Not "My" crisis: Social identity and followers' crisis responses to COVID-19.

Operationalizing social group identification as political partisanship, we examine followers' (i.e., US residents') affective experiences and behavioral responses during the initial COVID-19 outbreak in the United States (March to May 2020). In Study 1, we conducted content analyses on major news outlets' coverage of COVID-19 (N = 4319) to examine media polarization and how it plays a role in shaping followers' perceptions of the pandemic and leadership. News outlets trusted by Republicans portrayed US President Donald Trump as more effective, conveyed a stronger sense of certainty with less negative affective tone, and had a lower emphasis on COVID-19 prevention compared to outlets trusted by Democrats. We then conducted a field survey study (Study 2; N = 214) and found that Republicans perceived Trump as more effective, experienced higher positive affect, and engaged in less COVID-19 preventive behavior compared to Democrats. Using a longitudinal survey design in Study 3 (N = 251), we examined how emotional responses evolved in parallel with the pandemic and found further support for Study 2 findings. Collectively, our findings provide insight into the process of leadership from a social identity perspective during times of crisis, illustrating how social identity can inhibit mobilization of united efforts. The findings have implications for leadership of subgroup divides in different organizational and crisis contexts.

INTRODUCTION

In times of crisis, followers look to their leaders for explanation, reassurance, and direction (Probert & Turnbull, 2011). The twin characteristics of uncertainty and threat is said to crystallize individuals’ thoughts and visions and mobilize them into action, prompting the emergence of new leaders, and the maturation of existing ones (Boin & Hart, 2003; Conger & Kanungo, 1987; Pearson & Clair, 1998), while simultaneously unifying followers against a common threat (Drury et al., 2009; Flade et al., 2019; Gaertner et al., 2000). This mechanism rests on the assumption that followers unanimously follow their leader. However, this might not be the case in intergroup contexts where leaders and followers hold membership of disparate social groups (Hogg et al., 2012a; Pittinsky & Simon, 2007), yielding conflicting theories about how followers might mobilize during crises. Namely, group differences and boundaries may be overlooked by members of subgroups because navigating through the crisis serves as a common goal that guides their united efforts. Yet, crises could also serve as a contextual factor that promotes higher degrees of separation and polarization, especially if leaders encourage boundaries between subgroups (Kellerman, 2004) to support their own agenda (Hermann & Kegley, 1996).

These competing perspectives led us to question whether the COVID‐19 pandemic – the first major global health crisis of the 21st century – caused the American people, already heavily divided along political affiliations, to unite against this common threat or to further separate. Evidently, the American people, namely partisans, were polarized in their responses to this pandemic including perceptions of the severity of COVID‐19 (Funk & Tyson, 2021), mask‐wearing (van Kessel & Quinn, 2020), and COVID‐19 vaccination (Funk & Tyson, 2021). To continue to progress and to prepare for any inevitable future crises, it is important to understand the unfolding of such outcomes. Specifically, how does the interplay of follower characteristics, leadership, and the context shape follower responses in a crisis? Given the hyper salience of bipartisanship in the United States, we examine how social group identification relates to followers’ perception of leader effectiveness and their affective experiences and crisis management behavior.

In doing so, we contribute to the literature in several ways. First, although previous studies have well established that social identity impacts followers’ endorsement of leaders and perceived leader effectiveness (Giessner & van Knippenberg, 2008; Platow et al., 2006; van Knippenberg & Hogg, 2003), such studies were conducted in contexts characterized by limited uncertainty. Examining follower social identification and their perceptions of leadership during a global pandemic provides an opportunity for contextualizing the relationship and exploring its generalizability by using extreme or deviant purposeful sampling (see Patton, 2002).

Second, the current understanding of how followers’ perceptions and evaluations of leaders shape their subsequent affect and behaviors is less than robust because of the dominance of a theoretical approaches, as well as methodological limitations (Duck & Fielding, 2003). We advance knowledge by exploring how social identity ties to followers’ affect and crisis responses using a combination of methodological approaches. Our content analysis of media coverage adds to the existing body of literature by showcasing how social identity molds the way in which leadership is socially constructed and catalyzed by the mass media (Bligh et al., 2004a; Lau et al., 2020) during a crisis, and its role in creating division between followers.

