Can tragedy unify a polarized public?

The American public is largely polarized, ideologically and affectively. While studies continue to parse out the contours of this partisan division, the fact remains that whether through elected officials (Layman and Carsey, Reference Layman and Carsey2002; Druckman, Peterson, and Slothuus, Reference Druckman, Peterson and Slothuus2013), the media (Levendusky, Reference Levendusky2013), party activists (Layman et al., Reference Layman, Carsey, Green, Herrera and Cooperman2010), or neighbors (Brown, Reference Brown2022), and whether by means of issue-based sorting (Fiorina, Abrams, and Pope, Reference Fiorina, Abrams and Pope2005) or social identity-based sorting (Mason and Wronski, Reference Mason and Wronski2018), the American public has become affectively polarized. In response, many scholars have turned their attention towards strategies to depolarize partisans, with mixed results.

Recent research indicates that rally effects are not just limited to terrorist attacks or other similar anthropogenic mass tragedies. Importantly, studies find that natural disasters trigger a rally effect among at least co-partisans (Heersink et al., Reference Heersink, Jenkins, Olson and Peterson2022); that natural disasters, terrorist attacks, and pandemics have similar effects because rally effects are more strongly connected to centralization of government (Lago and Blais, Reference Lago and Blais2023); but studies of the effects of COVID-19 in particular find an emotionally driven rally effect in the Netherlands and other countries (Schraff, Reference Schraff2021; Hegewald and Schraff, Reference Hegewald and Schraff2024) as well as in Sweden (Johansson, Hopmann, and Shehata, Reference Johansson, Hopmann and Shehata2021), though such effects are short-lived.

The theoretical basis of many attempts to depolarize the public, successful and unsuccessful, rely upon efforts to recategorize or alter social identity conceptualization to see similarities with the partisan outgroup (Levendusky, Reference Levendusky2018; Simas, Clifford, and Kirkland, Reference Simas, Clifford and Kirkland2019; Druckman et al., Reference Druckman, Gubitz, Levendusky and Lloyd2019) or to bracket differences in a way that allows partisans to conceptualize outpartisans as less distant and extreme (and less threatening) (Druckman et al., Reference Druckman, Klar, Krupnikov, Levendusky and Barry Ryan2022; Marsh, Reference Marsh2022a). But, there are exogenous factors that may influence such perceptions without the active intervention of political scientists and journalists: traumatic events.

Research has indicated that traumatic experiences or events influence political behaviors such as turnout and protest engagement (Williamson, Trump, and Einstein, Reference Williamson, Trump and Levine Einstein2018; Ojeda and Pacheco, Reference Ojeda and Pacheco2019; Marsh, Reference Marsh2023) as well as contact with government (Cohen et al., Reference Cohen, Gunderson, Jackson, Zachary, Clark, Glynn and Leo Owens2019).Footnote ¹ Further, political scientists have for decades identified a “rally effect” in which some types of traumatic events, such as terrorist attacks and other threats to national security boost support for the president and cooperation among political elites across the partisan divide (Lee, Reference Lee1977; Norrander and Wilcox, Reference Norrander and Wilcox1993; Marcus and MacKuen, Reference Marcus and MacKuen1993). Beyond this, traumatic events that threaten the American public in some way, such as 9/11, trigger increased nationalism, patriotism, and national unity (Li and Brewer, Reference Li and Brewer2004). In short, traumatic events are psychologically challenging events, which oftentimes involve circumstances in which one feels a larger social group to which they belong is threatened. In particular, during some traumatic events, like 9/11, many perceive the event as threatening “America” and thus all those who identify as “American.” The traumatic event, therefore, at once 1) provides a foundation for recategorizing identity under the superordinate “American” identity, rather than subordinate identities, and 2) makes this superordinate identity more salient in the process of managing posttraumatic stress.

