; Tomz, 2007).

To be clear, we do not argue that all of these disadvantages are present all the time. Nor do we argue that overt action contains no unique downsides. We accept that in many cases leaders worry about reputational costs associated with making public commitments and backing down or pressure from domestic audiences to prevail in conflict once the body bags start coming home (Fearon, 1994; Gelpi, Feaver & Reifler, 2009). The Vietnam War, which put US credibility on the line and entailed mass casualties, is a good example of this. Nevertheless, we believe that the majority of overt operations, most of which involve far more limited commitments and force deployments than Vietnam, will entail roughly similar costs as an exposed covert operation. This is especially true for the kinds of interventions we focus on below: missions to subvert sovereign regimes.

The following hypothesis tethers together the two insights from above. First, we argued that dense ICT networks in the target state threaten plausible deniability. Second, and returning to our model in Figure 1, we argued that the risk of exposure makes covert action less attractive relative to the alternative policy choices, all else being equal. Whether leaders will pursue overt action or no intervention depends on many other factors that we control for in our analysis. Empirically, we assess the effect exposure has on the likelihood of both outcomes relative to covert action. In general, our expectation is:

H1: Leaders are less likely to use covert action as the density of ICTs in the target state increases.

Data

As noted, we focus our attention on the United States’ decision to intervene against a foreign government. We construct a country-year dataset (omitting the USA) for every country in the international system during the Cold War (1947–90). Each of the 4,516 observations represents an opportunity for the USA to intervene against a target in a particular year. We refer to them here as target-years. During the Cold War, the United States reached broadly into the affairs of states all over the world. For this reason, we consider all other states as potential targets of meddling by the USA in our main analysis. Summary statistics for the independent and control variables, and a table of all covert and overt interventions we code, are presented in the Online appendix.

Dependent variable: Intervention against incumbent regimes

We define foreign intervention as any attempt to subvert, undermine, or physically oust an incumbent regime. We include all cases where the military or intelligence services used force to overthrow a regime directly or through the support of locals. We also include any instances where the United States interfered significantly in the electoral processes of foreign countries by funding opposition parties during an election.

Our focus on interventions to subvert, undermine, or oust incumbent regimes fits well with our scope conditions. Our theory applies to cases meeting two criteria: (1) the use of overt force by the USA must be plausibly controversial with foreign or domestic stakeholders; and (2) the target must have incentives to expose the mission. Regime change operations involving force, meddling in foreign elections, and high-level militarized disputes fit these criteria well. We exclude other instruments of foreign policy, such as economic sanctions, limited interventions, and diplomatic or institutional lobbying, primarily because these options may not be sufficiently controversial. We also exclude cases where the United States intervened to support an incumbent. While these missions may be contentious at times, they often reinforce sovereignty norms and demonstrate the United States’ commitment to allies and partners. Coding in this way is a tough test of our theory because it limits the number of available missions. Even though the USA was more active than most other states during the Cold War, interventions that fit our scope conditions are still fairly rare.

Covert interventions

We identify secret US-led missions to subvert incumbent regimes using two recently released datasets. To capture acts of covert foreign-imposed regime change, we use data from Downes & O’Rourke (2016). This dataset has several advantages over other sources for our purposes. First, Downes & O’Rourke (2016) focus exclusively on interventions against incumbents. Second, and unlike the widely used Berger et al. (2013) data, it includes both successful and failed missions and verifies cases using declassified documents rather than secondary sources. Third, it restricts the sample to only years where United States forces were actively seeking to subvert a foreign regime.

To capture interventions in foreign elections, we use data from Levin (2016). Levin codes partisan electoral interventions undertaken by either the United States or the Soviet Union in which the goal was to overtly or covertly support an incumbent party or a challenger between 1946 and 2000. In order to ensure comparability with the Downes & O’Rourke (2016) data, we only included cases from Levin’s dataset in which the USA intervened covertly on behalf of a challenger during the Cold War. Taken together, we are left with 229 unique covert action observations, which is approximately 6% of the total dataset.

Overt interventions

For overt military interventions, we code episodes in which the United States deployed the military to subvert, undermine, or overthrow a target in a given year. To generate as complete a set as possible, we combined four datasets of overt intervention. In each case, we subset the data by direct US interventions or conflicts in which the United States materially supported rebel groups and other third parties against a target government.

