Do External Threats Unite or Divide? Security Crises, Rivalries, and Polarization in American Foreign Policy

Abstract

A common explanation for the increasing polarization in contemporary American foreign policy is the absence of external threat. I identify two mechanisms through which threats could reduce polarization: by revealing information about an adversary that elicits a bipartisan response from policymakers (information mechanism) and by heightening the salience of national relative to partisan identity (identity mechanism). To evaluate the information mechanism, study 1 uses computational text analysis of congressional speeches to explore whether security threats reduce partisanship in attitudes toward foreign adversaries. To evaluate the identity mechanism, study 2 uses public opinion polls to assess whether threats reduce affective polarization among the public. Study 3 tests both mechanisms in a survey experiment that heightens a security threat from China. I find that the external threat hypothesis has limited ability to explain either polarization in US foreign policy or affective polarization among the American public. Instead, responses to external threats reflect the domestic political environment in which they are introduced. The findings cast doubt on predictions that new foreign threats will inherently create partisan unity.

A prominent explanation for the increasing partisanship in American foreign policy since the early 1990s is the collapse of a common external threat, the Soviet Union, which fostered bipartisanship over foreign affairs during the Cold War.¹ An “external threat hypothesis” posits that security threats from foreign adversaries reduce polarization among political officials and the public. This argument has received renewed scholarly and popular attention in the current polarized era. Walt, for example, argues that a benign international threat environment is partly to blame for political dysfunction in the United States and Europe.² Brooks asserts that a security threat posed by a rising China would be good for national unity because “Americans would come together when we realized that we face a dangerous foreign foe.”³

While the external threat hypothesis may be intuitively appealing, it is also troubling. Leaders who subscribe to this narrative may believe they should amplify or create security crises to unify their publics. Yet in reality we have little systematic evidence about whether external threats foster partisan convergence over foreign policy in the United States. One reason this argument is pervasive is that it is difficult to falsify. The security studies literature primarily evaluates this hypothesis by tracing variation in the threat environment of the United States across historical periods. This approach, however, is unsatisfactory. Analyses of congressional foreign policy preferences show that polarization over foreign policy increased throughout the 1990s and 2000s,⁴ a pattern consistent with the external threat hypothesis. However, this timing also coincides with major changes in domestic politics, including the realignment of southern states with the Republican Party and institutional reforms in Congress that facilitated party unity.⁵ A more robust test of the external threat hypothesis would start with other testable hypotheses beyond the fact that Democrats and Republicans are more polarized presently than during the Cold War.

This paper begins with a conceptual framework for understanding the external threat hypothesis and introduces new evidence to evaluate it. I argue that narratives about the impact of external threats on polarization in the United States explain one of two dependent variables: converging political attitudes toward foreign adversaries, and increasing social cohesion. These variables map onto two concepts of partisan polarization. The first is preference or issue-based polarization, which refers to divergence between the preferences of the Republican and Democratic parties in a single issue area or across a group of issues. The second is social or identity-based polarization, commonly called affective polarization, which refers to an increasing tendency to be more favorable toward one's in-party and less favorable toward one's out-party.⁶

I use these dependent variables to outline mechanisms through which external threats could reduce partisan polarization. Through the information mechanism, threatening actions taken by foreign powers reveal information by credibly signaling the gravity of a foreign policy issue. Since information about national security is revealed asymmetrically, policymakers set aside partisan differences and defer to the executive, reducing polarization over foreign policy. Through the identity mechanism, security threats increase the salience of national identity relative to partisan identity, reducing affective polarization. I test these mechanisms in three studies that move beyond the era as the unit of analysis. The studies consider two types of security threat: an imminent threat triggered by a crisis, and a generalized sense of threat from a rival foreign power.

The first study tests an information mechanism by considering whether, in response to crisis events identified by the International Crisis Behavior project,⁷ legislators’ attitudes toward the country that triggered a security crisis converge along party lines. I extract from the Congressional Record all mentions of countries that ever triggered a crisis for the United States. I then use computational text analysis to measure the relative partisanship of speech about each foreign adversary. There is some evidence that historically, rhetoric around a foreign adversary was less polarized when the United States was engaged in a crisis with the country. However, this pattern no longer holds, and many contemporary crises are associated with greater polarization. Since the relative partisanship of legislators’ responses to crises are correlated with ex ante levels of polarization, my findings imply that responses to foreign threats reflect the domestic political environment in which they are introduced.

The second study tests an identity mechanism, using average partisan differences in presidential approval ratings as a proxy for affective polarization. The study examines popular responses to security crises in the United States based on presidential-approval data from Gallup News.⁸ I find that while periods of heightened threat reduce the partisan difference in presidential approval ratings, this effect dissipates quickly. I demonstrate that elite polarization and proximity to politicizing events like elections explain far more variation in the partisan gap in presidential approval ratings than crisis events. These results imply that patterns of affective polarization among the American public are not substantially related to foreign threats.

After finding little evidence that crises reduce domestic polarization, the third study analyzes the impact of a more generalized sense of heightened threat from a rival power. In a survey experiment with 2,500 American adults, respondents read intelligence reports about security threats the United States faces from China. I randomly vary characteristics of the reports and evaluate their impact on attitudes toward China and partisan identity. The results show no evidence for an identity mechanism: highlighting the China threat does not reduce affective polarization. Evidence for an information mechanism is mixed. Information about a foreign threat delivered by nonpartisan experts leads to convergence in attitudes toward a potential adversary. However, when the same information is accompanied by a partisan elite cue, attitudes are further polarized. The results show that new threats introduced in already polarized contexts are likely to sow greater division.

Collectively, these studies offer a new framework to unpack the claims by international relations scholars that the international system shapes domestic polarization. While each study has important limitations, together they demonstrate that the threat environment of the United States has limited ability to explain either patterns of polarization in foreign policy or affective polarization. The evidence is more consistent with the idea that polarization in domestic politics has spilled into foreign affairs. I argue that it is unlikely that threats from foreign adversaries will automatically create partisan unity, and attempts by American political officials to amplify threats could backfire.