One foundation of perceiving a shared superordinate social group identity is that the social group shares some “essence” such as shared attributes or heritage among members of that group (Yzerbyt, Rocher, and Schadron, Reference Yzerbyt, Rocher, Schadron, Spears, Oakes, Ellemers and Alexander Haslam1997; Brewer, Hong, and Li, Reference Brewer, Hong, Li, Yzerbyt, Judd and Corneille2004). This is the foundation of the recategorization strategy used in a successful depolarizing effort in Levendusky (Reference Levendusky2018), by reminding partisans of what makes them “American.” A second foundation is when members are seen as part of a single group because of a common problem that creates a common purpose (Hamilton, Sherman, and Lickel, Reference Hamilton, Sherman, Lickel, Sedikides, Schopler and Insko1998; Brewer, Hong, and Li, Reference Brewer, Hong, Li, Yzerbyt, Judd and Corneille2004). Mass tragedies such as 9/11 or the spread of a deadly disease trigger this second type of superordinate-identity-recategorization, as these are both threats that unite “Americans” together to solve a common challenge. It is in this way that great tragedy can unite a polarized public: by providing them with a common purpose in the face of serious threats to their own and co-members’ lives. But, such unifying potential is vulnerable to forces that maintain a fragmented public.

A theory of persistent partisan hostility

To understand why American partisans continue to demonstrate high levels of negative outparty affect even through events which should unify them and enable them to overcome differences, I develop a theory of persistent partisan hostility. This theory contends that traumatic events that 1) reveal or exacerbate inequities between subordinate groups and/or 2) emphasize issue cleavages between parties negate the depolarizing effects traumatic events can have (providing a means for Americans to recategorize outpartisans within the “American” superordinate identity).

First, by “traumatic event,” I mean events that threaten serious violence or death to a subject or their close friends or family members (APA, 2013), and are likely to shatter individuals’ perceptions of how the world works (Janoff-Bulman, Reference Janoff-Bulman1992). I conceptualize traumatic events as largely being either anthropogenic or natural and as being accidental or intentional (Marsh, Reference Marsh2023). Further, mass tragedies are more likely to be politically relevant, meaning they are more likely to induce some political behavioral outcomes when they are events that require governmental response by law or precedent, are of a scale and severity that pressures a government to respond, or if policy demanders or elites make the event political (Marsh, Reference Marsh2022b, Reference Marsh2023). Further, it is political elites who make politically relevant traumatic events partisan such that the processes of polarization amplify responses to such events (Marsh, Reference Marsh2022b). Policy demanders typically polarize a traumatic event to achieve policy goals or concessions, leveraging the event as a focusing event that leads to policy gains, while political elites typically polarize for electoral gain, seeking to shirk or shift blame (Marsh Reference Marsh2022a, b).

Importantly, the processes of this theory are dependent largely on elite framing in that elites can utilize either of these principles to counter or at least decrease the depolarizing tendencies of traumatic events. That is, if policy demanders (party activists or the media) or party elites (elected officials) use the mass tragedy to identify existing social group inequities that align with partisan identity or to emphasize issue or value cleavages between the parties, the event will resist depolarizing forces and could even further polarize the mass public.

One case exemplifying how elites can successfully intervene to counter depolarizing tendencies of a mass tragedy is the COVID-19 pandemic. From very early on in the pandemic, party elites polarized the public on attitudes towards COVID-19 and towards policies meant to respond to the pandemic (Green et al., Reference Green, Edgerton, Naftel, Shoub and Cranmer2020; Box-Steffensmeier and Moses, Reference Box-Steffensmeier and Moses2021). Such messaging from party elites largely relied upon decades of messaging about trust in science, such that Republican elite efforts to sow distrust in science laid the groundwork for Republican mass distrust in scientific policy solutions in response to COVID-19, while trust in science characterized Democratic attitudes (Hegland et al., Reference Hegland, Zhang, Zichettella and Pasek2022). More proximate to the pandemic, then-President Trump’s messaging influenced Republican mass behavior, while scientific research did the same for Democrats in structuring polarized COVID-19 policy attitudes (Golos et al., Reference Golos, Hopkins, Bhanot and Buttenheim2022).

