Do external threats increase bipartisanship in the United States? An experimental test in the shadow of China's rise

Abstract

Do external threats increase American bipartisanship? We subject this question to an experimental test. Leveraging the Biden and Trump administrations’ similar characterization of the China threat, we exposed American respondents to real-world primes about security threats from China, while randomizing the messenger of such primes. We find that the threat primes—regardless of the partisan identity of their messenger—boosted Democrats’ and Republicans’ support for assertive foreign policy in a largely parallel manner, thereby failing to reduce preference polarization. Importantly, there were no measurable changes across multiple indicators of affective polarization. These findings clarify the limits of external threats in uniting Americans, while also challenging recent perspectives that external threats—often colored by elite rhetoric—will further polarize the American public.

Do external threats unite or divide Americans? Early influential scholarship argues that security threats posed by adversaries can induce bipartisanship in the United States (Ostrom and Job, 1986; Marra et al., 1990; Brody, 1991; Gowa, 1998). Recent scholarship, however, suggests that external threats are unlikely to unify Americans in the shadow of hyperpartisanship (Myrick, 2021). Because political communication about external threats is often accompanied by partisan elite cues (Berinsky, 2009; Lenz, 2012), external threats may even further polarize the American public (Nicholson, 2012; Levendusky, 2013; Guisinger and Saunders, 2017; Barber and Pope, 2019; Merkley and Stecula, 2021).

In this paper, we contribute to this timely and ongoing debate by experimentally testing the effects of external threats on preference and affective polarization among US citizens. Exploiting the similar characterization of the China threat between the Biden and Trump administrations in recent years, we exposed American respondents to real-world primes about security threats from China, while randomizing the messenger of such primes. We then measured the downstream effects on a wide range of foreign policy preferences and a set of validated measures of affective polarization. Our research design thus conceptually replicates and innovatively builds on Myrick (2021). While Myrick's experiment examined the impact of Trump's threat prime on partisans’ political attitudes, we added a parallel threat prime from Biden as a new experimental condition. By building on an existing experimental protocol, we track whether and how partisan cues about external threats shape preference and affective polarization amid hyperpartisanship.

Our experimental innovation is theoretically motivated: in political communication of external threats, Biden's deviation from a dovish foreign policy stance typically held by the Democratic Party generates a more costly—and therefore more credible—signal to the public by “going against type” (Trager and Vavreck, 2011; Kreps et al., 2018; Mattes and Weeks, 2019; cf. Kertzer et al., 2021). Such deviation thus has the potential to induce attitudinal change toward China not only among Democrats but also among Republicans, conducive to a “parallel shift” in their preferences for more assertive foreign policy against China (Coppock, 2023). The current political environment provides another theoretical motivation of our replication. Since Myrick's contribution, Democrats and Republicans have gradually reached a consensus on the China threat.¹ Given this unique context, out-party communication of the external threat could activate partisan identities less, but instead boost both Democrats’ and Republicans’ support for taking a more assertive stance on China. Thus, the polarizing effect of out-party cues on foreign policy preferences—as documented by previous research—may not hold in the current political climate. Replicating Myrick's experiment in this special context would allow us to accumulate and progress scientific knowledge by elucidating the scope conditions of existing empirical findings and refining theoretical arguments (Jasny et al., 2011; Maxwell et al., 2015).

We find that threat primes about China increased Democrats’ and Republicans’ foreign policy assertiveness in a largely parallel manner. Contrary to prominent scholarship underscoring the role of elite cues in polarizing the public (e.g., Nicholson, 2012; Satherley et al., 2018; Merkley and Stecula, 2021), the parallel shifts in foreign policy preferences took place regardless of the messenger. This finding contrasts with Myrick (2021), who offered suggestive evidence that foreign policy preferences among Democrats and Republicans further polarized when they were subjected to Trump's communication about the China threat. Consistent with her key finding, however, we find no statistically significant decreases across all four preregistered measures of affective polarization. With a well-powered survey experiment that adopts a multidimensional approach to measuring preference and affective polarization (see Section 2), our conceptual replication not only reinforces Myrick's influential but preliminary finding that external threats cannot unite Americans amid hyperpartisanship (cf. Schwartz and Tierney, 2024), but also offers first-cut experimental evidence to challenge the pessimistic yet increasingly dominant view that external threats will further polarize the American public.