The Puzzling Nature of Polarization in American Foreign Policy

Given Senator Arthur Vandenberg's famous adage that “politics stops at the water's edge,” evidence of increased partisanship in American foreign policy is puzzling to international relations scholars. Whether polarization in foreign policy has eroded America's liberal internationalist orientation in foreign affairs remains contested.⁹ However, it is generally accepted that foreign policy debates have become more partisan since the Cold War, and particularly since the 2003 Iraq War. These developments have troubling consequences for foreign affairs. In polarized times, it is difficult to amass bipartisan support for the passage of international treaties or the use of military force.¹⁰ Highly polarized democracies have more volatile foreign policy,¹¹ which can make allies perceive them as less reliable partners.¹² Polarization also leaves democracies vulnerable to foreign interference¹³ and less able to project power abroad.¹⁴

While foreign policy remains less polarized than domestic policy,¹⁵ the increasing polarization in foreign affairs is reflected in public attitudes. Largely as a result of cues from political elites,¹⁶ public opinion around many foreign policy issues has diverged along party lines, including in matters related to national security. For example, Republicans and Democrats have different views on nuclear weapons,¹⁷ humanitarian intervention,¹⁸ multilateralism,¹⁹ and the use of force.²⁰ They also have different attitudes toward many foreign adversaries²¹ and distinct beliefs about what constitutes a threat to national security.²²

To American politics scholars attuned to polarization in domestic politics, it may seem unremarkable that foreign policy is being absorbed into the partisan divide. But to international relations scholars, these developments are surprising. Two strands of the literature conclude that partisan politics should play a minimal role in foreign policy, and national security policy in particular. One strand emphasizes how information asymmetries in foreign affairs make it less susceptible to polarization than domestic affairs. The constitutional authority vested in the president and the sensitivity of intelligence gathering allows the executive to exercise wide latitude over national security.²³ And in times of crisis, information asymmetries stymie polarization by suppressing criticism from the political opposition.²⁴ A second strand of literature emphasizes the incentives political parties have to maintain centrist foreign policies. Successful foreign policymaking requires the abandonment of short-term political interests in favor of long-term commitments. Recognizing the inevitability of executive turnover, scholars argue that regular elections induce a “partisan truce” in which both parties commit to and implement a moderate foreign policy agenda.²⁵

What we know theoretically and practically about foreign policymaking in the United States leads scholars to expect minimal preference divergence in foreign policy between the Republican and Democratic Parties. Yet this has not been the case over the past few decades. To explain this puzzle, some international relations scholars argue that changes to the threat environment of the United States have catalyzed partisan polarization and the social fragmentation that accompanies it.

The External Threat Hypothesis and Its Implications

The narrative that foreign threats facilitate partisan unity is rooted in an extensive literature about external threat and internal cohesion.²⁶ In Folkways, Sumner wrote, “The exigencies of war with outsiders are what make peace inside.”²⁷ Stein says that this observation could be of “considerable utility in studying the cohesion of American political parties … One could take a step backward and ask whether partisanship decreases when the entire nation is threatened.”²⁸ The application of this narrative to US foreign policy hinges on the idea that American grand strategy is tied to the existence of a common enemy which “has wonderfully concentrated the mind, and the energies, of the American nation.”²⁹

One variant of the external threat hypothesis focuses on the immediate impact of crisis and conflict in unifying political officials and heightening national identity among the public.³⁰ Other variants look at the effect of a state's security environment over long periods, noting that periods in which it was relatively benign—for instance, the decades following the Mexican–American War, World War I, and the Cold War—were associated with greater partisan conflict. Scholars in this tradition argue that the collapse of the Soviet threat explains the increasing polarization in American politics over the last thirty years.³¹ In some versions of this argument, the absence of external threat is the primary explanation for contemporary polarization. Bafumi and Parent, for example, argue that domestic politics will depolarize as China rises to challenge American hegemony.³²

Critically, the external threat hypothesis is not just an academic argument but a powerful idea about American foreign policy that continues to resurface. Mead bemoans the “schizophrenic” nature of American foreign policy created by the absence of Cold War threat.³³ Rodriguez laments the fading threat from Osama bin Laden, arguing that a new “threatening external enemy” is needed to reduce domestic polarization.³⁴ Salam wonders whether China or Russia should be America's “orienting enemy … that gives our grand strategy its ideological shape.”³⁵ Language reflecting the external threat hypothesis is so pervasive in the public discourse that Rothkopf caricatures American foreign policy as having a “rampant, untreated case of enemy dependency.”³⁶

At first blush, the external threat hypothesis and its application to post–Cold War American foreign policy seem intuitive. However, closer scrutiny reveals that this hypothesis is extremely difficult to falsify. The literature is complicated by two challenges. First, the independent variable—the threat environment of the United States—is commonly operationalized with respect to historical periods. When the unit of analysis is the historical period or era, attributing variation in polarization to changes in external threat is challenging because other features of domestic politics vary within and across these periods. Second, concepts like “threat” and “polarization” are defined inconsistently in the literature, resulting in many variants of the external threat hypothesis. Unless these underlying concepts can be clarified, the mechanisms that link threat and polarization will remain unclear. The next two sections address these conceptual challenges.

Defining Threat

Consistent with literature on the “second image reversed”³⁷—a term used to describe the effect of the international system on domestic politics—most evidence for the external threat hypothesis in the American context implicitly uses the historical period as the unit of analysis. This is problematic for drawing inferences because while the end of the Cold War removed the Soviet threat, it also coincided with domestic changes that precipitated polarization. For example, southern states realigned with the Republican Party,³⁸ and income inequality steadily rose,³⁹ prompting the Republican Party to shift rightward. Institutional changes in Congress in the 1990s incentivized greater party discipline and less bipartisanship.⁴⁰ Other notable trends include the development of a partisan, fragmented media environment⁴¹ and the increasing coherence of ideology and party identification among the public that reinforced elite polarization.⁴² Since these changes occurred simultaneously, a better evaluation of the external threat hypothesis would look for evidence beyond cross-temporal trends in polarization.

One way scholars address this challenge is to analyze the relationship between threat and domestic politics cross-nationally.⁴³ Bak, Chavez, and Rider, for example, look at the impact of strategic rivalries on domestic cohesion and conflict. The authors note the limits of a “purposefully broad level of analysis” and invite future work to study the effects of external threats with more granularity, emphasizing a lack of research on threat and polarization in particular.⁴⁴ I build on that suggestion by using two alternative conceptions of heightened threat in the American context.

The first approach is to conceptualize threat as a sudden shock to a state's security environment. This approach reflects the “rally ’round the flag” literature in international relations, which assesses domestic responses to imminent security crises triggered by foreign adversaries.⁴⁵ To operationalize this concept, in the first two studies, I use data from the International Crisis Behavior (ICB) project to identify past security crises.⁴⁶ The ICB project defines a crisis as “(a) a threat to basic values, with simultaneous or subsequent (b) high probability of involvement in military hostilities, and the awareness of (c) finite time for response to the external value threat.”⁴⁷ Conceptually, this approach is useful because it captures “threat” rather than entry into conflict. A crisis by definition involves a high probability of military involvement, but not all crises result in military action. Thus, crisis data measure what is perceived as threatening by foreign policy decision makers at the time rather than defining threat post hoc. Another advantage of using crisis data is that crises are associated with a finite response time. We can more easily observe the reactions of policymakers or the public and attribute them to crisis events.

The second approach is to assess the impact of a more generalized sense of heightened threat from a rival foreign power. This approach mirrors variants of the external threat hypothesis that conceptualize threats as emerging over time from a common enemy or a prolonged strategic rivalry.⁴⁸ Strategic rivals have a competitive relationship that could become militarized.⁴⁹ In contemporary American foreign policy, China is viewed as the “most likely” rival great power⁵⁰ and therefore the most plausible case for the external threat hypothesis. To operationalize a heightened China threat, in the third study, I use reports from the US intelligence community to manipulate threat perceptions in a survey experiment of American adults.

