Denying the Obvious: Why Do Nominally Covert Actions Avoid Escalation?

Abstract

In 2014, Russia denied that its military was assisting separatists in eastern Ukraine, despite overwhelming evidence. Why do countries bother to deny hostile actions like this even when they are obvious? Scholars have argued that making hostile actions covert can reduce pressure on the target state to escalate. Yet it is not clear whether this claim applies when evidence of responsibility for the action is publicly available. We use three survey experiments to test whether denying responsibility for an action in the presence of contradictory evidence truly dampens demand for escalation among the public in the target state. We also test three causal mechanisms that might explain this: a rationalist reputation mechanism, a psychological mechanism, and an uncertainty mechanism. We do find a de-escalatory effect of noncredible denials. The effect is mediated through all three proposed causal mechanisms, but uncertainty and reputational concern have the most consistent effect.

In 2014, Russians in military uniforms without insignia began providing combat support to separatists in eastern Ukraine.¹ External observers quickly identified them as Russian soldiers, but the Kremlin denied this for over a year.² Four years later, Russian defector Sergei Skripal was assassinated in the United Kingdom. Although the nerve agent used, Novichok, was distinctly Russian, Russia again denied involvement. Although Russia's pattern of denying the obvious may be particularly egregious, it is not unique. China denied that its balloon drifting across the United States was intended for spying. Israel has a long-standing policy of neither confirming nor denying assassinations. Pakistan has denied the involvement of its armed forces in conflicts over Kashmir. The United States likewise refused for years to acknowledge its drone campaign in Pakistan, despite widespread media reporting.

The literature suggests a possible motive for non-acknowledgment: limiting escalation. Keeping a hostile action covert is believed to lessen the likelihood of escalation by allowing the target to save face and creating ambiguity about responsibility.³ Yet it is not clear whether this argument applies in cases like the ones just mentioned, where the responsibility for the action is obvious. Can a denial still be face saving when it is an obvious lie? And can uncertainty about the perpetrator's identity exist in the presence of clear evidence? Raising further doubt, a recent survey experiment by Pischedda and Cheon finds no significant difference in target escalation preferences when an easily attributable action is denied versus when it is admitted.⁴

In this research note, we seek to more definitively assess whether denying obvious actions—which we call nominal covertness—can limit escalation, and if so, why. We use new survey experiments to test whether nominal covertness truly reduces preferences for escalation among the public in the targeted country. We also test three causal mechanisms that might explain this, including a rationalist face-saving mechanism, a psychological insult mechanism, and an uncertainty mechanism. Understanding the public reaction to nominally covert actions is important because information about these actions is publicly available, and public opinion can influence the responses of policymakers.⁵

We field our three survey experiments in samples of the US public. Subjects in each experiment read a fictional scenario in which a country—either Iran or Qatar—attacked US commercial ships in the Persian Gulf. All respondents were presented with evidence of the attacker's responsibility, but the control group was told that the country claimed responsibility, while the treatment group was told the country denied it. We find that a denial reduces respondents’ desire for an escalatory response. This effect is mediated through all three proposed causal mechanisms, but uncertainty and reputational concern have the most consistent effect. We find little evidence that the treatment effect differs based on dispositional attributes, indicating our results are likely to be relevant beyond the United States.⁶

While covert action has received increased scholarly attention in recent years, we are among the first scholars to focus on actions that are covert in name only. We are also among the first to explore public reactions to covert actions. Apart from Pischedda and Cheon,⁷ we are the first to directly compare target reactions to nominally covert versus overt attacks.⁸ Unlike Pischedda and Cheon, we find a significant effect of denials. Moreover, we go beyond their study by being the first to compare different causal mechanisms for the de-escalatory effect of nominal covertness. This contributes to theory by helping us better understand the purpose of nominal covertness.

Our findings also have important policy implications. For governments that are targeted by nominally covert actions, the willingness of some individuals to give credence to an adversary's denial may be an obstacle to rallying public support for a response. Our results show that nominal covertness is very effective at creating uncertainty, and it is likely to become even more effective over time as political polarization and distrust of traditional information sources increases. This suggests that polarization and distrust are not merely domestic political problems but also harm national security, by making the United States and other countries increasingly vulnerable to nominally covert attacks.