We first include militarized interstate disputes using Palmer et al. (2015). We code an overt intervention if a country engaged in a high-level MID (a hostility level of 3 or higher) against the USA. Second, we include observations from the UCDP/PRIO dataset where the USA directly fought against a state during a civil war or supported insurgents in their fight against the state in a given year (Melander, Pettersson & Themnér, 2016). Third, we include all US military interventions from the IMI dataset in which the United States was fighting against the incumbent government directly or with the rebels (Kisangani & Pickering, 2008; Pearson & Baumann, 1993). Finally, we again use data from Levin (2016) to

Table I.

Correlation between different measures of ICT and other development measures

	GDP	Industrial production	School enrollment	Telephone	Radio	TV	Newspaper
Industrial production	0.91
School enrollment	0.57	0.37
Telephone	0.27	0.30	–0.07
Radio	0.35	0.37	–0.04	0.83
Television	0.39	0.41	–0.01	0.83	0.83
Newspaper	0.35	0.37	–0.01	0.74	0.73	0.67
MCTI	0.39	0.41	0.02	0.88	0.96	0.91	0.86

capture intervention in foreign elections, this time restricting the data to cases where the USA openly interfered to support a challenger.

We count as intervention the first year in which the USA deployed forces or materially supported an opposition party against an incumbent as well as all subsequent years until the regime falls or the USA gives up. If a covert and overt intervention occurred in the same country in the same year, we code it as an overt intervention. This leaves us with 207 unique overt intervention observations, approximately 5% of the total dataset.

Independent variables: Communication capacity and media

Our assumed mechanism is a leader’s assessment of the likelihood that their sponsorship of a covert mission would be exposed. Above, we invoked the density of ICTs in a target state as one of the key indicators mission planners use to gauge the feasibility of plausible deniability. We identified the ease of communication between the government, citizens, and reporters – and the general awareness of the population – as key mechanisms that raise the probability that evidence will be recorded to expose a covert action. Drawing from the ICT literature, we use data from Warren (2014) on the density of media and communications technology. Warren begins with raw cross-national time-series data on the length of telephone cables laid, newspapers in circulation, and the presence of televisions and radios from the Banks Dataset and the World Bank. Warren then uses information from both sets to generate per capita measures for each ICT that take into account missing data and biased counting in both raw datasets.

Using these data, Warren (2014) generates a country-year media and communications technology per capita index (MCTI). The measure is constructed as follows:

M C T I_{i t} = \frac{T e l e v i s i o n s_{i t} + R a d i o s_{i t} + N e w s p a p e r s_{i t}}{P o p u l a t i o n_{i t}}

for country i in year t. This measure has become the industry standard for cross-national studies of ICTs in the substate political violence. However, it is also common for researchers to use the component parts separately when their theory is more specific. Potter & Baum (2014), for example, use only radio and television per capita measures because their theory describes media awareness. Therefore, the telephone line measure would be inappropriate.

Below, we first analyze the MCTI measure and then explore each of the four technology and media distribution elements as alternative specifications. We do this for two reasons. First, although our theory makes the same claim about the impact of media and communications technology, the causal mechanisms are distinct. The telephone lines and radio stations measure the density of communications infrastructure, which facilitates rapid communication between citizens, reporters, and governments in the target state. The newspaper and television variables capture media awareness, which increases the chance that citizens are alert to foreign issues and unusual political behavior. Thus, we expect that increases in each of these measures should be significant, albeit for different reasons.

Second, although the measures are correlated, there is still considerable variance among them as shown in the correlation matrix in Table I. These measures not only imperfectly covary, but they are also imperfectly correlated with other measures of infrastructure development and GDP. As such, analyzing them all presents a tough test for our theory. If we are right, high scores on any of these measures should give covert mission planners pause. A significant finding across all of them lends strong support to our theory.