Elites emphasized existing cleavages between Democrats and Republicans regarding trust in science and trust in government that ultimately led to a conflict extension (Layman and Carsey, Reference Layman and Carsey2002) in which partisans were divided on almost every COVID-19 policy (Druckman et al., Reference Druckman, Klar, Krupnikov, Levendusky and Barry Ryan2021b; Rodriguez et al., Reference Rodriguez, Gadarian, Goodman and Pepinsky2022). Geography played an important factor in that areas already geographically polarized demonstrated polarized adherence to COVID-19 policies, especially mask wearing (Baxter-King et al., Reference Baxter-King, Brown, Enos and Vavreck2022). Geography may be important in the dynamics of how traumatic events impact behavior via identity primes through the politicization of the event (Marsh, Reference Marsh2023). Further, while partisan hostility shaped COVID-19 attitudes and behavior, this relationship was muted in geographic locations with more severe outbreaks of the disease (Druckman et al., Reference Druckman, Klar, Krupnikov, Levendusky and Barry Ryan2021a). This indicates that more severe trauma exposure increases the likelihood of a tragedy breaking the pattern of partisan animosity.

Finally, the polarization of partisans through the COVID-19 pandemic was characterized by important differences in trust in government and medical officials, and perceived risk (Kerr, Panagopoulos, and van der Linden, Reference Kerr, Panagopoulos and van der Linden2021). Trust and risk perception undergo important changes in posttraumatic stress and posttraumatic growth responses to mass tragedies (Marsh, Reference Marsh2023). Two other important psychological responses in the wake of traumatic events are blame attribution and sympathy. When such events occur, elites try to avoid blaming themselves, while often also blaming outparty elites (Boin, Hart’t, and McConnell, Reference Boin, Hart and McConnell2009). Such blaming is an expression of partisan polarization when the mass public accepts and attributes blame of mass tragedies on outpartisans absent other reasons. As such, blaming outpartisans and having less sympathy for them when they die is an indicator of partisan hostility.

Expectations

Given this theory of persistent partisan hostility, I expect that partisan animosity will persist through the COVID-19 pandemic along with the psychological responses of blame attribution and sympathy. Democratic and Republican respondents are more likely to blame outpartisans for their own death from COVID-19 ( ${H_{1a}}$ ). The effects should be especially strong when the outpartisan is vocally anti-vaccination ( ${H_{1b}}$ ). Importantly, effects should be strongest among Democrats given the partisan nature of the COVID-19 response and vaccine hesitancy among Republicans ( ${H_{1c}}$ ).

Further, I expect Democrats and Republicans will be less sympathetic towards outpartisans who have died from COVID-19 ( ${H_{2a}}$ ). The effects should be especially strong when the outpartisan is vocally anti-vaccination ( ${H_{2b}}$ ).

Empirical strategy

Experimental design

To test these hypotheses, I construct and pre-register a traditional survey experiment.Footnote ² In this experiment, respondents read a short news story (see Figure 1) about a community leader who has died from complications related to a coronavirus infection. In August 2021, I fielded the survey experiment on a nationwide Lucid Theorem sample with quotas for age, gender, ethnicity, and region of 1,360 U.S. adults. I then replicated the study through a 1,000-individual sample on the 2021 Cooperative Election Study (CES) through YouGov, fielded in October 2021. YouGov uses stratified random sampling from its existing panel to approach national representativeness. The experiments were exactly the same in both studies. Henceforth, Study I refers to the Lucid Theorem study and Study II to the CES (YouGov) study.

Figure 1. Control condition.