1. External threats, partisan cues, and polarization

Early scholarship in international relations (IR) argues that heightened security threats posed by adversaries can make “politics stop at the water's edge.” Drawing on qualitative case studies, an influential literature asserts that external threats can unite Americans (Desch, 1998; Gowa, 1998). Analyzing how America's polarization historically corresponded to its shifts in power, scholars “predict that as China becomes a peer competitor polarization will gradually diminish […] and American foreign policy will regain its luster” (Bafumi and Parent, 2012, 27). Additional research argues that external threats can unite Americans by boosting social cohesion (Ostrom and Job, 1986; Marra et al., 1990). This is because international crises can reinforce national identity and promote cross-party unity (Mueller, 1970; Brody, 1991; cf. Seo and Horiuchi, 2024), conducive to ameliorating polarization (Levendusky, 2023).

Much of the American politics literature, however, would suggest that inducing bipartisanship is less likely in a highly polarized environment. Because elite communication is often filtered through partisan lens (Berinsky, 2009; Barber and Pope, 2019), external threats may not be able to unify Americans. In many cases, Democrats and Republicans simply “follow their leaders” by taking their cues and adopting their policy positions (Lenz, 2012), including foreign policy issues (Guisinger and Saunders, 2017). Due to hyperpartisanship, out-party cues may even backfire, not only failing to persuade partisans from the other side but also pushing strongly attached partisans to the other extreme (Nicholson, 2012; Levendusky, 2013; Satherley et al., 2018; Merkley and Stecula, 2021). The result is further polarization.²

While scholars continue to disagree on whether external threats will unite or divide Americans amid hyperpartisanship, empirical investigations that help to adjudicate the debate are scarce. One notable exception is Myrick (2021), who conducted an experiment in 2019 where Trump's political communication about the China threat dominated the media environment.³ Myrick randomly assigned respondents to read a short summary of the Worldwide Threat Assessment of the US Intelligence Community, while also randomizing whether the source cue that accompanied the summary was from nonpartisan experts or from the Trump administration. Measuring partisans’ posttreatment preferences for US foreign policy responses to China and feeling thermometer scores for in- and out-partisans, Myrick (2021) finds that “when [information about a foreign threat] is accompanied by a partisan elite cue, attitudes are further polarized” (923). Based on this finding, she concludes that “new threats introduced in already polarized contexts are likely to sow greater division” and that “attempts by American political officials to amplify threats could backfire” (923).

While Myrick's experiment makes an important contribution, it remains ambiguous whether and how the results would travel in the current context where the Democratic leader is the dominant messenger of the China threat. The ambiguity is further amplified when the Democratic Party, unlike the Republican Party, is typically associated with dovishness rather than hawkishness (cf. Kertzer et al., 2021). When Biden becomes the messenger of the China threat, he sends a clearer signal to the public by going against type (Trager and Vavreck, 2011; Kreps et al., 2018; Mattes and Weeks, 2019). Because such against-type signals are more costly, they are also more credible—not only to Democrats but also to Republicans. In turn, partisan preferences for assertive foreign policy against China may shift in parallel (Coppock, 2023). Although this theoretical expectation diverges from the prediction from Myrick (2021) and the “partisan intoxication” hypothesis dominant in American political behavior scholarship,⁴ it accords with the latest work showing that partisanship does not distort information processing of political messages (Tappin et al., 2023). Against the backdrop of polarization, partisans can update their prior beliefs amid disagreement (Fowler and Howell, 2023). How elite cues about external threats shape bipartisanship, however, remains an empirical question in IR.