Defining Polarization

A second reason the external threat hypothesis is difficult to falsify is that the mechanisms by which it operates are unclear. In part, this is because scholars invoke different dependent variables to explain the link between threat and polarization. The American politics literature distinguishes between two types of polarization: issue based and identity based. These two concepts create different variants of the external threat hypothesis. In one variant, foreign threats reduce the divergence in political preferences between Republicans and Democrats. Specifically, threats reduce preference or issue-based polarization by leading to shared perceptions of threat and convergence in attitudes toward the adversary. In a second variant, foreign threats increase social cohesion along political cleavages. In doing so, external threats reduce affective or identity-based polarization: increasing favoritism toward one's political in-group and dislike of one's political out-group.

Combining these definitions of threat (as arising from either a crisis event or a heightened security concern from a strategic rival) and polarization (as either issue based or identity based) leads to four variants of the external threat hypothesis (Table 1). Now that we have clarified these variables, the next sections use this framework to outline the mechanisms by which security threats may reduce domestic polarization, which are then tested across three empirical studies.

Table 1. Variants of the external threat hypothesis

An Information Mechanism: Converging Attitudes

How would external threats reduce issue-based polarization? One theory is that periods of heightened or prolonged threat cause convergence in attitudes toward an adversary by revealing information to political officials and the public. To political officials, threatening actions taken by an adversary, such as its invading a neighboring country or acquiring nuclear material, provide credible information about the gravity of a crisis or ongoing rivalry. Since political officials juggle an immense foreign policy portfolio, these actions focus policymakers on a shared objective. In heightening the “stakes of the game,” security threats cause policymakers to set aside partisan differences to resolve a pressing problem.

A key way security threats could reduce polarization is by suppressing criticism from the political opposition.⁵¹ Given the large informational advantage the executive branch has in national security issues,⁵² the political opposition tends to respond to heightened threat by being either “silent or supportive” of the president.⁵³ The opposition's tendency to defer to the executive on matters of national security should in theory lead to greater bipartisanship in response to a crisis.

Among the public, threats from foreign powers could also cause convergence in attitudes toward an adversary. Although citizens often hold coherent beliefs about foreign policy,⁵⁴ asymmetries of information between political officials and the public make the impact of elite cues stronger in foreign affairs relative to domestic affairs.⁵⁵ If the public observes elites responding to a crisis in a bipartisan manner, their evaluation of the crisis is less likely to be partisan.⁵⁶ These processes of partisan convergence among the public and political officials can be mutually reinforcing. Revealed information about external threats increases public attention to foreign policy issues and heightens scrutiny of political officials, prompting them to act in the national interest.⁵⁷

In sum, arguments in the literature suggest that by credibly signaling the gravity of the situation, threatening actions abroad focus policymakers on a common goal and suppress criticism from the opposition. The seriousness of the issue at hand and the greater public scrutiny that accompanies it should elevate foreign policymaking above traditional politicking. Therefore, a first implication of the external threat hypothesis is that threats, whether in the form of an imminent crisis or a prolonged threat from a rival foreign power, lead to partisan convergence in attitudes toward the adversary among political officials and the public. There are two variants of the Information Hypothesis:

H1A Heightened threat from a security crisis causes partisan convergence in attitudes toward the state that initiated the crisis.

H1B Heightened threat from a rival foreign power causes partisan convergence in attitudes toward the rival state.

An Identity Mechanism: Increasing Cohesion

External threats could also reduce identity-based polarization by increasing social cohesion. Individuals have multiple identities that are activated in different social contexts.⁵⁸ Threats induce bipartisanship by changing the relative salience of two identities: partisan identity and national identity. In American politics, partisanship is often considered a social phenomenon.⁵⁹ Partisan identity reflects attachment to one's political party. By contrast, national identity reflects an “internalized sense of belonging to the nation.”⁶⁰ National identity tends to be more “sticky” than partisan identity in America, because it is easier to change one's political party than one's nationality. It is also thought to be superordinate to partisan identity; some Americans may be Democrats and some Republicans, but all share a common national identity.

Theories of intergroup bias posit that one way to reduce animosity between conflicting social groups is to activate a superordinate identity.⁶¹ For example, Levendusky shows that increasing the salience of American identity reduces hostility to noncopartisans.⁶² The external threat hypothesis hinges on the idea that periods of heightened or prolonged threat reduce partisan animus, making national identity more salient than partisan identity. Huntington, for instance, laments the loss of a common enemy from the Cold War that created “identity and cohesion among people.”⁶³ Heightening the threat environment of the United States should lead citizens to define themselves in relation to a foreign adversary rather than in relation to one another, so animosity toward the out-party and affective polarization should decrease. There are two variants of the Identity Hypothesis:

H2A Heightened threat from a security crisis increases social cohesion and reduces affective polarization.

H2B Heightened threat from a rival foreign power increases social cohesion and reduces affective polarization.

Clarifying the concepts of threat and polarization and outlining the mechanisms that link them allows me to test the external threat hypothesis using units of analysis other than the historical period or era. I look for evidence of the information and identity mechanisms using three research strategies. The first two studies assess historical evidence for the information mechanism (study 1) and the identity mechanism (study 2) in response to security crises triggered for the United States. Study 3 probes the logic of both mechanisms in the context of heightened threat from a rival foreign power using a survey experiment about a rising China.

Study 1: Security Crises and the Information Mechanism

Study 1 tests the Information Hypothesis (H1A), which anticipates that heightened threat from security crises causes partisan convergence of attitudes toward an adversary. I use computational text analysis of the Congressional Record to assess whether rhetoric from Republican and Democratic legislators diverges or converges in response to crises triggered by foreign adversaries. The analyses show a weak negative correlation between heightened threat and polarization of rhetoric, an effect driven by responses to crises in the first half of the twentieth century. This relationship diminishes over time as rhetoric around security crises increasingly polarizes. Overall, crises do not appear to be strongly associated with partisan convergence in attitudes. Instead, reactions to crises reflect the relative polarization of the domestic environment in which they arise.

Research Design

Assessing whether external threats unite or divide the two major American political parties requires a measure of differences in attitudes toward foreign actors by party across time. For example, when the Soviets invade Afghanistan in December 1979, we need to observe how Republican and Democratic attitudes toward the Soviet Union converge or diverge in response to the crisis. A standard way to measure issue-based polarization is to use multidimensional scaling methods to estimate the positions of political parties based on congressional roll-call votes.⁶⁴ Yet analyzing roll-call votes in this context is difficult because votes related to specific adversaries are infrequent. For example, between 2000 and 2015, the United States was involved in four separate nuclear crises with North Korea, but only eighteen roll-call votes related to North Korea took place.⁶⁵

Inferring polarization from foreign policy votes can also be misleading because strong polarization may shape the scope or content of legislation. For instance, legislators may pass an expressive resolution condemning an adversary for human rights violations because there was strong partisan disagreement over taking more concrete foreign policy actions. If this expressive resolution is adopted unanimously, we could mistakenly infer that there was no polarization around the issue.