The Veneer of Covertness

The last decade has seen an upsurge in research on covert actions. Scholars have noted the importance of covert actions in a world where overt military interventions are increasingly costly,⁹ and declassified documents have allowed covert actions to be analyzed quantitatively.¹⁰ Despite this research activity, the very concept of covertness remains ambiguous. Scholars have pointed out that covert actions can have different levels of deniability.¹¹ The US government considers an action covert if the role of the US government is not “apparent or acknowledged publicly.”¹² Yet some scholars have defined covertness more broadly. Carson and Yarhi-Milo consider an action covert if it is unacknowledged and “most observing audiences do not know about or cannot attribute” it.¹³ O'Rourke offers an even broader definition, requiring only that the intervening state “does not acknowledge its role publicly.”¹⁴

The broadest definitions of covert actions can incorporate actions that are widely known, as long as the perpetrator does not acknowledge responsibility. We refer to actions that remain unacknowledged despite obvious evidence of responsibility as nominally covert.¹⁵ Some of these actions begin with the intention of secrecy but are unexpectedly revealed. In other cases, it does not appear that the actions were ever truly intended to be secret. But given that lying can be inefficient and cause public disapproval,¹⁶ why don't states simply acknowledge their activities once they have become obvious?

The literature suggests a desire to avoid escalation as a potential explanation for covertness.¹⁷ Carson argues that covertness saves face and reduces demands for revenge from the public.¹⁸ O'Rourke similarly claims that covertness reduces the perception of a reputational challenge.¹⁹ Others argue that undertaking operations covertly signals reassurance that the perpetrator wishes to avoid escalation.²⁰ Finally, the uncertainty created by covert actions can raise doubt about whom to punish.²¹ Based on this prior work about the de-escalatory benefits of covertness, we derive our first hypothesis:²²

H1 Public support for an escalatory response will be lower when a hostile action is denied than when it is done overtly.

Yet most of the scholarship used to derive this hypothesis focuses on actions that are fully covert, without clear public evidence of responsibility. Applying these explanations to actions that are covert in name only raises new questions. For example, why does the perpetrator's denial reduce the perception of a challenge and placate the public even when the public has evidence that the denial is a lie? And how can uncertainty about the perpetrator's identity persist despite evidence? Pischedda and Cheon's study raises further doubt about whether H1 applies to nominally covert actions. In a survey experiment with a hypothetical scenario, they find that whether Russia claims or denies responsibility for an explosion at a NATO base does not significantly affect US public support for an airstrike or war against Russia.²³

How Nominal Covertness Might Avoid Escalation

Here we consider how existing theories about the de-escalatory benefits of covertness might apply to nominally covert actions. Our focus is on public reactions in the targeted country. The public has received little attention in the covert action literature so far, based on the assumption that covert actions are known to elites only. Yet many covert actions are partially or fully visible to the public. Since evidence of responsibility for nominally covert actions is publicly available, it is important to understand how the public reacts to this information and the perpetrator's denials. Elite surveys show that the policy preferences of democratic elites are guided by public opinion on foreign policy.²⁴ The very fact that elites sometimes hide evidence of covert actions targeting their country from the public is further evidence that they view public opinion as a constraint.²⁵ Therefore, we cannot understand government reactions to nominally covert actions without understanding the public reactions that shape government decisions. Later, we consider the extent to which the same causal mechanisms might also directly influence elite reactions.

We compare three causal mechanisms that might explain how even nominal covertness can reduce public demand for escalation. Building on claims that non-acknowledgement can save face,²⁶ we consider both rationalist and psychological explanations for how nominal covertness can reduce the public's perception that escalation is necessary to restore reputation or status. Building on other scholarship,²⁷ we consider how a third mechanism, uncertainty, can operate even in the presence of public evidence. We end our discussion by comparing the implications of the mechanisms.

Reputational Mechanism

The first reason that the public may be less likely to demand escalation in response to a nominally covert action, as compared to an overt one, relates to reputation. Maintaining a reputation for resolve is believed to help countries deter challenges and make credible threats.²⁸ Research shows that policymakers are motivated to preserve reputation and that the public disapproves of elites who damage the country's reputation.²⁹ This suggests that the public may desire escalation in response to hostile actions to defend their country's reputation.

How might making an action nominally covert alleviate reputational concerns? One way is that the target might be able to save face by pretending to be unaware of the action or of the perpetrator's identity. Consider a professor with a policy against reading outside material during class. If the professor ignores a student reading on their phone, this will not necessarily weaken the professor's reputation because it is plausible that the professor did not notice. In the past, however, it was more common for students to read newspapers. Since holding up a newspaper is more obvious, this put the professor's reputation on the line to a greater extent. The same dynamics can apply to international relations. If the perpetrator does not acknowledge an action and the target government feigns ignorance about responsibility, then the public might view this feigning—even if not fully credible—as sufficient for avoiding reputational damage.