Third, our findings extend the thus far embryonic empirical literature on intergroup leadership. Although the importance of intergroup dynamics in leadership has been stressed in numerous writings (Hogg et al., 2012a, 2010; Pittinsky & Simons, 2007; Tee et al., 2013), relatively few studies have examined this phenomenon (Hohman et al., 2010; Rast et al., 2015, 2018; van der Stoep et al., 2020). We hope our focus on intergroup leadership in the presence of a global crisis can contribute to this body of literature by further contextualizing and refining its consequences for intergroup members. Overall, our paper views leadership as a social construction process driven by social identity and uses multiple theories and samples to advance our understanding of how social identification associates with the way followers respond to major exogenous shocks: cognitively, emotionally, and behaviorally.

SOCIAL IDENTITY, INTERGROUP LEADERSHIP, AND CRISIS

Much empirical research has demonstrated the influence crises have on the leadership process. Followers look to leaders for sensemaking during crises; they also count on leaders to guide them through times of hardship with a vision and shared purpose (Boin & Hart, 2003; Meindl, 1995). Leaders who emphasize visions and values promote followers’ hope and optimism and convey a sense of collective efficacy in overcoming crisis and adversity (Shamir et al., 1993; Shamir & Howell, 1999), which creates a “psychological platform for group members to coordinate efforts to tackle stressors” associated with crises (Van Bavel et al., 2020, p. 466). Thus, crises serve as an important contextual factor for the emergence of effective and charismatic leadership (Bligh et al., 2004b; Conger & Kanungo, 1987; Pearson & Clair, 1998). In a series of studies conducted by Williams, Pillai, and colleagues, they consistently found that perceptions of crisis were predictive of charismatic leadership and voting behaviors among voters in the US presidential elections (Pillai et al., 2003; Williams et al., 2012, 2009). President George W. Bush was considered as more charismatic, both objectively and subjectively (Bligh et al., 2004a, 2004b), and received a higher approval rating after the 9/11 attacks compared to before (Langer, 2001). Based on this body of research, we shall expect followers to unite and look to their leaders for guidance in a crisis. Successful navigation during such a time of uncertainty should lead followers to perceive the leader as more effective and charismatic. Yet, research on social identity and leadership suggests this might not always be the case, particularly in an intergroup context. Following, we provide an overview of the literature on social identity theory of leadership (Hogg, 2001) and apply it in our hypothesis development.

Social identity of leadership theory

The central tenant of social identity theory of leadership (SITL) is that the social groups that we belong to largely influence our attitudes, cognition, and behaviors. Those who are considered prototypical in their perspective social groups (a prototype is an approximate set of attributes that define the group) are more likely to be liked by their fellow group members (Hogg, 1993). Such popularity grants prototypical individuals power and influence to lead the group (Haslam et al., 1995; Hogg et al., 2012b).

Hogg et al. (2012b) discusses that leader evaluations are much of an attribution process through the lens of SITL. Because ingroup, prototypical leaders are influential and liked by their ingroup members, effective leadership and charisma are attributed to these leaders. In other words, the perceived effectiveness of group leaders is a function of their prototypicality rather than their leadership behaviors, particularly in groups with hyper identity salience (Fielding & Hogg, 1997; Hains et al., 1997; Hogg et al., 1998).

Because ingroup leaders are trusted by their followers (Haslam & Platow, 2001b), followers believe that such leaders’ behaviors are beneficial to the group (group‐oriented; Hogg et al., 2012b), even when they are not (Hogg et al., 2012b; Platow & van Knippenberg, 2001; Platow et al., 2006; van Knippenberg & van Knippenberg, 2005). Indeed, studies have shown that ingroup/prototypical leaders enjoy followers’ trust and positive evaluations regardless of them being selfish vs group‐oriented (Platow et al., 2006; van Knippenberg & van Knippenberg, 2005) or when they succeed or fail (Giessner & van Knippenberg, 2008; Giessner et al., 2008). Followers seem to rationalize prototypical leaders’ non‐group‐oriented behaviors, and hence, their support for their leaders. Thus, according to SITL, the key to effective leadership (in the eyes of followers) is the leader's prototypicality.