2. Experimental design

To investigate whether external threats induce bipartisanship and how partisanship may complicate this process in the contemporary political climate, we conducted a preregistered survey experiment on American adults through PureSpectrum in September 2023, using quota sampling to match census benchmarks in sex, age, and race. We provide our preanalysis plan in Appendix I, where we also document our consistencies with and deviations from the preregistration. Because our experiment focuses on political polarization, in what follows we only analyze respondents who identified as Democrats or Republicans in pretreatment questions ($N = 3335$). Appendix A provides information about the sample and recruitment procedure.

Using simple randomization, we assigned respondents to one of the four experimental conditions: (1) no prime (control group), (2) Biden threat prime, (3) Trump threat prime, and (4) nonpartisan threat prime. The treatment groups then read the following text, which distilled the major insights from real-world intelligence reports (Myrick, 2021):

Intelligence reports from [the Biden administration/the Trump administration/nonpartisan experts] warned 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 these reports, [President Biden and his cabinet officials/President Trump and his cabinet officials/non-partisan experts] said that:

• • China is aggressively pursuing its goal of building a world-class military that will enable it to project power globally and offset U.S. military superiority.

• • China is using intelligence services to steal information and has become the top threat to U.S. technological competitiveness.

• • China is actively using subsidies and trade policy to give its firms a competitive advantage.

Subsequently, we asked respondents to write down what they thought about the report in order to reinforce the threat prime. Respondents in the control group, on the other hand, were not given this report and were not asked to complete this task. The procedure thus replicates Myrick (2021), allowing us to connect our experimental test to the existing empirical literature.

To examine the treatment effects on preference polarization, we measured individual preferences for a set of US foreign policies on a 5-point Likert scale. These policies include expanding US military spending, restricting the exchange of scientific research between the United States and China, and reducing international trade with China—policies that already saw preference polarization between Democrats and Republicans, such that Republicans’ approval was higher according to nationally-representative surveys (Chicago Council on Global Affairs, 2021). We also measured preferences for offering financial support to US companies that are in strategic industries, whose approval from Republicans was not higher than that from Democrats (Chicago Council on Global Affairs, 2021)—thus allowing us to expand our empirical scope about the treatment effects on preference polarization. For affective polarization, we measured the difference between in-party and out-party feeling thermometer scores, evaluation of out-partisans’ positive and negative traits, trust toward the out-party to do what is right for the country, and the level of social distance from out-partisans (Druckman and Levendusky, 2019). Following Myrick (2021), we also asked respondents how much they agreed that “China poses a threat to the United States” to gauge another source of attitude polarization between partisans. Our survey instrument is available in Appendix B.1.

Two design features are useful to note. First, our treatments are factually grounded. The annual threat assessments of the (nonpartisan) US Intelligence Community contain those messages about China under both the Biden and Trump administrations. Thus, our experiment is a conceptual replication of Myrick (2021) that draws on the same real-world materials to develop the threat primes. Our main innovation is the addition of Biden's cue that is both factual and unique to the current political context, thereby allowing us to investigate the public opinion impact of a more complete set of partisan cues while holding the threat primes constant.

Second, our measurement approach is direct and thorough. The set of foreign policies we examine directly maps onto the areas in which China is posing a threat to the United States according to the threat prime. The measures of affective polarization, on the other hand, are all validated and widely used in the American politics literature (Levendusky, 2023) and extend beyond the conventional feeling thermometer item that Myrick's experiment solely relied on. This measurement feature is crucial because existing scholarship has shown that the other three measures of affective polarization adopted in our experiment are conceptually and empirically distinct from the feeling thermometer item (Druckman and Levendusky, 2019). Incorporating additional measures thus allows us to investigate the treatment effects on affective polarization across multiple dimensions, compared to an alternative approach that uses the feeling thermometer item to capture only one specific dimension of affective polarization. In Appendix B.2, we further compare and contrast our experimental design with Myrick (2021), and explain the logic and merit of each deviation.