Even when the target government does not feign ignorance, nominal covertness might reduce perceptions of reputational damage. If the perpetrator openly admitted the operation, individuals might interpret this as a deliberate challenge. As O'Neil explained,³⁰ the essence of a challenge is that it is intended to be a challenge. Schelling similarly says, “If you are publicly invited to play chicken and say you would rather not, you have just played.”³¹ Overt actions might therefore be interpreted by the public as a dare that their country cannot decline without losing face. In contrast, going to the trouble of denying an action shows respect for the target's reputation, indicating the action is not intended as a dare. This should reduce beliefs that it is necessary to escalate for reputational reasons, which suggests the following hypotheses.

H2a Perceptions of reputational damage among the public in the targeted country will be lower when a hostile action is denied than when it is done overtly.

H2b The effect of nominal covertness on the public's escalation preferences will be mediated through beliefs about reputational damage.

Psychological Mechanism

The next mechanism through which nominal covertness might reduce escalation pressure is psychological. We expect individuals in the targeted state to feel less insulted and humiliated by nominally covert actions than by overt actions. Humiliation is a negative emotional reaction to loss of status,³² and insults are disrespectful actions that trigger humiliation.³³ Humiliation is important in international relations because it enhances the psychological payoff of revenge and can shift cost sensitivity, leading decision makers to value the benefits of conflict more and the costs less.³⁴ Thus, insults and the resulting humiliation often prompt states to take escalatory action to restore their status.

Importantly, feelings of humiliation typically result from the perception that an insult is intentional. Gilbert notes that humiliation, in contrast with shame, imposes blame on another for a loss of status.³⁵ Likewise, Miller says that humiliation is not merely the perception of having low status but the feeling of being “put into that state by another person.”³⁶ Thus, as with the reputational argument, intentionality is important when considering whether a hostile action provokes humiliation. In the case of nominally covert actions, the veneer of covertness suggests that the perpetrator is not intentionally disrespecting the target's status. Therefore, members of the public may perceive a lesser insult and feel less humiliated, dampening their emotional desire for revenge. This yields another pair of hypotheses:

H3a Members of the public in the targeted country will be less likely to perceive an insult when a hostile action is denied than when it is done overtly.

H3b The effect of nominal covertness on the public's escalation preferences will be mediated through the degree to which respondents feel insulted.

It might be counterargued that implausible denials can be interpreted as insulting to observers’ intelligence. This could be true if they were interpreted as attempts to deceive, but we argue that those savvy enough to realize denials are false are likely to understand that they serve another purpose. For example, research shows that Trump supporters who recognize statements by Trump as false perceive the statements as signals of willingness to flout norms rather than attempts to hoodwink them.³⁷ Similarly, the logic of the insult mechanism predicts that nominal covertness will be perceived as an attempt to avoid disrespect rather than to deceive.

Uncertainty

A third way that nominally covert operations might provoke less public demand for escalation is through uncertainty about the perpetrator's identity. Hedgecock and Sukin's survey experiment shows that the degree of certainty reported by the target government about responsibility for an attack has a large effect on public support for retaliation.³⁸ However, they consider a scenario in which no evidence is available beyond the target government's terse statement, and the accused government makes no response. Can uncertainty also play a role when there is more obvious evidence of responsibility? We argue that even when credible evidence is publicly available, some individuals will doubt the evidence and believe denials.

One reason for doubt is distrust of authority. Without the perpetrator's admission, evidence of their identity will usually come from the government or media. Beyond releasing a statement of attribution, governments can release more specific intelligence to prove their case. The media can convey independent information from a wider variety of sources. Yet public trust in both the government and the media has declined in the United States and other countries.³⁹ In the US, faulty intelligence about Iraq's “weapons of mass destruction,” and the media's failure to question it, contributed to this decline. Trust in expertise in general has also declined.⁴⁰

Another reason some individuals might remain uncertain in the face of evidence is motivated reasoning, a tendency to reject information that threatens a preferred worldview. For example, research shows that hawkish individuals are reluctant to accept positive information about Iran.⁴¹ In the case of evidence about responsibility for hostile acts, both dovish individuals who dislike military action and individuals with a positive attitude toward the perpetrator may reject the evidence.

Individuals who remain uncertain about responsibility will be less likely to support an escalatory response, due to fear of directing it at the wrong state. This yields the last set of hypotheses:

H4a Certainty about responsibility among the public in the targeted country will be lower when a hostile action is denied than when it is done overtly.

H4b The effect of nominal covertness on the public's escalation preferences will be mediated through certainty about responsibility.

Implications of the Mechanisms

These three mechanisms are not mutually exclusive. They may coexist and reinforce each other. However, distinguishing among them is important, not only for theoretical reasons but also for understanding (1) the applicability of our findings to other audiences and (2) which countries are most likely to be targeted by nominally covert actions.