Instead of relying on roll-call votes, this study uses partisanship of speech—specifically, computational text analysis of the Congressional Record from the 43rd to the 114th Congress—to estimate issue-based polarization with respect to attitudes toward adversaries across time. The intuition is that if Republican and Democratic political officials use systematically different language to describe the same country at the same time, they have different perspectives on how to approach relations with that country.

The full text of the Congressional Record provides a much richer data source about attitudes toward foreign adversaries. During the same period (2000–2015) discussed previously, legislators made 4,468 speeches discussing North Korea. Relying only on roll-call votes would mask considerable debate in Congress on US–North Korean relations. Text data can also reveal partisan dimensions of an issue that are unobservable in roll-call votes. For instance, if the majority party prevents legislation from reaching the floor precisely because an issue is too polarized, we can still observe dissent from the minority party in the Congressional Record in the absence of a floor vote.

Since I am interested in understanding US relations with adversaries in the context of security threats, I start with the thirty countries that have ever triggered a security crisis for the United States, as defined by the ICB project.⁶⁶ I manually extend the ICB data backwards to 1873, the start of the Congressional Record, for crises involving the United States. I extract what ICB refers to as the “triggering entity,” the actor(s) that triggered an international crisis for the US government.

This process identifies seventy-one crises triggered by thirty countries from 1873 to 2015 (see the online supplement). I extract all speeches about each crisis country from the digitized Congressional Record (which includes floor debates, conference committee reports, and statements submitted by legislators) in which these countries are discussed.⁶⁷ To measure the relative partisanship of speech about each country, I use a supervised machine-learning method from Peterson and Spirling, who estimated polarization using the text of British parliamentary debates.⁶⁸ I first create a database of congressional speeches in which any of the thirty crisis countries is mentioned. The speeches are cleaned and segmented by country and congressional session, such that the unit of analysis is a “country-session.” Country-sessions in which there are not a minimum of fifty Republican and fifty Democratic speeches are dropped from the analysis. Vocabulary is fixed across all the speeches; words used infrequently (i.e., that do not appear in at least 200 unique speeches) are removed from the data.⁶⁹ Within each country-session, I use the Peterson and Spirling method to run different classifiers (machine-learning algorithms) and select the classifier that best predicts legislators’ parties based on their speech.⁷⁰

I use this particular method to develop the proxy for polarization because it has higher construct validity than other common approaches in computational text analysis. As Goet explains in a review of methods for measuring partisanship in speech, “In contrast to their unsupervised siblings, such supervised models attempt to identify which speakers use a vocabulary that is similar to speakers from one versus another party, ensuring that variation in word use is related to a stable construct.”⁷¹ In other words, supervised machine-learning methods ensure that the measure of polarization is capturing variation tied to the speaker's political party rather than an unknown latent dimension that may not be directly related to partisanship.

The predictive accuracy of the classifier, which in theory ranges from 0.5 to 1, is a proxy for polarization. At 0.5, the classifier correctly predicts the party of the speaker based on their speech 50 percent of the time—effectively, a random guess. This is an environment of low polarization in which there is no major distinction between Republican and Democratic speech. At 1, the classifier correctly predicts the party of the speaker 100 percent of the time, indicating an environment of maximum polarization. Using a fixed vocabulary across all country-sessions, the result is a data set of thirty countries between from the 43rd to the 114th Congressional sessions (1873–2017). Each country-session has a dependent variable: a measure of the predictive accuracy of the best performing classifier, which captures the polarization of rhetoric. The advantage of this approach is that I am able to compare levels of polarization across foreign adversaries, across congressional sessions, or both.

Results

To test the Information Hypothesis (H1A), I assess the relationship between security crises and partisanship of speech about the relevant adversary. The models in Table 2 regress the predictive accuracy of the classifier on indicators for whether a crisis with the country occurred during that congressional session. The models include country fixed effects, session fixed effects, and both simultaneously. The Information Hypothesis anticipates a negative relationship between crisis events and polarization of rhetoric. Some coefficients in Table 2 are negative, but the results are inconsistent. The magnitude of effect sizes is small, ranging from 0.04 to 0.2 standard deviations of the dependent variable. Only in model 1, which holds fixed the country that triggered the crisis but does not account for common shocks across time, is the coefficient negative and statistically significant at conventional levels. In sessions following a crisis, the negative relationship disappears. Model 5 implies that country-sessions following the crisis are actually more polarized, but this finding is inconsistent across models.

Table 2. Crises and polarization of congressional rhetoric

Notes: Standard errors clustered by country (1, 4), session (2, 5), or country and session (3, 6). *p < .1; **p < .05; ***p < .01.

There are a few possible critiques of this analysis. One is that the null results could be a function of a noisy dependent variable. However, there are compelling reasons to believe that this variable is capturing the partisanship of speech. First, as mentioned, the supervised learning method is tied explicitly to the speaker's party. Second, in addition to dropping commonly used words and phrases, this method drops words appearing in less than 200 unique speeches, so that idiosyncratic speech will not induce noise into the measure. Third, this approach has a good face validity. Among the most polarized country-sessions the classifiers identify are Libya (2013–2014), Iraq (2007–2008), and Guatemala (2013–2014). These reflect partisan debates about the 2012 attack on the US consulate in Benghazi, the January 2007 surge in Iraq, and the 2014 Central American migrant crisis, respectively. Moreover, analyzing the words that discriminate between Republican and Democratic speech shows they are substantive. Take, for example, the most “Republican” words about Iran in the 114th session (2015–2016): energy, billion, terrorist, Islamic, and ISIS. These words are connected to partisan debates. Billion, for instance, refers to the controversy over the USD 1.7 billion in cash the Obama administration sent to Iran in January 2016—assets and accumulated interest frozen in the United States after the 1979 Iranian hostage crisis. The move became a heated partisan issue, with congressional Republicans criticizing the Obama administration for returning the frozen assets.

Another possible critique is that the crises coded by the ICB project are not substantial enough to reduce polarization. To show that the null results are not driven by “low-threat” crises, I replicate this analysis by subsetting to “high-threat” crises, defined in three different ways in Table 3. These models replicate the analysis using only crises that escalated to military violence (models 1–3), involve the Soviet threat (models 4–6), or were coded by the ICB project as having a perceived high gravity of threat (models 7–9). The coefficients mirror those in Table 2. Their magnitude remains small: between 0.02 and 0.3 standard deviations of the dependent variable. The conclusions show that even “high-threat” crises are not consistently associated with major decreases in polarization of rhetoric around the country that triggered the crisis.