3. Results

While respondents did not significantly increase their threat perceptions of China upon exposure to threat primes,⁵ they reported higher support for assertive US foreign policy, regardless of the messenger (Figure 1). The increases in support were predominantly bipartisan—even in the presence of out-party elite cues. To illustrate this point, Figure 2 shows the conditional average treatment effects (CATEs) on Democrats and Republicans. If elite cues had a backlash effect, the CATEs on Democrats in the Trump condition and on Republicans in the Biden condition would both be negative. However, as Figure 2 shows, none of the 24 point estimates of CATEs are negative; all of them are positive regardless of whether the threat prime was delivered by an in-party, out-party, or nonpartisan administration.

Figure 1. Average treatment effects on preferences for US foreign policy. Note: Plotted point estimates indicate the average treatment effects on foreign policy assertiveness, i.e., mean differences in preferences for assertive US foreign policy between the treatment and control groups. Higher values indicate stronger preferences for the assertive foreign policy. Error bars represent 95 percent confidence intervals.

Figure 2. CATEs and differences-in-CATEs for democrats’ and republicans’ US foreign policy preferences. Note: CATEs refer to the average treatment effects conditional on a respondent characteristic, which is partisanship in this case. Table S3 shows the regression estimates. Figures S2–S5 show the control mean and distribution of responses. Error bars represent 95 percent confidence intervals.

To unpack the impact of partisan threat primes on preference polarization, Figures S2–S5 in Appendix D show the distribution of foreign policy preferences by partisanship and experimental condition. In the control condition, there were gaps between Democrats’ and Republicans’ preferences for expanding military spending, restricting scientific exchange, and reducing bilateral trade, such that Republicans had higher baseline levels of support for these assertive foreign policies against China (i.e., indication of preference polarization with Republicans being more hawkish). However, none of the treatment conditions appeared to meaningfully widen these pre-existing gaps. If elite cues had further polarized partisans’ preferences, the difference-in-CATEs estimates reported in Figure 2 (i.e., CATE on Democrats minus CATE on Republicans) would have been negative for this set of foreign policies.⁶ Against this empirical implication of heightened polarization, none of the 12 difference-in-CATEs estimates were negative with statistical significance at our preregistered threshold of p = 0.05. While there is some indication that Trump's threat prime widened the gap between Democrats’ and Republicans’ military spending preferences ($\hat {\beta } = -0.20$, SE = 0.12, p = 0.07), the effect size is relatively small given the 5-point range of our dependent variables. The only statistically significant, but positive, difference-in-CATEs estimate is the impact of Biden's threat prime on Democrats’ and Republicans’ preferences for restricting scientific exchange with China (second panel in Figure 2; $\hat {\beta } = 0.24$, SE = 0.11, p = 0.03). Because Democrats’ baseline support for this policy was lower than Republicans’ baseline support, the positive difference-in-CATEs estimate indicates not only that Biden's threat prime was more effective in boosting Democrats’ support for the policy but also that Biden's cue converged policy preferences between Democrats and Republicans (Figure S3).

Overall, it is notable that none of the partisan gaps in policy preferences were widened across all foreign policies examined in our experiment. All experimental stimuli increased Democrats’ and Republicans’ preferences for assertive US foreign policy, and randomized partisan cues—be they communicated by in-party, out-party, or nonpartisan elites—neither converged nor diverged these preferences, with only one exception that suggests convergence (i.e., preference depolarization) rather than divergence.⁷

These results contrast with the existing literature's finding that Democrats’ and Republicans’ policy preferences would further polarize if they were subjected to out-party communication of the China threat (Myrick, 2021).⁸ We believe this is due to the unique time context of our experiment, where a bipartisan consensus about the China threat had largely been formed in the United States.⁹ In Appendix F, we compare the data from our control group with Myrick's control group data, showing that our 2023 survey saw escalated perceptions of the China threat among both Democrats and Republicans vis-à-vis Myrick's 2019 survey. With such shared beliefs about China, most respondents would likely find information about the China threat credible, regardless of the information source. Because there is not much partisan debate over whether China is a threat, anti-China information from even a partisan source would not necessarily activate respondents’ partisan identities. The timely nature of our survey, therefore, might explain why support for a more assertive position against China increased in parallel among Democrats and Republicans even when they were exposed to one-sided partisan cues about the China threat, rather than diverging as Myrick's experiment suggested.