We focus on the public reaction in the targeted state, but there are also other potentially important audiences for nominally covert actions. The first is elites in the targeted state. Elites may experience less uncertainty because they have more information.⁴² Yet the reputation and insult mechanisms are likely to be highly relevant to elites. Scholars agree that elites are obsessed with maintaining their country's reputation.⁴³ Moreover, foreign policy elites are probably more likely than the public to identify themselves closely with their country and therefore take insults personally. Thus, to the extent that the reputation and insult mechanisms exist among the public, it is likely that they function at least as strongly among elites. This suggests that, if we find a de-escalatory effect of nominal covertness in the public sample that is mediated by the reputation and insult mechanisms, it will have a similarly de-escalatory effect among elites (even leaving aside public pressure on elites).

Another relevant audience for understanding the effect of nominally covert actions is the public and elites in third-party states that could potentially punish the perpetrator or assist the target. We expect the reputation and insult mechanisms to generally have little applicability to third-party audiences, unless the third-party state has staked its reputation on defending the target. Even when reputation is at stake, the situation is likely to have less emotional resonance than a direct attack, reducing the relevance of the insult mechanism. In contrast, we expect the uncertainty mechanism to apply at least equally to third-party audiences, which may be motivated to reject evidence of responsibility as an excuse to avoid involvement. This suggests that, if we find a de-escalatory effect of nominal covertness in the target public sample that is mediated by the uncertainty mechanism, it will have a similarly de-escalatory effect among third-party audiences. Therefore, the support we find for the different mechanisms will affect our predictions about where our results will travel.

The various mechanisms also have different implications for which countries are most likely to be targeted by nominally covert attacks. Perpetrators are most likely to employ nominal covertness when it has the greatest de-escalatory benefit. If nominal covertness is particularly good at reducing reputational concern, then it would be especially useful against countries that have high reputational stakes because of numerous security commitments. Alternatively, if nominal covertness is particularly good at decreasing the perception of an insult, then it would be good to use against countries with strong honor cultures. Finally, if the main benefit of nominal covertness is generating uncertainty, then it is most likely to be effective against countries where citizens distrust evidence from their own government and media. Thus, understanding which mechanisms make countries vulnerable to nominal covertness can help us predict which countries will be targeted and how frequently. It is also a first step toward developing strategies to counteract nominal covertness.

A final caveat is that the importance of each mechanism might vary with the perpetrator's identity. For example, when the perpetrator is a more hostile country, its denials might create less uncertainty. The reputational stakes might also be higher, which could either enhance or detract from the ability of denials to ease reputational concerns. We explore this potential variation by using scenarios with two different attackers.

Research Design

We test our hypotheses in three preregistered survey experiments fielded on representative samples of the US public provided by Lucid. We use experiments to eliminate biases arising from strategic behavior. Real-world leaders are most likely to employ nominal covertness when it is expected to be most effective at preventing escalation, possibly creating upward bias in its observed effectiveness. The targeted government may also strategically influence which evidence of responsibility comes to light. Experiments can eliminate these confounding factors by holding all aspects of a scenario constant except for the denial.

To maximize emotional engagement, our experimental vignettes used real country names and read like short news reports, though they were labeled as fictional. Subjects in each experiment read a vignette in which a country attacked US oil tankers in the Persian Gulf, sinking two of them. All respondents were told that independent observers identified the attacking country, but the control group was told the country claimed responsibility, while the treatment group was told the country denied it.

In the main experiment, fielded in September 2022 in a sample of 954 subjects, respondents were told that observers identified the attacking country as Iran. Identifying Iran, a distrusted US adversary, as the attacker creates a hard test for the effect of denials. Our other experiments are variations on the first. In our second experiment, fielded in September 2022 in a sample of 477 subjects, subjects were told that observers identified the attacking country as Qatar, a nation with less adversarial relations with the US. Comparing experiments 1 and 2 will allow us to evaluate how much the hypothesized effects vary with the identity of the attacker. In our third experiment, fielded in July 2023 in a sample of 492 subjects, we again identified Iran as the attacker, but we included stronger evidence of Iranian responsibility.

Vignettes

The first two experiments had identical wording, except for the country name. After consenting to participate, participants were presented with two pretreatment attention-check questions. Those who answered either question incorrectly were dropped from the survey. Participants were then shown the scenario, which began:

On March 22nd, 2023,⁴⁴ sources reported that United States oil tanker ships were attacked in the Persian Gulf. Two of the US ships were sunk. Independent observers said the attacking ships looked like Iranian [Qatari] naval vessels.