Control variables

Our proxies for the probability of exposure vary with the target government’s capacity, human development, war-fighting potential, and levels of infrastructure development. To isolate the effect of communications technology and media awareness, we included a series of controls intended to measure these concepts but not a target’s capacity to communicate or its media awareness. We include the target’s revenue over its expenditure, primary school enrollment per capita (Banks & Wilson, 2014), and GDP (Gleditsch, 2002). Similarly, we include a measure of their military capabilities using the CINC score index from the Correlates of War Project (Singer, Bremer & Stuckey, 1972). These measures capture much of the covariance between our communications variables and general factors having to do with development. We also include a control for the land area of the target state and mountainous terrain (Fearon & Laitin, 2003). The residual effect of our measures isolates the effect of the capacity to communicate.

Next, we include control variables that vary with the theoretical concept – the choice of intervention – as well as the USA’s potential interest in the target. We include logged trade with the United States to control for a target’s level of dependence on the United States, distance between the target’s capital and Washington, DC (Gleditsch, 2002), and the existence of an alliance between the target and the United States (Gibler, 2009).

We also included controls that may make military intervention against a target government unpopular for US audiences. For example, we expect that US presidents will be reluctant to intervene against open and democratic societies or their allies. To account for this, we include the Polity IV measure of the target’s regime type (Marshall & Jaggers, 2002) as well as a measure of press freedom in the target state (Van Bell, 1997).

In addition to possible domestic constraints on intervention, we account for several factors related to the target’s foreign policy, separate from the USA, that may generate escalatory risks in the case of an attempted intervention. We include dummy variables that identify if the target is involved in a third-party military dispute or threatened dispute (Palmer et al., 2015). Because we focus on the Cold War, we also control for target–Soviet trade (logged), distance between the target’s capital and Moscow (logged) (Gleditsch, 2002), and the existence of an alliance between the target and the Soviet Union (Gibler, 2009).⁶ We do not include a population control since the independent variables are ICTs per capita.

Summary statistics

As a first cut, we plot fitted parametric curves for the Media and Communications Technology Index across time, broken out by types of intervention in Figure 2(a). The green dashed curve shows the average MCTI for countries that were not targets of intervention across time. Consistent with global investments in media and communications technology, the average MCTI score increases across time. The solid red and dotted blue curves plot the average MCTI scores over time for targets of covert and overt interventions, respectively. Notably, the average score for targets of covert intervention remains roughly constant across time and is lower than for targets of overt interventions as well as the average state that did not experience intervention of any kind. The predictive line for overt interventions is on average consistent with rates of ICT growth across the sample as a whole. Figure 2(b) swaps out the aggregate MCTI measure for telephone cabling per capita. The interpretation is similar.

To illustrate that ICTs are a unique sort of technology with respect to their relationship to types of intervention and not simply measuring industrialization or human development, we construct the same plot but replace the ICT proxies with rates of industrial production (c) and primary school enrollment (d). In these comparisons, the targets of covert action are indistinguishable from the yearly averages of states that did not experience intervention.

Although we cannot draw strong inferences from these plots, the general trend in the data suggests that the relationship between ICTs and instances of covert action differs from both overt action and other types of technologies. The data suggest that access to communications and media technologies across all states increases over time, yet the targets of covert interventions consistently have low access. This appears to be unique to ICTs and is not present in broader measures of development or industrial production.⁷

Research design

We begin our analysis with a statistical model that reflects the basic causal model depicted in Figure 1. In our theoretical setup, the president chooses between no intervention, covert intervention, and overt intervention.

Figure 2.

Comparing targets of intervention outcomes to technological trends across time

As such, we model a multinomial logistic regression with three outcome categories that reflect these three options. We use placebo tests to rule out alternative mechanisms and measurement error. Additionally, it is possible that some states are more likely targets of intervention than others. To account for this, we also estimate a two-stage model where the first stage models selection into intervention, and the second the type of intervention (covert or overt) that takes place.⁸

Multinomial logit: A three-way decision model

We first estimate a multinomial logistic (MNL) regression where the dependent variable is the type of intervention, if any, in a target-year. The dependent variable is coded as a 0 when no intervention took place, 1 in years in which a covert intervention transpired, and 2 in years with overt intervention. MNL regression with three categories estimates three two-way comparisons. We make predictions about two of these: covert intervention versus no intervention and overt intervention, respectively. We omit the third comparison – no intervention versus overt intervention – from our main analyses; we return to this comparison in the placebo tests below. The models include all the controls listed above. We include time polynomials (Carter & Signorino, 2010) to account for factors that may affect US decisions to intervene globally at any specific point in time. These include variation in US preferences for intervention, the chance of leaks, and US–Soviet tensions.⁹

Table II.