What about affective polarization? Across all treatment conditions, there were no measurable decreases in all indicators of affective polarization (Figure 3). Although previous research suggests that external threats can make politics stop at the water's edge by promoting national identity (Mueller, 1970) and that priming national identity is conducive to ameliorating affective polarization (Levendusky, 2023), the data at hand do not offer empirical evidence that external threats can depolarize Americans on the affective dimension—consistent with Myrick's finding that hinges on one specific measure and dimension of affective polarization (i.e., in-party and out-party feeling thermometer scores). Thus, not only do we successfully replicate Myrick's important finding, but we also use an additional elite cue, an improved measurement approach, and a well-powered experiment to reinforce the finding in the contemporary political context.

Figure 3. Average treatment effects on affective polarization. Note: Plotted point estimates indicate the average treatment effects on affective polarization, i.e., mean differences in indicators of negative out-party affect between the treatment and control groups. All variables are rescaled to take the range from 0 to 1, with higher values indicating higher levels of negative out-party affect. Error bars represent 95 percent confidence intervals.

In Appendix E, we extend our analysis to Independents and show that the threat primes did not appear to improve Independents’ affect toward Democrats or Republicans. We also explore heterogeneous treatment effects on foreign policy preferences by nationalism and cooperative internationalism—two important predictors of hawkish attitudes in IR scholarship—in Appendix D. To probe the mechanism behind our results, we analyze the open-ended responses to trace how respondents from across the political spectrum reacted to our threat primes. Using structural topic modeling (STM) to explore how respondents’ language in the writing task varied across experimental conditions and partisan orientations (Roberts et al., 2014), we show that how Democrats and Republicans reacted to the threat primes was largely unaffected by their messenger (Appendix H).¹⁰ Reading through the open-ended responses, we find that some respondents, although predisposed to having low trust with the Biden or Trump administration due to their partisan identity, were willing to believe the content of the out-party threat prime. For example, one self-identified strong Democrat wrote: “I don't trust much coming from the Trump administration, but I do find this report credible.” On the other side, a strong Republican commented on the Biden threat prime: “I think it is a (surprisingly) honest report. China is a threat to the nation at the moment as the sitting president has no interest in protecting the nation from foreign wars.”

Collectively, our findings cut against much speculation in the existing literature that partisan cues disseminated in a polarized environment would distort information processing among American partisans. Scholars adhering to the view of “partisan intoxication”—or Americans’ strong psychological attachments to partisanship such that “if they say ‘yes,’ we say ‘no”’ (Satherley et al., 2018; see also Nicholson, 2012; Merkley and Stecula, 2021)—would expect Democrats/Republicans to blindly accept Biden's/Trump's cue and reject Trump's/Biden's cue, regardless of their preexisting views and the opponent party's views. However, we find evidence contrary to this perspective.

How externally valid are our findings? While our study was conducted amid hyperpartisanship, we examined an issue area where a bipartisan consensus had largely been formed. The prevailing “China consensus” might be conducive to suppressing potential backlash, as partisans could still agree with belief-affirming messages disseminated by out-party elites. This dynamic suggests that our findings may be less applicable to country and time contexts where domestic consensus about the external threat has not yet been formed. If our speculation is correct, then similar elite cues about the Russia threat in the current time context should also be limited in shaping preference and affective polarization in the United States,¹¹ yet elite cues about the China threat may generate different effects on Taiwan's polarization due to the different understandings of the threat, in the first place, among supporters of Kuomintang and the Democratic Progressive Party (see Carothers, 2023).