Participants in the treatment condition were told, “Despite this, the Iranian [Qatari] foreign minister stated: ‘The Iranian [Qatari] government denies any involvement in the attacks.’” In contrast, participants in the control condition were told, “Later, the Iranian [Qatari] foreign minister confirmed: ‘The Iranian [Qatari] government claims responsibility for the attacks.’” For both groups, the scenario ended with the sentence, “The United States has a large naval base in the region, but it is unclear whether it will respond.”

Our first two experiments attributed the assessment of responsibility to “independent observers.” This wording was intended to keep the vignette simple and imply that the observers were credible because of their independence. We acknowledge, however, that the lack of detail about the observers might leave room for doubt about their reliability. Therefore, our third experiment gave more detailed evidence of responsibility. Experiment 3's vignette was mostly identical to experiment 1, but we replaced the sentence about independent observers identifying Iranian ships with this longer text:⁴⁵

Crew members aboard the targeted ships as well as sailors on other ships in the vicinity identified the attacking ships as Iranian naval vessels. The US government later released satellite imagery showing that the ships came from Iran. Intelligence released by other governments and investigative reporting by independent news outlets also provided evidence that Iran's leadership ordered the attacks.

This level of evidence goes beyond what is typically available in reality, but we wanted to see whether denials could have an effect even in the most extreme scenario of obvious responsibility.

We considered various issues in developing our vignettes. First, we considered the attack's target. We chose the scenario of sinking US commercial ships because we expected it would be sufficiently provocative to make some members of the public favor escalation, but not such a grave threat that escalation was a foregone conclusion. Second, we considered the military power of the attacker. The attackers in our vignettes are weaker countries than the United States. Since escalation against weaker countries is less devastating, we expected this to allow more variation in escalation preferences.

Third, we considered including partial denials, ambiguous statements, or total silence about covert actions as additional treatments. Total silence is the norm for truly covert actions, but less common for nominally covert actions because public evidence puts pressure on governments to respond.⁴⁶ In contrast, partial denials and ambiguous statements about nominally covert actions are quite common. For example, Israel's standard response is to neither confirm nor deny. We expect that partial denials, ambiguous statements, and silence probably have a lesser de-escalatory effect than total denials. Partial denials and ambiguous statements often imply responsibility, and even silence may be taken as a tacit admission of guilt. However, for this initial study, we chose to make the most straightforward comparison by comparing only total denials with open admissions.

Measurement of Variables

After reading the scenario, participants were asked about their escalation preferences and then about reputational concerns, perception of an insult, and certainty about responsibility, in random order. We measured escalation preferences by asking respondents to rate the extent to which they favored or opposed various policy options, including (1) publicly condemning Iran/Qatar, (2) imposing economic sanctions, (3) conducting airstrikes, and (4) declaring war and invading. Based on H1, we expect support for all escalatory options to be lower following a denial. To test H1, we created a dependent variable by taking the average of the standardized responses for the four escalation options and scaling it between 0 and 100. A higher value indicates greater preference for escalation.⁴⁷

We measured reputational concern by asking respondents to rate the likelihood of three outcomes if the US did not respond to the attack: (1) US credibility, prestige, or reputation would suffer; (2) other countries would be emboldened to attack the US and its allies; and (3) the US would be unable to deter other countries from attacking. We averaged the standardized ratings of the three options and scaled the average between 0 and 100. To measure perception of an insult, we asked, “As a US citizen, how insulted do you feel based on the events in the above scenario?” with answers on a five-point scale, giving us an ordinal variable. Finally, we measured uncertainty by asking, “Based on what you read, do you think that Iran/Qatar was responsible for the attacks?” with answers on a four-point scale, yielding another ordinal variable.

After answering these questions, respondents saw a manipulation-check question, asking them whether the attacker denied or acknowledged responsibility.⁴⁸ In each experiment, over 93 percent of subjects answered correctly. We do not drop subjects who answered incorrectly,⁴⁹ but the high correctness rate provides confidence that treated subjects truly received the treatment. Finally, we asked questions measuring dispositional traits, including militant assertiveness,⁵⁰ national chauvinism,⁵¹ trust in media and government,⁵² trust in other nations,⁵³ and foreign policy interest.

We estimate three versions of each statistical model: one with no controls, one controlling for dispositional attributes, and one controlling for demographic attributes.⁵⁴ As indicated in our pre-analysis plan, we focus on the version with dispositional controls. The others, shown in the online supplement, support the same substantive conclusions.