Effect of MCTI on the intervention decision

Variables	(1)	(2)
Variables	No intervention relative to covert intervention	Overt intervention relative to covert intervention
MCTI	0.0267**	0.0267**
	(0.00527)	(0.00742)
Third-party dispute	–0.721**	1.899**
	(0.170)	(0.338)
Area (log)	1.104**	1.153*
	(0.290)	(0.455)
CINC (log)	–55.65**	–3.334
	(13.43)	(18.84)
Democracy	–0.0378*	–0.0694*
	(0.0162)	(0.0308)
Distance to Washington, DC (log)	0.0535	–1.094**
	(0.190)	(0.338)
Distance to Moscow (log)	–0.762**	–1.175**
	(0.177)	(0.308)
Trade with USA (log)	0.212**	–0.197*
	(0.0624)	(0.0890)
Trade with USSR (log)	–0.215**	–0.0904
	(0.0607)	(0.0907)
US ally	–0.685**	0.260
	(0.265)	(0.452)
USSR ally	–0.979**	–0.488
	(0.258)	(0.418)
GDP (log)	–0.218*	0.0973
	(0.103)	(0.171)
Industrial production	–7.79e–07	–1.98e–05**
	(1.45e–06)	(6.17e–06)
Primary school enrollment	5.87e–05**	4.66e–05*
	(1.83e–05)	(2.32e–05)
Press freedom	–0.111	0.321*
	(0.0858)	(0.163)
Mountainous terrain	–0.00394	–0.0166
	(0.00412)	(0.00854)
Constant	11.10**	17.10**
	(2.600)	(4.805)
Time polynomials	√	√
Pseudo R ²	0.2556	0.2556
Observations	4,349	4,349

Standard errors in parentheses. **p < 0.01, *p < 0.05.

We estimated five models in total, one for each of the ICT measures. Each model is identical in all respects except for the choice of ICT measure. Table II presents the complete results with controls for Warren’s main MCTI. We plot the results of the ICT coefficients from all five models in Figure 3. Each row represents a different way of measuring media density taken from all five models. The same control variables were used in each specification.

As already mentioned, we make specific predictions about two of the three two-way comparisons produced by the MNL models: the likelihood of no intervention relative to covert action (Figure 3(a)) and overt action relative to covert action (Figure 3(b)). The point estimates are the coefficients for ICTs of the separate models with 95% confidence intervals. Positive values indicate increased ICTs make covert action less likely.

No matter how we measure the independent variable, increases in ICTs decrease the likelihood of covert action relative to the two other options. This is true regardless of whether we focus on individual proxies for communications infrastructure or media awareness. The effect is statistically significant for ten coefficients across five different measurement strategies for ICTs, providing powerful support for our argument. As the political awareness and capacity of various actors to communicate increases, the likelihood that US presidents will pursue covert regime change relative to no intervention or overt action decreases.

Placebo test

A natural concern is that the ICT measures used above vary with factors other than the probability of getting caught during a covert operation. If our controls do not account for the omitted variation, the results will be biased owing to missing variable bias. We construct a placebo test that rules out other mechanisms that may explain the result. A placebo test is a secondary test that identifies differences between our theoretical predictions and competing mechanisms. We identify features of our model that we would expect to see if mechanisms other than the ones we propose are present. If we can show that those findings are absent, we can increase our confidence that our model is properly specified.

Recall that MNL with three outcomes estimates three dependent binary logit models, generating separate coefficients for each two-way comparison. Researchers typically omit one of these comparisons. In our theory, we make no prediction for the effect of ICTs on the decision between overt intervention and no intervention. As a result, we have not discussed this comparison. However, we can use our null expectation to ease concerns about missing variable bias on the comparisons we care about. After we control for confounding factors, we expect our measures of ICTs to have no significant effect on the decision between overt intervention and no intervention. This would not be the case for alternative mechanisms that might drive the observed relationship between ICTs and covert intervention but are unrelated to the dynamics of exposure described earlier.