Table 3. “High-threat” crises and polarization of congressional rhetoric

Notes: Standard errors clustered by country (1, 4, 7), session (2, 5, 8), or country and session (3, 6, 9). *p < .1; **p < .05; ***p < .01.

Another potential concern is that country-sessions in the data set are not missing at random; their appearance may be correlated with the ex ante level of polarization. However, this selection process should make it easier to observe partisan convergence in response to crises. Country-sessions with few speeches (suggesting that the issues do not merit political debate) are unlikely to be associated with substantial disagreement between the parties. This analysis compares country-sessions in which there is a crisis to country-sessions in which there is some baseline level of disagreement, rather than to country-sessions without much debate. This is an easier test of the external threat hypothesis and should increase confidence in the null findings.

Another possibility is that the relationship between crises and polarization of rhetoric is changing over time. Figure 1 plots the interaction between crisis and year, including country fixed effects. We see that historically, rhetoric around a given country was less polarized in congressional sessions where a crisis with that country was taking place. This association has disappeared over time. Recently, crises have been associated with slightly more partisan divergence in rhetoric. However, this interaction effect should be interpreted with caution, since it is only statistically significant at the 90 percent confidence level and is driven by a handful of crises prior to 1940.

Figure 1. Crises and polarization of congressional rhetoric across time

Since 1940, polarization of rhetoric around foreign crises mirrors trends in overall polarization (Figure 2), which was low in the first few decades of the Cold War and increases starting in the 1970s. Rather than the absence of foreign threat eroding bipartisanship, it appears that partisan polarization driven by domestic politics has spilled into foreign policy. The results from study 1 show a weak and inconsistent association between heightened threat from international crises and partisan convergence in rhetoric around crisis countries, failing to lend robust support to the Information Hypothesis (H1A).

Figure 2. Polarization of rhetoric around international crises

Study 2: Security Crises and the Identity Mechanism

Study 2 tests the Identity Hypothesis (H2A), which predicts that external threats foster social cohesion and reduce affective polarization. I evaluate this claim by looking at how security crises affect changes in the partisan gap in presidential approval ratings, a proxy for affective polarization. Matching crisis events with biweekly Gallup polls from 1953 to 2012,⁷² I do find that security crises are associated with smaller partisan gaps in presidential approval ratings. However, this effect is short lived, providing weak support for the Identity Hypothesis. There is some evidence that “high-threat” crises have more persistent effects. But overall, domestic political variables, such as the level of congressional polarization and proximity to presidential elections, explain much more of the partisan gap in presidential approval than the threat environment.

Research Design

Assessing the Identity Hypothesis (H2A) requires a consistent, fine-grained proxy for affective polarization. Affective polarization is commonly measured by “feeling thermometers,” such as those used by the American National Election Studies (ANES) to capture attitudes toward an individual's in-party and out-party. However, the ANES are not run frequently enough to allow researchers to isolate the effects of crisis events. As an alternative, I follow other scholars in using presidential-approval data to construct measures of affective polarization.⁷³ I use a data set containing 1.8 million observations across over 1,450 nationally representative public opinion surveys run by Gallup News⁷⁴ to proxy affective polarization as the “partisan difference in presidential approval.” The advantage of this data set is that Gallup draws national samples biweekly or monthly, providing consistent observations directly before, during, and after crisis events.

This approach mirrors the literature on the “rally ’round the flag effect,” which describes how engagement in conflict increases presidential popularity.⁷⁵ But rather than using presidential approval as the dependent variable—a common approach in that literature—I calculate the partisan difference in presidential approval (Figure 3). I first average approval ratings—where Approve, Neutral, and Disapprove are coded as 1, 0, and –1, respectively—among Republicans and among Democrats in each survey (n = 1,457). The “partisan difference in presidential approval” is the average approval among respondents who share the same party as the president (“in-party”) minus the average approval among respondents from the other party (“out-party”).⁷⁶ A strong, positive correlation (r = 0.87) between the dependent variable in this study and the ANES affective polarization measure (see online supplement) increases confidence in this measure.

Figure 3. Partisan difference in presidential approval

As in study 1, I conceptualize periods of heightened threat as occurring during security crises experienced by the United States, as defined by the ICB project.⁷⁷ To compare the relative importance of threat to other macro-level variables, I aggregate data on domestic economic and political conditions. The economic conditions include the percent change in real disposable personal income per capita from the previous quarter⁷⁸ and the unemployment rate in the previous month.⁷⁹ The political variables include an indicator for whether the survey took place in an election year, an indicator for divided government, and the average level of congressional polarization in the previous session.⁸⁰

Results

In Table 4, the unit of analysis is the Gallup survey (n = 1,457). I regress the average partisan difference in presidential approval for each survey on indicators for whether the survey was run directly before, during, or directly after a crisis event. I then compare the explanatory power of these models to models containing economic variables (change in disposable income from the previous quarter and change in unemployment from the previous month) and political variables (divided government, election year, and congressional polarization).⁸¹ All models include president fixed effects with robust standard errors but are robust to other modeling choices (see online supplement).

Table 4. Crises and partisan gaps in presidential approval

Note: *p < .1; **p < .05; ***p < .01.

The Identity Hypothesis (H2A) anticipates a negative relationship between crisis events and affective polarization, which is proxied here by the partisan difference in presidential approval. The coefficient on during crisis is negative: surveys fielded during crises are associated with smaller partisan gaps in attitudes toward the president. Holding fixed the president in office, during international crises, the partisan gap in presidential approval narrows by roughly 0.15 to 0.25 standard deviations of the dependent variable. However, this relationship is short lived. By the next survey—typically only a week or two later—the effect decreases by 50 percent and is no longer statistically significant.

The models in Table 4 further show that domestic political variables explain more variation in the average partisan difference in presidential approval than the threat environment does. Standardizing the coefficients in model 6 reveals that a 1 standard deviation increase in congressional polarization corresponds to a 0.75 standard deviation increase in the partisan gap in presidential approval. Across all models, the estimated coefficient on an indicator for whether the survey is run during an election year is two to four times as large in magnitude as the coefficients on whether the survey occurred during or after an international crisis. Statistics for model selection like the AIC and BIC are significantly lower in models that contain domestic political variables, indicating that they better fit the data. In the online supplement, I further use random forests to illustrate that domestic variables are far superior in predictive power to threat-related variables.

What about the impact of “high-threat” crises? As in study 1, Table 5 replicates the analysis using three different measures of severe crises: crises that escalated to military violence (models 1 and 2), involved the Soviet threat (models 3 and 4), or were perceived to pose a grave threat ex ante (models 5 and 6). The coefficient estimates are similar in direction and magnitude to those in Table 4, but there is some evidence that “high-threat” crises may have more persistent effects. Of these three alternative measures, crises that escalated to military violence have the strongest negative association with the partisan gap in presidential approval. It remains evident, however, that domestic political variables still hold much more explanatory power relative to crisis variables.