US Focus and Relevance Abroad

We ran our experiments in the United States for several reasons. First, the US has been targeted by many nominally covert actions, including Russian election interference, Chinese hacking, and Iranian assassination plots. Second, given the US military's exceptional power, US escalatory responses have implications for the entire world. For example, the US launched two major wars in Afghanistan and Iraq, upending regional dynamics and causing hundreds of thousands of deaths, in response to the September 11 attacks. Third, public opinion clearly influences US foreign policy. For example, public opinion prompted President Kennedy to take a hard line in the Cuban Missile Crisis and influenced President Clinton's intervention in Kosovo.⁵⁵

Nonetheless, it is useful to consider the applicability of our research to other countries targeted by nominally covert actions, such as Ukraine and Iran. One important scope condition for the policy relevance of our research is regime type. Our results are most relevant to democracies, because democratic governments are most responsive to public opinion. While responsiveness may vary somewhat among democracies based on factors like election timing and voters’ attention to foreign policy, we do not expect any democratic regime to systematically ignore public opinion on nominal covertness. Therefore, we expect our results to have broad relevance to democratic regimes beyond the US. But they may also have some relevance for nondemocracies. These regimes often cannot totally disregard public opinion, and—as discussed earlier—some of our hypothesized mechanisms may also apply to autocratic elites.

Another consideration is balance of power. The US military is the world's strongest, and our experimental scenarios feature attacks by much weaker countries. If we fielded our surveys in militarily weaker countries, or if we simply used a scenario with a more powerful attacker, we would probably see less variation in escalation preferences. Many people might believe that escalation against an equally or more powerful country is too costly regardless of whether an attack is overt or nominally covert. This suggests an important caveat: the size of our treatment effect might decline as the military balance becomes less favorable due to compression of the range of escalation preferences. Substantively, this means that employing nominal covertness might not make as much of a difference for countries with other means of deterring escalation.

A third concern is that US and foreign citizens might have systematically different dispositional characteristics. For example, US citizens might be more hawkish or nationally chauvinistic. If so, this could be the greatest limitation on our results’ applicability abroad. While regime type might affect policy relevance, and a less favorable military balance might weaken the treatment effect, systematically different dispositional attributes could more fundamentally alter the treatment effect. Bassan-Nygate and colleagues find that many treatment effects can be replicated across different countries and argue that treatment effects are least likely to travel when they are known to be heterogeneous based on dispositional attributes.⁵⁶ This provides a metric for evaluating how likely our treatment effect is to travel.

If our treatment effect is heterogeneous across any dispositional traits, theory suggests the traits of hawkishness (militant assertiveness), national chauvinism, and trust as the most likely candidates. Hawkish and chauvinist individuals might be more concerned with reputation and insults. Meanwhile, individuals who are more trusting of other countries and less trusting of the media and government might be more likely to believe denials over other evidence. To test for heterogeneous treatment effects, we interact our measures of these traits with the treatment.

Experiment 1 Results

Figure 1 plots the regression coefficients from our first experiment. Model 1 is an ordinary least squares model with the continuous escalation-preference measure as the dependent variable. The denial treatment has a highly significant and negative effect on support for escalation. H1 is thus supported based on our preregistered test. Compared to admission of responsibility, the denial treatment decreases the preference for escalation by nearly fourteen percentage points.

Figure 1. Regression coefficients from experiment 1

Because the aggregated escalation preference measure is difficult to interpret substantively, we show the support for each policy option individually in Figure 2. The mean level of support for every action against Iran is significantly lower in the denial (treatment) condition than in the overt (control) condition. The decline in support is largest in percentage terms for the two most escalatory options, an airstrike and war: nominal covertness decreases support for an airstrike by 22 percent and support for war by 19 percent. This provides further support for H1.

Figure 2. Support for escalation options by treatment condition in experiment 1

But why do preferences for escalation decline with a denial? The other models shown in Figure 1 address this question. Model 2 is an ordinary least squares model predicting reputational concern. The denial coefficient is negative and highly significant, indicating that nominal covertness reduces concern about reputational damage. This supports H2a. Model 3 is an ordered logit predicting perception of an insult. Again, the denial coefficient is negative and highly significant. This means the denial reduces the perception of an insult, supporting H3a. Finally, model 4 is another ordered logit with certainty about the attacker's identity as the dependent variable. The negative and highly significant denial coefficient indicates that a denial decreases certainty that Iran was responsible, supporting H4a. So far, the results are supportive of all three hypothesized causal mechanisms: reputational concern, perception of an insult, and uncertainty.