Figure 3.

The effect of different ICTs on the decision to intervene (MNL estimates)

To illustrate the intuition, consider that wealthy and powerful states are more important and might be subject to more interventions. If our control variables failed, media density may actually be capturing the target’s overall wealth, not political awareness or the capacity to communicate. If true, this confounding mechanism might explain why we see a relationship between ICTs and intervention. However, if this mechanism was indeed at work, it would also imply that an increase in ICTs positively predicted overt intervention against no intervention. The reason is straightforward: while we expect ICTs used as a proxy for the chances of getting caught conducting covert operations to have no bearing on the likelihood of no intervention relative to overt intervention, if ICTs are capturing a target’s wealth and power, they might have an effect here. We can therefore analyze the MNL results for the comparison between overt intervention and no intervention. If there is no systematic effect across the different ICT measures, we can rule out wealth as a factor that confounds our results.

Table III identifies the predictions for a series of competing causal mechanisms that might confound our result in a similar way as described above. If any one of these variables – GDP, military capability, state capacity, democracy, or an ongoing war – influenced the level of communications and media infrastructure in a target and is not adequately captured by our control variables, it may bias the results for the comparisons we care about, namely the two-way comparisons that include covert action. However, it would also bias the results in the comparison we do not care about (overt vs. no intervention).¹⁰ Figure 3(c) presents the coefficients estimated for overt versus no intervention in our models. If our measure is biased, we should see a constant effect of ICTs in this comparison. We do not. The effect is insignificant across each ICT specification. We contrast that with ten of ten positive and significant coefficients that support our hypotheses in the first two columns. In short, it would be highly unusual to see these results if our results were confounded.

Selection into intervention

Depending on the political preferences of target states, domestic-political issues, and various geopolitical dynamics, the USA may be more or less likely to select a given state as a target of intervention. If true, the results from our MNL models, which treat all three intervention options as unordered categorical variables, would be biased. To account for this, we use a Heckman selection model with a probit correction in both

Table III.

Comparing endogeneity with the theorized effect for media density

Endogeneity concerns for media density	Direction of bias for media density
Endogeneity concerns for media density	No intervention vs. covert	Overt vs. covert	Overt vs. No intervention
GDP → More important targets	–	(.)	–
Advanced infrastructure → Resistant to intervention	+	(.)	–
Military capability → Difficult to defeat	+	(.)	–
Democracy → Normative restrictions	+	–	–
War → Damaged infrastructure + increased interest	–	(.)	(.)
Theorized effect	+	+	(.)

stages to approximate a two-stage decision model. In the first stage, a president decides whether to intervene or not. In the second, the president decides between covert or overt intervention.

The first stage of the model requires a probit regression of form:

I n t e r v e n t i o n_{i t} = γ X_{i t} + γ W_{i t} + ∊_{1}

The outcome variable equals 1 if the USA engaged the target (i) in any sort of intervention – covert or overt – in a given year (t). The model includes two sets of predictors X_it, W_it. The former (X) predict both the decision to intervene in the first place and the type of intervention (covert or overt). These variables include the independent variable of interest – ICT technology – and will appear again in the second stage.

The latter (W) are chosen carefully to include variables that predict selection into intervention but not the difference between covert and overt action. We argue three variables fit this criterion because they (1) increase the United States’ interest in the target but (2) make both covert and overt operations equally more difficult and (3) do not increase the normative costs from engaging in overt intervention in the first place or getting caught afterwards.

Following the international relations literature, we first included physical distance as a proxy measure for the United States’ interest in the politics of a target state. We expect the USA to be more interested in states that are geographically proximate and are therefore more likely to target them for intervention. However, distance makes all types of military missions more difficult whether they are secret or not. Distance also has no independent effect on the legitimacy of intervention. As a result, the pros and cons of covert and overt intervention should not vary too much with distance. We believe that similar arguments apply for the GDP of target states and their military capabilities (measured in CINC scores).