Table 5. “High-threat” crises and partisan gaps in presidential approval

Notes: Models include controls for polls occurring directly before a crisis, which are omitted for presentation. *p < .1; **p < .05; ***p < .01.

One possible critique of these analyses is that security crises may not be exogenous to partisan gaps in presidential approval. However, security crises are more likely to be exogenous to domestic polarization than US use of military force (see the online supplement). This is one advantage of using data on crises rather than on American entry into militarized interstate disputes, which often involves a strategic choice by policymakers partly informed by domestic considerations.⁸² To further assuage this concern, the models in Tables 4 and 5 contain indicators for whether the Gallup survey was run directly before a crisis event (before crisis). The small and statistically insignificant coefficients generate confidence that crises are somewhat exogenous to the partisan difference in presidential approval.

Overall, study 2 provides weak evidence for the Identity Hypothesis (H2A) by demonstrating that periods of heightened threat from security crises are associated with smaller partisan differences in attitudes toward the president. However, in most cases, this effect is small and short-lived. While arguments that the international threat environment is a major driver of polarization are common in security studies, these analyses often do not directly compare the explanatory power of security threats to other domestic variables. Study 2 finds that the explanatory power of these claims may be limited when applied to affective polarization among the American public. Elite polarization and politicizing events like elections explain far more variation in the partisan gap in presidential approval than reactions to crisis events.

Study 3: Testing Both Mechanisms in the Context of a Rivalry

The first two studies, which observe domestic responses to security crises, fail to provide robust evidence for the external threat hypothesis. Study 3 considers an alternative conception of security threat as a more generalized sense of heightened threat from a strategic rival. A survey experiment, fielded to a nationally representative sample of 2,500 American adults, heightens a security threat to the United States from China and assesses its impact on polarization. The results show mixed evidence for the information mechanism and no evidence for the identity mechanism. First, external threats do lead to partisan convergence in attitudes toward China, as anticipated by the Information Hypothesis (H1B), but only when they are reported by nonpartisan actors. When the same threat is accompanied by a partisan elite cue, attitudes diverge. This pattern illustrates how when information about a threat is introduced in a partisan context, it can have polarizing rather than unifying effects. Second, I find little evidence that threats improve attitudes toward the out-group or reduce affective polarization as anticipated by the Identity Hypothesis (H2B).

Research Design

Although studies 1 and 2 do not find significant support for the external threat hypothesis, they have important limitations. First, they are unable to directly measure the outcomes of interest—partisan convergence in attitudes toward a potential adversary (study 1) and affective polarization (study 2)—relying instead on proxy measures. Second, the concept of threat in these studies is operationalized as a crisis event. However, some variants of the external threat hypothesis stress that a state's threat environment is shaped by strategic rivals.⁸³ Third, while the first two studies find that threats may not inherently reduce domestic polarization, they cannot speak to where the theoretical logic behind the external threat hypothesis breaks down.

To address these limitations, in study 3, I field a survey experiment to 2,500 American adults that heightens a security threat from a rising China. The primary objective of the experiment is to explore whether having respondents read a threat report about China induces partisan convergence in attitudes toward China (information mechanism) or reduces affective polarization (identity mechanism). A secondary objective is to probe the underlying logic of these mechanisms. To do so, I vary the characteristics of the threat report to explore conditions under which threats are more likely to reduce polarization.

I focus on China because it is a “most likely” case for the external threat hypothesis in contemporary US politics. Experts believe China has the potential to pose not only economic and military but also ideological threats to the United States.⁸⁴ The rise of China also has been explicitly linked to the external threat hypothesis in scholarly and popular discourse.⁸⁵

I fielded the survey experiment in June 2019 via the Lucid Academic Marketplace.⁸⁶ Lucid maintains a marketplace to recruit survey respondents, targeting demographic quotas based on age, sex, ethnicity, race, and region in each sample to mirror the population of American adults.⁸⁷ Participants were randomly assigned to one of five conditions, represented by the dark-gray boxes in Figure 4. Twenty percent were assigned to the control group (“No threat”) and received no prime. The other 80 percent read a vignette based on the Worldwide Threat Assessment of the US Intelligence Community, presented annually to Congress by the director of national intelligence.⁸⁸ The vignette provides real information on the traditional and nontraditional security threats China poses to the United States. If the external threat hypothesis is true, respondents who read about the China threat should exhibit less polarization relative to the control group.

Figure 4. Survey treatment conditions

To explore the underlying logic of the external threat hypothesis, I also vary the vignette in two ways to see whether there are conditions under which threats reduce polarization. First, to probe the Information Hypothesis (H1B), I randomize the source cue that accompanies the report. The intuition is that the Information Hypothesis should be the most likely to hold when information about the threat is communicated by an actor who is explicitly nonpartisan. By contrast, in a polarized environment, if information about a threat is communicated by a partisan actor, the out-party will not view the source as credible. To test this, I randomize the source cue in the experiment and pair it with a visual cue. For half of the respondents, the report is attributed to “nonpartisan experts,” and they see a picture of the actual report cover of the Worldwide Threat Assessment. For the other half, the report is attributed to the Trump administration, and they see an image of President Trump.⁸⁹

Second, to probe the logic of the Identity Hypothesis (H2B), I randomize the frame of the report: whether or not the security threat is described as threatening American identity and values. The intuition is that the Identity Hypothesis should be most likely to hold when threats have an ideological component. Threats that are explicitly anti-American should be most likely to create an in-group/out-group distinction, heightening national identity and reducing affective polarization. In the ideological frame, the additional text (also adapted from the Worldwide Threat Assessment) emphasizes how China's intentions contrast with American values like the promotion of democracy and human rights. In the nonideological frame, this text is excluded. The vignette reads:

A recent report from [CUE: nonpartisan experts / the Trump administration] says that the risk of conflict between the United States and China is higher than any time since the end of the Cold War. According to the report, [CUE: experts / President Trump and his cabinet officials] say that:

• • China is aggressively expanding its economic and military influence, as well as its nuclear capabilities.

• • China is using intelligence services to steal information and spy on US citizens.

• • China has the ability to launch cyber attacks that can disrupt critical infrastructure—such as electric grids or natural gas pipelines—in the United States.

• • [FRAME: NULL / As the Chinese Communist Party continues to expand its power, the report predicts that a coming “ideological battle” between the United States and China will threaten support for democracy and human rights globally.]

To reinforce the threat prime, respondents reflect on its findings in an open-ended response:

In a few sentences, tell us what you think about this report. A copy of the report is below for your reference.

Participants then answer questions about their partisan identity and attitudes toward China. To evaluate the Information Hypothesis (H1B), participants are asked about their threat perceptions of China and appropriate policy responses. To measure threat perceptions, they report the extent to which they agree or disagree with the statement: “China poses a threat to the United States.” The response options range from “strongly disagree” (1) to “strongly agree” (7) on a Likert scale. To measure attitudes toward policy responses, respondents are asked how acceptable or unacceptable it is for the United States to use different instruments of foreign policy to respond to China. The main results focus on how acceptable it is for the United States to “use military force against China.”⁹⁰ The outcomes are measured on a Likert scale from “very unacceptable” (1) to “very acceptable” (5). If priming external threat makes Democrats and Republicans converge in their attitudes toward China, this would substantiate the Information Hypothesis (H1B).