Because leadership is a social‐cognitive group process (Haslam & Platow, 2001a) and prototypicality is socially constructed, leaders may focus on the construction of their prototypicality to garner followers’ support. For example, leaders may define the core norms and values for their group and convince their followers that themselves are the embodiments of such norms and values. They do so through tactical use of rhetoric, communication, and behaving in rhetoric‐consistent ways (Merritt & Monin, 2011; Stone et al., 1997). Reicher and Hopkins (1996) call these leaders “entrepreneurs of identity” (also see Reicher et al., 2005; Klein et al., 2007), highlighting their attempt or ability to construct and manage the group prototypes (Hogg & Reid, 2006; Seyranian, 2014; Seyranian & Bligh, 2008). Through controlling their group prototypicality, these leaders enhance their own prototypicality and secure their leadership position (Hogg et al., 2012b).

Leaders in intergroup contexts

Leaders often find themselves in the position to lead individuals that subscribe to multiple social groups (e.g., cross‐functional teams, mergers, and acquisitions); the presence of intergroup relations adds a layer of complexity to this model. Outgroup leaders are often viewed as highly non‐prototypical and untrustworthy by outgroup followers and thus, produce lower performance (Hogg et al., 2012b). This is why, often, corporate mergers fail – as the leader of the merged organization is still viewed as outgroup member post‐merger (Ullrich & van Dick, 2007), and as a result, these leaders are less likely to be trusted, viewed as charismatic, effective, or mobilize their followers. Therefore, an intergroup leader's challenge is to develop a common ingroup identity of followers who identify with different subgroups. In this case, leaders may choose to actively engage in identity entrepreneurship (Reicher & Hopkins, 1996, 2001) to enhance their prototypicality to both ingroup and outgroup followers by appealing to their group identities and the relations between them (intergroup leadership; see also Hogg et al., 2012a).

Hogg et al. (2012a) coined the term “intergroup relational identity” to describe the need for intergroup leaders to balance between superordinate identity and subgroup identities and frame the two as mutually beneficial to unite their followers. To do so, Hogg and colleagues suggest that leaders can a) champion intergroup collaboration, b) highlight the intergroup relational identity consistently and publicly, and c) collaborate with other leaders within subgroups. Done successfully, not only should intergroup members have more favorable evaluations of the intergroup leader (Kershaw et al., 2021), but they should also be more likely to feel psychologically empowered (van der Stoep et al., 2020) and have more positive interactions with inter‐subgroup members (Kobayashi et al., 2021).

Intergroup leadership in the context of crisis

Crises serve as a contextual catalyst for intergroup leaders to unite followers of different subgroups; yet, leaders often cede the opportunity to do so (Pittinsky & Simon, 2007). Subgroups leaders, especially those with precarious power, would even exploit inter‐subgroup conflicts to secure their own positions (Bekkers, 1977; Oberschall, 2000; Rabbie & Bekkers, 1978). When leaders feel threatened in their positions, they are more likely to instill intergroup competition to secure support from ingroup members (Oberschall, 2000; Rabbie & Bekkers, 1978). According to SITL, this tactic would be effective because it accentuates the difference between subgroups and reinvigorates the ingroup's distinctiveness and the leader's prototypicality (Hogg, 2001). The presence of crisis serves as a catalyst for the effect of prototypicality on perceived leader effectiveness (Cicero et al., 2010, 2020; Rast et al., 2013).

In sum, research on crisis and leadership suggests that followers tend to perceive leaders as more effective and charismatic during crises. We question the prevalence of this effect in a threatening intergroup context where social identity may inhibit leaders’ ability to unite followers and manage the crisis. History also suggests that leaders do not necessarily see crisis management as their priority in a threatening situation. How does social identity impact ingroup/outgroup followers’ responses to such leadership in times of crisis? The COVID‐19 pandemic offers a unique opportunity to examine the role of social identification, leadership, and context in followers’ responses in a crisis.