We then compute coefficients for the probability of selection γ, which we use to estimate the predicted probabilities of selection into intervention for each observation in the set λ*. We then estimate the second stage probit equation that factors in these probabilities:

O v e r t A c t i o n_{i t} = β X_{i t} + λ^{*} + ∊_{2}

where the outcome is a binary variable equal to 1 if the United States engaged against the target in an overt intervention in a given year and 0 if it was covert. Years with no intervention are omitted (and captured in the first-stage model).

The results from both stages are presented in Table IV. The main result is the coefficient for the MCTI score in the second stage. Consistent with the results from the MNL models presented above as well as our theoretical expectation, countries with greater access to ICT technology are more likely to be the subject of overt interventions than covert interventions even after we account for selection into an intervention. Indeed, the results are actually stronger and more significant in this model than in our core MNL.

Notably, MCTI positively predicts intervention in the first stage (selection into intervention). This implies that the USA does not want to intervene, either covertly or overtly, against more technologically advanced states. This is consistent with broader research that finds interventions are less likely to succeed against states with advanced technologies. This intuitive result increases our confidence in the MNL models because it shows that the selection effect works against our finding by reducing the prospect of intervention all around.

Conclusion

The threat of exposure is an important constraint leaders face when authorizing covert operations. We focused on the pathway that varies with properties within the target of intervention, namely the chance that local actors will witness a covert action and expose it to the world. The presence of dense ICT networks aids this process. Our results were robust to many empirical tests.

Table IV.

Effect of MDI on the intervention decision, Heckman selection

Variables	(1)	(2)
Variables	First stage	Second stage
MCTI	0.0155**	–0.00528**
	(0.00515)	(0.00176)
Third-party dispute	1.051*	0.642**
	(0.416)	(0.0687)
Distance to Moscow (log)	–0.0325	0.161*
	(0.178)	(0.0656)
Trade with USA (log)	–0.00411	–0.164**
	(0.0718)	(0.0230)
Trade with USSR (log)	0.0791	0.0853**
	(0.0813)	(0.0216)
USSR ally	–0.604*	0.378**
	(0.277)	(0.107)
US ally	–0.0680	0.366**
	(0.332)	(0.105)
Press freedom	0.277*	0.0836*
	(0.131)	(0.0363)
Mountainous terrain	0.00797	0.00112
	(0.00513)	(0.00172)
Democracy	–0.0703**	0.0106
	(0.0216)	(0.00666)
CINC (log)		30.16**
		(6.024)
Distance to Washington, DC (log)		–0.290** (0.0747)
GDP (log)		0.0882*
		(0.0404)
Industrial production		–2.22e–06*
		(1.07e–06)
Primary school enrollment		–2.26e–05**
		(6.08e–06)
Constant	–1.290	–1.423
	(2.413)	(0.993)
Observations	4,349	4,349

Standard errors in parentheses. **p < 0.01, *p < 0.05. Wald χ² = 43.24.

Thinking about constraints on covert intervention has important implications for the future study of hidden statecraft. Indeed, states often view deception as beneficial. But without appreciating relevant risks, existing research cannot explain why covert action does not happen all the time. Future research might consider in more detail how the costs and risks interact and what the consequences are for various secret missions. For instance, when and why constituents care about particular covert actions, and how they punish interveners who get caught, is an important question.

Finally, our theory implies that Cold War-style covert actions may be a relic of a bygone era. The proliferation of cell phones with cameras and high-speed Internet to the far corners of the world means that there is always a high risk that evidence will be recorded. As such, any operation that is designed to remain secret runs an extraordinarily high risk of exposure. Of course, technology has not been entirely bad for proponents of covert action. The advent of unmanned aerial vehicles and cyber warfare in particular has enabled states to conduct covert operations across the globe from increasingly remote locations; attribution in these circumstances is possible, but difficult. Understanding how leaders grapple with these trade-offs is another important area for future research.

Footnotes

Replication data

The dataset, codebook, and do-files for the empirical analysis in this article, along with the Online appendix, can be found at . All statistical analyses were conducted using Stata 15 IC. Figures were generated using R version 3.2.2.