To determine how much each of these mechanisms is responsible for declining support for escalation, we combine the pieces of the causal chain using mediation analysis.⁵⁷ Table 1 shows the results. In every case, the average causal mediation effect (ACME) is significantly different from zero. This indicates that all three mediators have a meaningful effect and that Hypotheses 2b, 3b, and 4b are all supported. However, the certainty respondents feel about the perpetrator's identity has the largest mediating effect. Decreased certainty in the denial condition is estimated to account for 32 percent of the decline in public desire for escalation. In contrast, decreased reputational concern accounts for 22 percent, and decreased perception of an insult accounts for 10 percent.

Table 1. Mediation analysis results from experiment 1

Notes: 95% confidence bounds in parentheses. **p < .05.

Experiment 2 Results

Figure 3 shows regression coefficients from experiment 2, which was identical to experiment 1 except for identifying Qatar as the attacker. H1 is again supported. In model 1, the denial treatment has a negative and significant effect on escalation preferences. The substantive effect is approximately 50 percent larger than in experiment 1. Figure 4 shows that support for each individual escalation option also declines significantly in the denial condition. The negative and significant coefficients of the treatment variable in the other three plots in Figure 3 indicate that Qatar's denial also reduces reputational concern, perception of an insult, and certainty. Thus, Hypotheses 2a, 3a, and 4a are again supported. Overall, these findings are very similar to the previous experiment.

Figure 3. Regression coefficients from experiment 2

Figure 4. Support for escalation options by treatment condition in experiment 2

Table 2 shows the mediation analysis results from the second experiment. All three mediators have ACMEs that are significantly different from zero, meaning that Hypotheses 2b, 3b, and 4b are supported. Certainty, which already had the strongest effect in the previous experiment, is an even more important mediator in experiment 2. Here certainty mediates 60 percent of the total effect, with the next-most-important mediator (reputational concern) mediating 23 percent. The greater effect of certainty in this experiment is probably due to denials from a friendlier country seeming more plausible. Overall, these results show that the perpetrator's identity does influence the operation of the causal mechanisms. However, while the mediation effect sizes change, the ordering of their importance remains consistent. This suggests that we can draw general, and not merely attacker-specific, conclusions about the relative importance of the mechanisms.

Table 2. Mediation analysis results from experiment 2

Notes: 95% confidence bounds in parentheses. **p < .05.

Experiment 3 Results

Our third experiment identifies Iran as the attacker, as in experiment 1, but provides much more evidence, including information from eyewitnesses, the US government, foreign governments, and independent media. Figure 5 shows the regression coefficients. The effect of a denial on escalation preferences is still negative, but smaller—less than a quarter of the size in experiment 1. The coefficient falls short of significance at the standard threshold, although it is significant with 90 percent confidence. In an alternative specification with demographic controls (Table A19), it reaches significance with 95 percent confidence. Therefore, we cannot say that H1 is supported according to our pre-analysis plan, but there is weak evidence that a denial dampens escalation preferences. Figure 6 shows that a denial also dampens preferences for each individual escalation option, but the differences are small and are significant for only the condemning and sanctioning options.

Figure 5. Regression coefficients from experiment 3

Figure 6. Support for escalation options by treatment condition in experiment 3

The effect of a denial on the perception of an insult is insignificant in Figure 5. However, Iran's denial continues to have a negative and highly significant effect on reputational concern and—despite all the evidence—certainty about Iran's responsibility. The mediation analysis results (Table 3) similarly show that certainty and reputational concern have ACMEs that are significantly different from zero, but the perception of an insult does not. Therefore, the certainty hypotheses (H4a and H4b) and reputation hypotheses (H2a and H2b) are supported in experiment 3, but the insult hypotheses (H3a and H3b) are not.

Table 3. Mediation analysis results from experiment 3

Notes: 95% confidence bounds in parentheses. **p < .05.

A surprising aspect of the mediation analysis results is that the direct effect is positive (although insignificant) in the row for certainty. This suggests that were it not for the restraining effect of lower certainty, hearing an Iranian denial would make the public more eager to escalate. The fact that certainty's ACME and the direct effect push in opposite directions requires us to calculate the proportion mediated by certainty differently and makes the proportions mediated by certainty and reputational concern difficult to compare.⁵⁸ Nonetheless, the evidence suggests that both mediators contribute to dampening escalation preferences.

Interactions with Respondent Characteristics

Finally, we consider across all three experiments whether the treatment effect varies with militant assertiveness (hawkishness), national chauvinism, international trust, trust in news, or trust in government.⁵⁹ We analyze interactions between each of these five dispositional variables and our treatment indicator. We estimate interaction models predicting not only escalation preferences but also the three mediators. We estimate a separate model for each interaction, to avoid multicollinearity.