Our findings are also limited in elucidating the impact of other forms of threat primes that are more vivid. While our experimental treatments were factually grounded such that they drew directly from security reports, real-world elite rhetoric on external threats is often less sterile but more emotionally charged. Using a vivid threat prime by “present[ing] respondents with a significantly more detailed, concrete, emotionally-engaging, and dramatic threat condition that also involves images as well as text,” Schwartz and Tierney (2024) find that it can reduce affective polarization when combined with “information that there is consensus among both Democratic and Republican policy elites about the severity of the Chinese threat” (11–12). Future research could build on our experimental protocol, as well as the vivid threat prime developed by Schwartz and Tierney, to unpack how vividness and elite cues may interact to shape American polarization.

4. Conclusion

Caveats notwithstanding, our paper makes two general contributions. First, it challenges the pessimistic view that partisan communication of external threats will further polarize the American public in a hyperpartisan context. Across all treatment conditions with varying partisan cues, as well as all measures of preference and affective polarization that our experiment comprehensively covered, we find no evidence that political communication of external threats diverged Americans’ foreign policy preferences or that it worsened Democrats’ and Republicans’ affect toward each other. These findings help to adjudicate ongoing debates about the impact of external threats on American polarization, especially in the shadow of China's rise (Brooks, 2019; Myrick, 2021; Friedrichs and Tama, 2024), suggesting that external threats would unlikely create further divisions among Americans. Juxtaposing our results with recent findings by Schwartz and Tierney (2024) would even suggest that external threats, if anything, are more likely to unite than divide. While they find that external threat primes—if vivid and bipartisan—could reduce affective polarization, we show with robust evidence that political communication about external threats—if less vivid and nonpartisan or even one-sided—does not heighten preexisting polarization.

Second, this paper contributes to the literature on partisans’ information processing in American politics. One strand of scholarship argues that parties and their leaders often have strong impacts on public opinion, regardless of their messages (Berinsky, 2009; Lenz, 2012; Barber and Pope, 2019). In an extreme form, partisans may even update their policy preferences negatively in response to out-party cues (Nicholson, 2012; Merkley and Stecula, 2021). Another strand of scholarship starts to question this form of “partisan intoxication” (Fowler, 2020), arguing that partisans can favorably update their beliefs in response to new persuasive information (Coppock, 2023), even if the messenger is an out-party leader. Focusing predominantly on ideologically polarized and domestic policy issues, Tappin et al. (2023) and Fowler and Howell (2023) find from US survey experiments that out-party cues did not backfire and could even be persuasive. Our timely study, which uncovers similar findings in the policy domain of international security, strengthens this nascent yet incomplete literature through a new empirical lens, shedding light on the boundaries of elite cues in biasing public opinion in the realm of American foreign policy.

Our finding on the bipartisan increases in foreign policy assertiveness has implications for US–China relations. Despite rampant affective polarization among the US public, elites from both sides of the aisle, as our finding suggests, can be confident not only that their rhetoric about the China threat will not backfire in terms of public approval of their hawkish policy orientations but also that they will find a domestic audience receptive to this kind of rhetoric. The potential political gains from deploying such assertive rhetoric against China speak to the increasing hawkishness displayed by Democratic and Republican elites in the public sphere.¹² Chinese leaders and diplomats, on the other side, have employed similarly assertive rhetoric against the United States (e.g., “wolf warrior diplomacy”) to rally domestic support (Xu, 2024). The escalating mutual hostility between the two nations, in turn, has raised tensions and exacerbated US–China relations (Weiss, 2019). To investigate the prospect of de-escalation, future research should test whether and how elite rhetoric that downplays the China threat can reduce Democrats’ and Republicans’ hawkishness amid hyperpartisanship. Interesting dynamics may emerge when Republican elites “go against type” by adopting a dovish stance on China.