HYPOTHESIS DEVELOPMENT

In the reported studies, we operationalized social group identification as political affiliation and tested its relationships with perceptions of leader effectiveness and followers’ affect and crisis management behavior in the context of the COVID‐19 pandemic. Specifically, referencing previous SITL research in the US political context, we considered bipartisanship as an indicator of inter‐subgroup relations in the United States (Alabastro et al., 2013; Hohman et al., 2010; Rast et al., 2015). From a value‐based perspective, partisans subscribe to one party versus another based on alignments of values and beliefs with the party (Swigart et al., 2020). Broad distinctions between conservatives and liberals (roughly corresponds to the Republican and the Democratic party, respectively) includes one's belief in a) social change versus tradition, b) equality versus hierarchy, and c) emphasis on contextual factors versus personal agency (Jost et al., 2009; Swigart et al., 2020). Conservatives, compared to liberals, also tend to prioritize individual freedom over collective egalitarian goals or government interventions (Jost, 2017). Recently, social identity theory has been used as an alternative lens to view partisanship in the field of political science (Iyengar & Krupenkin, 2018; Van Bavel & Pereira, 2018). Through this lens, partisanship is partly viewed as symbolic labels (Devine, 2015; Malka & Lelkes, 2010). In other words, partisans’ cognition, attitudes, and behaviors are driven by not only their values and beliefs but also their social identification (Swigart et al., 2020). We adopt this view of social group identification as political affiliation in the current studies.

According to research on crisis and leadership (Bligh et al., 2004a, 2004b; Pearson & Clair, 1998; Pillai et al., 2003; Williams et al., 2009, 2012), we might expect the American people to turn to their leader (the US President, Donald Trump), trusting him to effectively handle a national and global crisis. Yet, politically, the United States seems to be more divided than ever (Pew Research Center, 2017). The magnitude of this political polarization can potentially overshadow the need for unity amid a crisis (van Bavel et al., 2020), such that the American people remain divided in their trust and endorsement of their President. We, thus, first explore whether polarization in the United States prevails during the early stage of the COVID‐19 crisis and determines followers’ perceived effectiveness of their leader. Echoing previous research on social identity and leader evaluations (Giessner & van Knippenberg, 2008; Giessner et al., 2008), we consider followers’ self‐categorization as Democrat or Republican (i.e., party affiliation) as tantamount to social identification with the party. Such treatment of social identification is also supported by voting patterns of partisan voters. According to different polling sites, about 94% of registered partisan voters (both Republicans and Democrats) voted for their party candidates in the 2020 US Presidential election (ABC, 2020), which suggests identification with the party.

The unfolding of the leadership process depends on the context, the followers, and of course, the leader (Gleibs & Haslam, 2016). According to SITL (Hogg, 2001; van Knippenberg & Hogg, 2003), followers’ endorsement of a leader depends on the extent to which the leader enhances the group's identity and well‐being. Scholars have argued that Trump's success in the 2016 US Presidential election partly, at least, owes to his skills as an “entrepreneur of identity” (Reicher & Haslam, 2017). Namely, Trump was able to articulate his constituents’ sense of reality by contextualizing their problems. By providing concrete solutions to the problems that he highlighted, Trump further secured his influence over this group of followers by demonstrating his ingroup orientation (Reicher & Haslam, 2017).

In the context of the COVID‐19 pandemic, we speculate that Trump similarly served as an “entrepreneur of identity.” However, rather than unifying his subgroup followers, his approach seems to lead to further division between them (Bekkers, 1977; Rabbie & Bekkers, 1978). Namely, as a leader of the superordinate group (the President of the United States), he only appealed to his ingroup followers (the Republicans). He negotiated his ingroup identity, for example, by repeatedly referring to COVID‐19 as the “Chinese virus” (BBC, 2020), reinforcing the idea that China is to be blamed for the pandemic. According to political psychologists, conservatives are more likely to prefer a sense of order and structure, are less tolerant of uncertainty, and are more inclined to justify unequal economic systems compared to liberals (Jost, 2017). Thus, recognizing China's rapid development in recent decades threatens the social hierarchical status of the United States, Republicans consider China as a rival country (Silver et al., 2020). In Trump's 2016 US Presidential campaign, he constructed anti‐china rhetoric, describing China's trade policy as “unfair,” and soon after his inauguration, raged a trade war against China (BBC, 2016). This tactic appeals to his constituents’ party identity. Resuming this effective tactic––blaming China for the COVID‐19 crisis––Trump enhances his prototypicality, and hence perceived effectiveness, among his ingroup followers.