To evaluate the Identity Hypothesis (H2B), respondents use “feeling thermometers” to characterize attitudes toward members of their political out-party and in-party. Participants rate their favorability toward “Democrats” and “Republicans” on a scale from from 0 to 100. The difference between the rating of a respondent's in-party and out-party is a proxy for affective polarization.⁹¹ If respondents who receive the threat primes exhibit lower levels of affective polarization relative to the control group, this would substantiate the Identity Hypothesis (H2B). If threat primes that also contain an ideological frame reduce affective polarization more than primes with no ideological frame, this would lend further support to the Identity Hypothesis.

Results

To test the Information Hypothesis (H1B), I consider whether reading the threat report caused partisan convergence in attitudes toward China. To measure threat perceptions, the survey asks whether respondents agree or disagree with the statement, “China poses a threat to the United States,” recording answers on a seven-point Likert scale.⁹² Consistent with the Information Hypothesis (H1B), threat primes increase the perception that China poses a threat to the United States overall. However, the primes have different impacts on the polarization of attitudes.

When the threat is communicated by nonpartisan experts, Republicans and Democrats converge in their perceptions of China (Figure 5).⁹³ The partisan difference in threat perceptions is about three times larger in the control group than in the nonpartisan condition. However, communicating the same threat but attributing it to President Trump makes Democrats and Republicans diverge in their perceptions of China. Relative to the nonpartisan condition, the partisan condition increases perceptions of threat among Republicans but has no effect on perceptions of threat among Democrats. In the partisan condition, the partisan difference in threat perceptions is six times as large as in the nonpartisan condition and twice as large as in the control group.⁹⁴

Figure 5. Does China pose a threat to the United States?

To measure partisan convergence in attitudes toward policy responses, the survey asks respondents how acceptable or unacceptable it is for the United States to use military force against China. Figure 6 shows the percentage of Democrats and Republicans who believe it is acceptable to use military force in each of the treatment conditions. Reading the threat report increases support for the use of force. However, when the same threat report is accompanied by a partisan cue, attitudes diverge along party lines. Under the partisan condition, the gap increases by about nine percentage points, roughly doubling in size.

Figure 6. Is it acceptable for the United States to use military force against China?

The results show mixed support for an information mechanism. A threat introduced by nonpartisan experts causes convergence in threat perceptions of China but not in policy responses. Most importantly, when the same threat is introduced by President Trump, both attitudes toward China further polarize. These findings provide insight into the results of study 1, which found no association between crises and partisan convergence of rhetoric around the country that initiated it. Recall that study 1 also showed that historically, responses to crises were less polarized. One explanation for the study 1 results that is consistent with study 3's experimental findings is that threats introduced in nonpartisan contexts can cause convergence in attitudes, but when information about threats enters a polarized domestic context and filters through a partisan media environment, the external threat hypothesis is unlikely to hold. Study 3 illustrates that in a hyperpartisan environment, a simple partisan cue caused many out-party respondents to reject information about a China threat. In such an environment, we should anticipate that political actors will have difficulty conveying information about threats in a nonpartisan manner.

Next, I evaluate the Identity Hypothesis (H2B), which anticipates that external threat primes reduce affective polarization. As a corollary, we might expect threats framed in an explicitly ideological way to reduce affective polarization more than security threats alone. In Table 6, the dependent variable, affective polarization, is measured as the difference in “feeling thermometer” ratings (0 to 100 degrees) between the respondents’ in-party and out-party. Models with demographic controls include covariates for political party, sex, age, race, higher education, and region.

Table 6. Do external threats reduce affective polarization?

Note: *p < .1; **p < .05; ***p <.01.

The results in Table 6 show that priming threat decreases the difference between in-party and out-party ratings by approximately one to three percentage points on a 100-point scale relative to the control group. This effect is small and not statistically significant. Even respondents who read threat reports with an ideological frame that emphasized the threat that China poses to American values do not exhibit significantly less affective polarization (models 3 and 4).⁹⁵

In sum, study 3 provides mixed evidence for the Information Hypothesis (H1B) and no evidence for the Identity Hypothesis (H2B).⁹⁶ Heightening the salience of a security threat to the United States from China did not affect the level of affective polarization in the sample, even when the report was framed in an explicitly ideological way. However, priming threat did lead to convergence in attitudes toward China across party lines when information was delivered by nonpartisan experts. In contrast, when the report was attributed to President Trump, Republicans perceived China as even more threatening, and Democrats did not update their attitudes relative to the control condition. As a result, attitudes toward China diverged further along partisan lines.

Discussion

The three studies reported here demonstrate that despite the prominence of the external threat hypothesis, we lack strong evidence that foreign threats explain much variation in domestic polarization. While each study has important limitations, they collectively move research on external threat and domestic polarization beyond the historical period to different levels of analysis. A critic of this research may argue that no threat to the United States has matched or will match the Soviet threat. Yet even if we concede that the Soviet threat did induce bipartisanship in foreign affairs,⁹⁷ reliance on this single case illustrates why the external threat hypothesis is so sticky in the international relations literature. That is, the Soviet case is used both to generate and to test the theory, rendering it unfalsifiable. But if the external threat hypothesis applies to only one case, it is much more limited in its explanatory power than existing research suggests.

Perhaps the closest analogy to the Soviet threat in the twenty-first century is the threat that emerged from terrorist attacks in the United States on 11 September 2001. One could imagine a counterfactual in which 9/11 ushered in decades of bipartisan cooperation over counter-terrorism. Instead, the bipartisan aftermath of the attacks and the reduction in affective polarization that accompanied it quickly unraveled. Within two years, both the partisanship of congressional rhetoric around foreign adversaries and the partisan gap in presidential approval reached their greatest extent to date. Arguably, the divisiveness of the years that followed had less to do with the nature of the terrorist threat and more to do with the institutional changes and demographic trends that continued to polarize American politics.

A second question that follows naturally from this project is: If the international threat environment is not a major driver of polarization in American foreign policy, what is? While this paper cannot directly speak to that question, the evidence is more consistent with the idea that polarization over domestic policy has spilled into foreign affairs. This logic is related to the concept of “conflict extension” in American politics,⁹⁸ a process in which issue areas that cut across party lines are gradually absorbed into the partisan divide by party activists. While recent work maps the extension of party conflict into a range of social issues,⁹⁹ future research could map this process in greater detail in foreign policy.