Acknowledgments

The authors are listed alphabetically and contributed equally. We are grateful to Brandon Bartels, Meredith Blank, Austin Carson, Alexander Downes, Anita Gohdes, Eric Grynaviski, Daniel Krcmaric, Philip Potter, two anonymous reviewers, JPR Associate Editor Nils Weidmann, and participants at the Lansing B. Lee, Jr Seminar in Global Politics at the University of Virginia for helpful feedback that improved the article.

Notes

References

Bailard

Catie S

(2015) Ethnic conflict goes mobile: Mobile technology’s effect on the opportunities and motivations for violent collective action. Journal of Peace Research 52(3): 323–337.

Banks

Arthur S

Wilson

Kenneth A

(2014) Cross-national time-series data archive (www.cntsdata.com).

Baum

Matthew A

(2004) Going private: Public opinion, presidential rhetoric, and the domestic politics of audience costs in US foreign policy crises. Journal of Conflict Resolution 48(5): 603–631.

Berger

Daniel

Easterly

William

Nunn

Nathan

Satyanath

Shanker

(2013) Commercial imperialism? Political influence and trade during the Cold War. American Economic Review 103(2): 863–896.

Berinsky

Adam J

(2007) Assuming the costs of war: Events, elites, and American public support for military conflict. Journal of Politics 69(4): 975–997.

Brown

Jonathan N

(2013) The sound of silence: Power, secrecy, and international audiences in US military basing negotiations. Conflict Management and Peace Science 31(4): 406–431.

Brown

Jonathan N

Marcum

Anthony S

(2011) Avoiding audience costs: Domestic political accountability and concessions in crisis diplomacy. Security Studies 20(2): 141–170.

Carson

Austin

(2016) Facing off and saving face: Covert intervention and escalation management in the Korean War. International Organization 70(1): 103–131.

Carter

David B

Signorino

Curtis S

(2010) Back to the future: Modeling time dependence in binary data. Political Analysis 18(3): 271–292.

10.

Crabtree

Charles

Darmofal

David

Kern

Holger L

(2015) A spatial analysis of the impact of West German television on protest mobilization during the East German revolution. Journal of Peace Research 52(3): 269–284.

11.

Daugherty

William J

(2004) Executive Secrets: Covert Action and the Presidency. Lexington, KY: University Press of Kentucky.

12.

Downes

Alexander B

Lilley

Mary L

(2010) Overt peace, covert war? Covert intervention and the democratic peace. Security Studies 19(2): 266–306.

13.

Downes

Alexander B

O’Rourke

Lindsey A

(2016) You can’t always get what you want: Why foreign-imposed regime change seldom improves state relations. International Security 41(2): 43–89.

14.

Fearon

James D

(1994) Domestic political audiences and the escalation of international disputes. American Political Science Review 88(3): 577–592.

15.

Fearon

James D

Laitin

David D

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

16.

Finnemore

Martha

(2003) The Purpose of Intervention: Changing Beliefs About the Use of Force. Ithaca, NY: Cornell University Press.

17.

Gelpi

Christopher

Feaver

Peter D

Reifler

Jason

(2009) Paying the Human Costs of War: American Public Opinion and Casualties in Military Conflicts. Princeton, NJ: Princeton University Press.

18.

Gibbs

David N

(1995) Secrecy and international relations. Journal of Peace Research 32(2): 213–228.

19.

Gibler

Douglas M

(2009) International Military Alliances, 1648–2008, v4.1. Washington, DC: CQ.

20.

Gleditsch

Kristian Skrede

(2002) Expanded trade and GDP data. Journal of Conflict Resolution 46(5): 712–724.

21.

Godson

Roy

(1995) Dirty Tricks or Trump Cards: US Covert Action and Counterintelligence. Washington, DC: Brassey’s.

22.

Greene

Joshua D

Sommerville

Nystrom

Leigh E

Darley

John M

Cohen

Jonathan D

(2001) An fMRI investigation of emotional engagement in moral judgment. Science 293(5537): 2105–2108.

23.

Hastedt

Glenn

(2005) Public intelligence: Leaks as policy instruments – the case of the Iraq war. Intelligence and National Security 20(3): 419–439.

24.