Conversely, this tactic is unlikely to work on his outgroup followers (Democrats) because of the Democrats’ lower endorsement of inequality compared to Republicans (Jost, 2017). Trump also did not demonstrate a commitment to his outgroup followers’ well‐being. In the early onset of the pandemic, Democrat‐leaning congressional districts (which tend to be metropolitan and urban areas with high population density where viruses can spread quickly and easily) observed higher death rates compared to Republican‐leaning congressional districts (which tend to be suburban and rural areas; Jones & Kiley, 2020). Democrats were also more likely to consider COVID‐19 as a major threat compared to Republicans (March 2021; Funk & Tyson, 2021). Despite the toll and concerns observed among Democrats, Trump “downplayed” the severity of COVID‐19 in his rhetoric (especially during the early onset of the pandemic; Chan, Chiu, et al., 2021; Cohen et al., 2020). In numerous announcements, Trump claimed that the situation was “under control” (McCarthy, 2020) despite the rising death rate associated with COVID‐19. In late February 2020, he asserted that COVID‐19 is but seasonal influenza that would “disappear” one day (Yamey & Gonsalves, 2020). Compared to prototypical leaders, outgroup leaders ought to engage in group‐oriented ways in order to convince outgroup followers of their positive group orientation to garner their trust and support (Hogg et al., 2012b). Trump's rhetoric and behavior suggested otherwise.

Therefore, we reason that in the face of a global health crisis, Trump focused on appealing to his ingroup followers through affirming the ingroup identity and highlighting his prototypicality. This tactic should lead his ingroup followers to perceive him as an effective leader. On the contrary, his minimization of the severity of COVID‐19 highlights his non‐prototypicality and lack of concern for his outgroup followers. Thus, we hypothesize that political affiliation is related to perceived leader effectiveness: Republicans (ingroup followers) are likely to consider Trump as more effective in managing the crisis than Democrats (outgroup followers; Hypothesis 1).

Leader effectiveness and followers’ response

Taking on a social identity perspective to understand the leadership process in the COVID‐19 pandemic, we also examined how followers’ partisanship and identification with their leader related to their affect and behaviors. Ample research has demonstrated the role of affect in responses to crises (Fredrickson et al., 2003; Sommer et al., 2016) and trauma (Tedeschi & Calhoun, 2004). Emotional regulation has great implications for how people maneuver in times of crisis, and whether an individual is to derail or thrive from this experience. On one hand, the pressure and uncertainty elicited by crises are often associated with negative affective responses such as distress, anxiety, frustration, and fear, which can fuel feelings of helplessness and lack of efficacy (Yu et al., 2008). Cognitive energy is also likely to be subsumed by negative affect (Beal et al., 2005), which influence one's abilities to pivot and be agile in a time of uncertainty (Brief & Weiss, 2002). On the other hand, positive affective responses such as happiness and joy are also critical for individuals navigating a crisis. Positive affect is linked to greater efficacy, optimism, and perseverance, which are important factors for coping with adversity or crises (Judge et al., 1999). Positive affect is also important in cultivating the flexibility and resilience that are crucial for post‐traumatic growth (Kaplan et al., 2013; Tugade & Fredrickson, 2004). Post 9/11, individuals who exhibited higher positive emotions were less likely to develop depressive symptoms and experienced more post‐crisis growth (Fredrickson et al., 2003).

Taking on a follower‐centric approach to leadership, Dasborough et al. (2009) drew from affective events theory and theorized that a leader's favoritism and followers’ attribution of leader insincerity can cause the experience of negative emotions among followers. Extending upon this line of work, Tee et al. (2013) posited that leaders’ emotion expression and behaviors can drive followers’ affective, cognitive, and behavioral responses in an intergroup context. When leaders are perceived to be expressing emotions that are incongruent with the group's situation, followers view them as less trustworthy and have less confidence in them. Such unfavorable perceptions may lead to the experience of negative emotions for the followers (Van Dijke & De Cremer, 2010). Conversely, leaders’ appropriate emotional expression leads to a more intense experience of positive emotions (Bucy, 2000). These effects should be stronger when the group experiences situational ambiguity (Bucy, 2000; Stewart et al., 2009) and when the leader is viewed as an outgroup member in an intergroup context (i.e., nonprototypical leader; Tee et al., 2013).