If domestic partisan conflict is extending to foreign affairs, national security threats should not independently cause convergence in political attitudes or reduce social cohesion. Rather, reactions to external threat should be correlated with ex ante levels of domestic polarization. The results from all three studies more closely align with these expectations than the predictions outlined by the external threat hypothesis. Study 1 showed a small and inconsistent negative association between security crises and partisan convergence in congressional rhetoric. Instead, the partisanship of speech in relation to crisis events has steadily increased since the 1970s, which reflects trends observed in domestic polarization. In study 2, while crises were associated with a brief narrowing of the partisan gap in presidential approval ratings, variation in this outcome was better explained by congressional polarization and proximity to politicizing events like elections. In study 3, partisan convergence or divergence in attitudes toward China was conditional on whether the threat report was accompanied by a nonpartisan or partisan cue.

To clarify, the results of this research do not suggest that external threats will never reduce polarization. But they do show that in the contemporary political climate, new security threats are unlikely to be unifying. As study 3 finds, in a highly polarized environment, it becomes difficult for political actors to credibly communicate information about threats in a nonpartisan manner. Presidents are seen as inherently political actors, and members of the political opposition have incentives to criticize or actively oppose them rather than to simply defer to their judgment. Recent scholarship in American politics finds that policies championed by the president are more sharply polarized.¹⁰⁰ This research shows that national security threats are not immune to such dynamics. Instead, we should anticipate debates about new, salient threats to be divisive.

Extensions

There are many possible extensions to this research on threat and domestic polarization. One would be to distinguish between different types of threats, such as economic shocks, humanitarian disasters, or public health crises. Albertson and Gadarian distinguish between unframed threats, which require no political messaging to instill fear (e.g., spread of infectious diseases), and framed threats (e.g., security of national borders), which are easier to filter through partisan lenses.¹⁰¹ In theory, the former should be more likely to reduce polarization than the latter. However, the American response to the threat of the coronavirus disease (COVID-19) in 2020 demonstrates that major threats to public health are still susceptible to politicization. The partisan differences in behavior, threat perception, and policy responses to the pandemic illustrate how information about a relatively unambiguous threat is easily politicized in a highly polarized environment.¹⁰²

A second extension is to consider how threats with different distributional consequences may lead to greater polarization. For instance, economic shocks from a rising China that disproportionately affect different areas of the United States might lead to divergent threat perceptions of China. Likewise, foreign adversaries that pose a greater threat to one party are likely to be divisive. For example, Russian interference in the 2016 US presidential election was widely perceived to advantage the Trump campaign, leading Democrats to view Russia as a much greater threat relative to their Republican counterparts' view.¹⁰³

Another next step is to explore variation across comparative contexts. The findings in this paper speak to the United States, which is unique in its political system, geographic location, size, and economic power. The external threat hypothesis may be more likely to hold in states that are small or bordered by adversarial powers. For instance, Gibler traces how territorial threats from Turkey led to political centralization in Greece, and Sekulic explains how the removal of the Soviet threat precipitated Yugoslavia's dissolution.¹⁰⁴ Another characteristic that may moderate the relationship between threat and polarization is the nature and alignment of political cleavages. In places where foreign policy is orthogonal to a central partisan cleavage, foreign threats may be more likely to reduce polarization. By contrast, in places where foreign policy is central to partisan cleavages—such as in South Korea and Israel—periods of heightened threat could be even more likely to sow division. These and related questions could probe the conditions under which the external threat hypothesis holds outside of the American context.

Conclusion

The narrative that external threats foster bipartisan consensus, and domestic politics polarizes in their absence, permeates academic and political discourse. Yet the implications of this argument—that crises and other environments of high threat are “good” for domestic politics—are troubling. This argument is prominent because it is intuitively appealing and difficult to falsify. Many evaluations of the external threat narrative operationalize the threat environment of the United States as congruent with different eras of US foreign policy, leaving us with few observations to evaluate whether the hypothesis is true. The claim that American politics was less polarized during the Cold War than it is today is hard to dispute; however, this assertion alone is insufficient to link foreign threats to domestic polarization.

This paper describes two mechanisms through which foreign threats might create partisan unity. Through the information mechanism, threatening actions reveal information that results in partisan convergence of attitudes toward an adversary. Threats credibly signal the gravity of a foreign policy situation, creating a shared objective among policymakers, which results in the political opposition deferring to the executive. Through the identity mechanism, foreign threats foster social cohesion by heightening the salience of national identity relative to partisan identity. This increases favorability toward the out-party, reducing affective polarization. I tested these mechanisms in three studies, using historical, observational data to evaluate responses to security crises (study 1 and 2), and using experimental data to evaluate the effects of heightened threat from a rival foreign power (study 3).

To evaluate the information mechanism, I explored whether threats generated by security crises caused partisan convergence in Congress. I used supervised machine learning to measure the polarization of congressional rhetoric around foreign adversaries. I then assessed whether there was less partisan rhetoric about an adversary in congressional sessions in which that adversary triggered a crisis for the United States. The analysis showed no systematic association between heightened threat and polarization of rhetoric. Instead, legislators’ responses to threat reflected the relative partisanship of the political environment into which new crises were introduced.

To evaluate the identity mechanism, I looked at how security crises affected affective polarization among the American public. I used the partisan gap in presidential approval to proxy for affective polarization. Drawing on historic polling data, I found that crises were associated with a smaller partisan difference in presidential approval ratings, but these effects were small and short-lived. Polarization among political officials and politicizing events like elections explain the variation in partisan gaps in presidential approval better than the international threat environment does.

I next considered an alternative concept of a security threat by heightening a generalized sense of threat from China in a survey experiment. Twenty-five hundred American adults read a report about major security threats that the United States faces from China. The report varied in two ways: the source of the report and the frame of the report. In half of the cases, the report was attributed to “nonpartisan experts”; in the other half, to the Trump administration. The frame of the report also varied: in half of the reports, the security threat was framed as an ideological threat to American values. Through the information mechanism, priming threat should cause partisan convergence in perceptions of an adversary and policy responses. Through the identity mechanism, priming threat should reduce affective polarization, and these effects should be larger in response to foreign threats that explicitly challenge American values. The survey results showed no support for the identity mechanism but mixed support for the information mechanism. The external threat hypothesis was most likely to hold when the threat report was accompanied by a nonpartisan cue. By contrast, when the report was attributed to President Trump, attitudes further polarized. These results suggest that when information about a security threat enters into a politicized context, we should not anticipate the threat to inherently reduce polarization.

What can we conclude about the external threat hypothesis? Taken collectively, these studies find that it is unlikely that partisan polarization over US foreign policy or affective polarization among the American public has been substantially shaped by America's threat environment. It is much more likely that the demographic and institutional changes that precipitated partisan polarization over domestic policy have spilled over into foreign affairs. The primary takeaway is that we have little evidence that foreign threats systematically reduce domestic polarization in the American context. This is not to say that security threats will never generate partisan unity, rather that we should expect external threats with substantive, long-lasting effects on polarization to be very rare. Claims that crisis events or emergent threats from rival powers will automatically bind the United States together should be met with a healthy dose of skepticism.