Respondents were randomly sorted into one of six experimental conditions outlined in Table 1.Footnote ³ The treatments combined all combinations of manipulating partisan information as well as information about the deceased man’s vaccination status and stance. As such, there are six conditions: 1) Control – no information about partisanship or vaccination status/stance, 2) Anti-Vax – the victim is given no partisan identity, but identified as an anti-vaccination advocate who is unvaccinated, 3) Democrat – identified as a Democrat with no information about vaccination status/stance, 4) Republican – same as condition 3, but Republican, 5) Anti-Vax Democrat – a combination of conditions 2 and 3, and 6) Anti-Vax Republican – a combination of conditions 2 and 4. These conditions allow me to test the effect of both partisan identity and vaccination stance in assessing my dependent variables of interest.Footnote ⁴ I model the effects separately for partisans and other relevant subgroups, but leverage alternative modeling strategies in the SI, section D. I pre-registered that I would also include difference-in-means tests as a main part of the analysis. I deviate from this part of the pre-registered analysis plan because regression analyses are more appropriate and easily interpretable to readers, and given reviewer suggestions. As such, I include the difference-in-means tests in the SI, section E.

Table 1. Experimental design

I provide figures of the estimated effects of each treatment condition compared to the control in the SI, section D, but in the main text, I utilize regression models in partisan subgroups with an interaction between the two types of treatment: 1) victim partisanship (no partisanship information, Democratic activist, Republican activist) and 2) victim anti-vaccination stance (no information, anti-vaccination).

In a second set of analyses, I run the same model, but on different subsets of the data or with additional variables added as controls. In these specifications, I subset by respondent vaccination status as well as exposure to a close friend or family member dying from COVID-19 or a COVID-19-related illness. I also run specifications on the full sample with a number of control variables including: respondent vaccinated (1 if respondent has at least one dose of a two-dose vaccine, 0 otherwise), covid death (1 if a close friend or family member died of COVID-19 or related complications and 0 otherwise), covid concern (101-pt scale of how concerned the respondent is about the COVID-19 pandemic), and vaccine mandate (101-pt scale of how strongly they believe the government should require a vaccination for all who can medically receive the vaccine).

Limitations

One important point to consider is that the treatment conditions seek to increase external validity, but in a way that may make it more difficult to identify the specific mechanism. The treatments identify the victim as a party activist and an anti-vaccination advocate. As such, it is not possible to distinguish the effects of partisanship from partisan activist and being unvaccinated or being anti-vaccination for oneself, compared to advocating for others to also not be vaccinated. Further, the treatments do not provide information about the behaviors of the victim, meaning respondents are imputing their own assumptions about masking, social distancing, and other behaviors.

In explaining the empirical design, I explain the reason for such choices, but it is important to note again the limitations such design choices impose. I have specified the models in such a way as to better contextualize this particular limitation. Namely, I have interacted with two treatment types, analyzed multiple subgroups separately, and included a number of control variables that should parse out some of the threats to identifying the effect of partisanship and anti-vaccination.

Another limitation is that I only test a single event, while my theory proposes to explain a broader range of types of traumatic events. Previous research indicates that there are rally events after natural events, such as a pandemic, but that there have been mixed results because the effect is conditioned by shared partisanship between elected officials and constituents (Heersink et al., Reference Heersink, Jenkins, Olson and Peterson2022). Further, after COVID-19, executives enjoyed an initial rally effect, but one which eroded in the long-term as the pandemic “normalized” (Johansson, Hopmann, and Shehata, Reference Johansson, Hopmann and Shehata2021). This is important because previous work often conceptualizes political responses to traumatic events in relation to rally effects and the incentives they provide for leaders.

9/11 provided a clear external enemy in a way that natural disasters (and pandemics) often do not. As policy demanders constrain elites, however, this dynamic is changing. Even still, there were limited efforts to blame external enemies for COVID-19 (though some actors, including then-Pres. Trump, did blame China explicitly). My theory of persistent partisan hostility implies that despite these differences, we have reason to believe that such patterns of outpartisan hostility and polarization will perpetuate as long as elites are able to polarize. I argue that one way they do this is by identifying existing social group inequalities that align with existing partisan identity and/or by emphasizing issue or value cleavages between the parties. In this study, I have tested this theory in the context of COVID-19, but future work should investigate how this theory does or does not travel to other natural events and to anthropogenic tragedies.