Kisangani

Emizet F

Pickering

Jeffrey

(2008) International Military Intervention, 1989–2005. Inter-University Consortium for Political and Social Research, Data Collection No. 21282, University of Michigan, Ann Arbor.

25.

Krcmaric

Daniel

(2017) Information, secrecy, and civilian targeting. Working paper, Northwestern University.

26.

Larson

Jennifer M

Lewis

Janet I

(2017) Ethnic networks. American Journal of Political Science 61(2): 350–364.

27.

Levin

Dov H

(2016) When the great power gets a vote: The effects of great power electoral interventions on election results. International Studies Quarterly 60(2): 189–202.

28.

Lowenthal

Mark M

(2015) Intelligence: From Secrets to Policy, 6th edition. Thousand Oaks, CA: CQ.

29.

Marshall

Monty G

Jaggers

Keith

(2002) Polity IV project: Political regime characteristics and transitions, 1800–2002 (http://www.systemicpeace.org/inscrdata.html).

30.

Melander

Erik

Pettersson

Therése

Themnér

Lotta

(2016) Organized violence, 1989–2015. Journal of Peace Research 53(5): 727–742.

31.

O’Rourke

Lindsey A

(2013) Secrecy and security: US-orchestrated regime change during the Cold War. Unpublished PhD dissertation, University of Chicago.

32.

Palmer

Glenn

D’Orazio

Vito

Kenwick

Michael

Lane

Matthew

(2015) The MID4 dataset, 2002–2010: Procedures, coding rules and description. Conflict Management and Peace Science 32(2): 222–242.

33.

Pearson

Frederic S

Baumann

Robert A

(1993) International Military Intervention Data, 1946–1988. Inter-University Consortium for Political and Social Research, Data Collection No. 6035, University of Michigan, Ann Arbor.

34.

Potter

Philip B

Baum

Matthew A

(2014) Looking for audience costs in all the wrong places: Electoral institutions, media access, and democratic constraint. Journal of Politics 76(1): 167–181.

35.

Poznansky

Michael

(2015) Stasis or decay? Reconciling covert war and the democratic peace. International Studies Quarterly 59(4): 815–826.

36.

Rustow

Walt W

(1967) Memorandum from the President’s Special Assistant (Rostow) to President Johnson. Foreign Relations of the United States, 1964–1968 Volume XXIII, Congo, 1960–1968, Document 544 (https://history.state.gov/historicaldocuments/frus1964-68v23/d544, accessed 13 June 2017).

37.

Sartori

Anne E

(2005) Deterrence by Diplomacy. Princeton, NJ: Princeton University Press.

38.

Shapiro

Jacob N

Weidmann

Nils B

(2015) Is the phone mightier than the sword? Cellphones and insurgent violence in Iraq. International Organization 69(2): 247–274.

39.

Singer

Bremer

Stuart

Stuckey

John

(1972) Capability distribution, uncertainty, and major power war, 1820–1965. In: Russett

Bruce

(ed.) Peace, War, and Numbers. Beverly Hills, CA: Sage, 19–48.

40.

Szoldra

Paul

(2014) Without realizing it, Russian soldiers are proving Vladimir Putin is lying about Eastern Ukraine. Business Insider 31 July (http://www.businessinsider.com/russian-soldiers-social-ukraine-2014-7).

41.

Tomz

Michael

(2007) Reputation and International Cooperation: Sovereign Debt Across Three Centuries. Princeton, NJ: Princeton University Press.

42.

Treverton

Gregory F

(1987) Covert Action: The Limits of Intervention in the Postwar World. New York: Basic Books.

43.

Van Bell

Douglas A

(1997) Press freedom and the democratic peace. Journal of Peace Research 34(4): 405–414.

44.

Warren

(2014) Not by the sword alone: Soft power, mass media, and the production of state sovereignty. International Organization 68(1): 111–141.

45.

Warren

(2015) Explosive connections? Mass media, social media, and the geography of collective violence in African states. Journal of Peace Research 52(3): 297–311.

46.

Yanagizawa-Drott

David

(2014) Propaganda and conflict: Evidence from the Rwandan genocide. Quarterly Journal of Economics 129(4): 1947–1994.

Supplementary Material

Please find the following supplemental material available below.

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

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

0.00 MB