The results appear in Tables A23–A37 in the online supplement. We first consider the fifteen models predicting escalation preferences (model 1 in each table). The interaction coefficient is statistically significant with 95 percent confidence in only three of these fifteen models: when interacting the treatment with national chauvinism in experiment 3, when interacting the treatment with international trust in experiment 1, and when interacting the treatment with government trust in experiment 3. We find similarly weak and inconsistent evidence in the models predicting the mediators. Across forty-five models we find only six interaction coefficients that are significant with 95 percent confidence, and there is no pattern in which ones are significant.

Overall, the rare and inconsistent significance of the interaction coefficients suggests that the US public's reaction to nominal covertness is not particularly contingent on any of the dispositional factors we considered. This allows optimism that our results are relevant to other countries, where dispositional attributes may be distributed differently.

Conclusion

This research note provides the first direct evidence that making hostile actions nominally covert reduces preferences for escalation among the public in targeted states. Across all three experiments, we see evidence that a denial, even in the face of evidence, has a de-escalatory effect. This evidence is strong in experiments 1 and 2, but weaker in experiment 3, where it falls short of our preregistered standard. A denial's de-escalatory effect is strengthened when it comes from a friendlier country (as in experiment 2) but weakened by stronger evidence (as in experiment 3). We find that uncertainty about responsibility, reduced reputational concern, and lesser perception of an insult all contribute to dampening demand for escalation after a denial. However, only uncertainty and reputational concern have a consistent effect across all three experiments. Uncertainty is the most important mediator in the first two experiments and continues to play an important role even in the face of extensive evidence in experiment 3.

For governments considering undertaking hostile actions, these findings suggest that even a thin veneer of covertness reduces the escalation risk. This suggests denials can moderate escalation risks not only for intelligence and special forces operations, but even for larger actions involving conventional forces. This helps explain why denials of responsibility for starting military skirmishes are so common.⁶⁰ Our evidence of the effectiveness of denials even has relevance for situations outside international relations where actors face allegations of wrongdoing, such as in domestic politics or business.

Our findings have implications for when nominal covertness is most likely to be effective and when it will be employed. The support we find for the reputation and insult mechanisms speaks to the international reputation literature by showing that reputational concerns are real, but also that they can be mitigated. This suggests that nominal covertness can be an effective strategy for reducing escalation risks even when targeting countries that would otherwise be most likely to escalate—those with interdependent commitments and cultures that emphasize honor.

The implications of the uncertainty mechanism are even more timely. The willingness of respondents to believe denials reflects substantial distrust in the sources of evidence. Thus, even as the modern information environment makes covert actions harder to hide,⁶¹ they are likely to remain easy to deny. Moreover, current trends suggest that such distrust is likely to increase. The rise of cyberattacks makes evidence of responsibility harder for laypeople to understand, increasing the need for the public to trust experts. At the same time, other trends in the United States and elsewhere are increasing skepticism of expertise and mistrust of the government. A fragmented media landscape causes people to question information from sources once considered authoritative. Political polarization also reduces trust in the government and expertise, as different parties promote different versions of the truth and some elites attack government credibility. Finally, the ability of artificial intelligence to create persuasive fake evidence may further reduce trust in real evidence. These trends, coupled with our evidence for the importance of the uncertainty mechanism, suggest that political polarization and fragmented media not only create domestic political problems but also pose a risk to national security. Knowing that even a poorly disguised lie can dampen support for escalation, adversaries are likely to increasingly target the United States and other polarized countries with nominally covert actions.⁶²

Another important innovation of our research is bringing greater focus on the public to the covert-action literature. Many covert actions are at least partially visible to the public, and therefore the targeted country's public is likely to influence government responses to them. At the same time, we believe our results are relevant to additional audiences beyond the target country public. We argued earlier that the reputation and insult mechanisms are most relevant to elites. Finding that the internal logic of these mechanisms is valid, we expect them to apply to elites also. We argued that the uncertainty mechanism is relevant to third-party states. Our results also validate the internal logic of this mechanism, meaning it is likely to apply among third-party audiences too. Taken together, this leads us to expect that nominal covertness will also dampen escalation preferences among elite and third-party audiences. The finding that our results are not very contingent on dispositional attributes also suggests they can travel to other countries. A caveat is that a less favorable military balance may shrink the treatment effect and that governments vary in their responsiveness to public opinion.

Ultimately, it would be valuable for future research to test the effect of nominal covertness in different samples and scenarios. Another direction for future research is exploring the effect of partial denials, ambiguous statements, and silence about covert actions. A third direction is to explore the effect of denials by countries that public opinion is more divided on, such as Russia. Fourth, it would be valuable to investigate how the perpetrator's own domestic audience responds to nominal covertness. Finally, future research could further explore other benefits of nominal covertness, beyond avoiding escalation.