Based on the summarized research, we argue that given followers’ social group identification, Trump's leadership tactic results in differentiated crisis responses from ingroup versus outgroup followers. Namely, the low severity (lack of urgency) conveyed by Trump regarding COVID‐19 mismatches his outgroup followers’ (Democrats’) expectation for their country leader's expression of situationally appropriate sentiment in a crisis. Such a mismatch further signals to the outgroup followers that their leader is ineffective (Bucy, 2000; Stewart et al., 2009; van Knippenberg, 2011) and might not be acting in their best interests (Michie & Gooty, 2005). As a result, outgroup followers experience more negative and less positive affect (Tee et al., 2013). In contrast, Republicans should be more likely to perceive Trump as effective and because they were less concerned with the COVID‐19 virus at the beginning of the crisis (around March 2020 in the United States), they are likely to perceive Trump's emotional responses as appropriate. This also echoes the literature on political ideologies which stresses that conservatives are more likely to reject science (van der Linden et al., 2021) and CDC's guidelines which call for behavioral change (e.g., mask‐wearing and lockdowns) as they threaten their individual freedom and are more likely to embrace Trump's sentiment that signals stability. Thus, we hypothesize that political affiliation is associated with followers’ affect via perceived leader effectiveness: Democrats will consider Trump as less effective as a leader and experience higher levels of negative affect (Hypothesis 2a) and lower levels of positive affect (Hypothesis 2b) as compared to their Republican counterparts.

Relatedly, research on social identity and leadership suggests that the perceptions of leadership impact followers’ behaviors. Drawing from research on charismatic and transformational leadership (Bass & Avolio, 1993; Conger & Kanungo, 1987; Shamir et al., 1993) and social identity theories, scholars argue that leaders who are perceived as transformational (or charismatic) are able to successfully mobilize their followers and curate a sense of unity and belongingness, which leads their followers to shift from focusing on self‐interest to collective interest (De Cremer & van Knippenberg, 2002). Indeed, research studies have demonstrated that perception of charismatic leadership, via a heightened sense of belongingness, indirectly predicts follower cooperation (De Cremer & van Knippenberg, 2002), as well as helping and compliance behavior (Den Hartog et al., 2007). Perception of transformational leadership is also associated with followers’ innovativeness and affective organizational commitment (Zhu et al., 2013). In a study conducted by van Knippenberg and van Knippenberg (2005), they also demonstrated that when leaders engage in group‐oriented behaviors such as self‐sacrifice, followers perceive them to be more effective and in turn, attain higher performance. This effect is stronger for leaders with low prototypicality. Collectively, these findings support a central tenant of SITL––nonprototypical leaders must demonstrate their positive group‐oriented motives to convince their followers that they are trustworthy and effective in order to successfully guide them to strive toward a common goal (van Knippenberg, 2011).

In crises, leaders’ guidance is of particular importance in inspiring followers’ responsive behaviors (Bligh et al., 2004a, 2004b; Hohman et al., 2010). On April 3, 2020, CDC amended their original recommendation regarding mask‐wearing and urged the public to wear masks in public spaces. In his address, Trump announced CDC's mask guideline but repeatedly emphasized that it is voluntary (Cathey, 2020). Thereafter, he appeared in public events without wearing a mask and continued to violate behavioral guidelines set forth by CDC (Cathey, 2020). Trump served as a role model for his followers (Bandura, 1977; Brown et al., 2005) as Republicans were similarly less willing to wear facemasks in public (van Kessel & Quinn, 2020). Trump's administration also defended his and his followers’ rejection of CDC's safety guideline (at campaign rallies) as one's exercise of their First Amendment rights (Cathey, 2020). That is, Trump and his administration's emphasis on individual freedom once again appeals to his ingroup followers’ ideals, further affirming their identity during this crisis.

As Trump's rhetoric and behaviors contradict the Democratic party identity as well as their welfare, his outgroup followers are likely to continue to see him as an outgroup leader, less likely to view him as effective, and trust him and follow his suit, including deviance from the COVID‐19 safety guidelines. Indeed, Democrats are more likely to follow CDC's guidelines and they are largely concerned with individuals not wearing a mask in the public (Kerr et al., 2021; van Kessel & Quinn, 2020). We attribute this, partly, to their social group identification and perception of leader effectiveness. Therefore, we hypothesize that political affiliation will predict follower crisis management behaviors via perceived leader effectiveness as follows: Republicans are more likely to consider Trump to be managing the COVID‐19 pandemic effectively, and in turn, are less likely to engage in COVID‐19 preventive behaviors compared to Democrats (Hypothesis 3).

THE CURRENT STUDIES

In recent years, scholars have observed a wide division in terms of information networks among partisans (Brady et al., 2017). Partisans are particularly drawn to polarized news sources or partisan “echo chambers” which provide information that further affirms their attitudes and beliefs (Bakshy et al., 2015; Lau et al., 2020; Lelkes et al., 2017). Such a divided pattern of information consumption has led the American people to arrive at different conclusions and thus different emotions and behaviors regarding the pandemic (Van Bavel et al., 2020). Evidently, a recent study conducted by Gollwitzer et al. (2020) found that media consumption pattern was related to physical distancing (also see Hart et al., 2020). As the media both passively mirror and actively sway the mood, attitudes, and behaviors of the public (Bligh et al., 2004a), including in times of crisis (Althaus, 2002), studying media coverage can help explicate understandings of partisanship, followers’ perceptions of leadership, and their behavioral responses to COVID‐19.

Thus, we employed a quantitative content analysis and two survey studies to test the derived hypotheses. In Study 1, we compared the media outlets that are trusted by Republicans and by Democrats to provide some insight into how the mass media, as a contextual factor, fuels political polarization regarding COVID‐19 and leadership perceptions. In Study 2, we employed a cross‐sectional survey to compare Republicans’ and Democrats’ responses to the COVID‐19 pandemic and tested Hypotheses 1–3. To capture the evolving nature of affect and examine how follower responses unfold during the pandemic, for Study 3, we employed a longitudinal survey examining the same key variables as in Study 2 with samples of Republicans, Democrats, and Independents. Overall, our three studies provide a comprehensive picture of how follower social identification and leadership perceptions were shaped and evolved during the pandemic, and how they were associated with followers’ cognitive, affective, and behavioral reactions.

STUDY 1

In Study 1, analyzing news online, we seek to confirm the role of the mass media in contributing the polarization of perceptions of leadership and responses to COVID‐19.

Method

Sample

We referenced Pew Research Center's report on US adults’ media consumption pattern (Jurkowitz et al., 2020) and limited our sample to the most trusted and most distrusted news outlets by Democrats and Republicans in order to account for selective exposure and confirmation bias – phenomena that are prevalent in partisan news consumption (Knobloch‐Westerwick et al., 2017; Lelkes et al., 2017). CNN, MSNBC, NBC, NPR, New York Times, and PBS were determined as outlets trusted by Democrats, and ABC, Breitbart, FOX, The New York Post, and the Rush Limbaugh show represented news outlets trusted by Republicans.1 Using Nexis UNI and Media Cloud (an open‐source platform that tracks the contents of online news; see Media Cloud.org), we conducted a target keyword search for “COVID‐19 or coronavirus” AND “Trump” to assess all references to COVID‐19 from January 20, 2020 (the first COVID‐19 case in the United States), to April 11, 2020 within our target sample. We decided to focus on the early onset of the COVID‐19 outbreak in the United States because one essence of a crisis––uncertainty (Boin & Hart, 2003; Pearon & Clair, 1998) ––is amplified during this period due to the limited knowledge in existence regarding the COVID‐19 virus. To fully capture how leadership and crisis are portrayed in these outlets, we did not further specify any inclusion/exclusion criteria.1 The final sample consists of 4319 news articles and TV/Radio news transcripts (n = 2643; n = 1676).

Procedure and measures

DICTION 6.0 (Hart et al., 2012) and Linguistic Inquiry and Word Count (LIWC) 2015 (Tausczik & Pennebaker, 2010) were utilized to analyze media coverage. DICTION 6.0 is a content analysis software designed to analyze political discourse that has been used to study semantics in management as well as political psychology (Bligh et al., 2004a, 2004b; Lau et al., 2020). This program searches for unique words generated from a collection of dictionaries, known as word‐lists, to determine the tone of a passage. We followed Lau et al. (2020) and utilized the praise dictionary (e.g., dear, delightful, bright) to capture public perception of leader effectiveness. We also employed the ambivalence dictionary (e.g., perhaps, might, almost) to examine whether news outlets trusted by Republicans would convey a stronger sense of certainty than news outlets trusted by Democrats.

LIWC is a text analysis program that searches for words that constitute an array of psychologically meaningful categories (see Tausczik & Pennebaker, 2010). LIWC supplements DICTION 6.0 in that it is designed and has been shown to be adequate in capturing emotionality in texts (Moore et al., 2017). We utilized the positive emotion (e.g., love, nice, sweet) and negative emotion (e.g., hurt, ugly, nasty) dictionaries in LIWC 2015 (Pennebaker et al., 2003) to assess the affective tone of the media coverage. A custom dictionary was created using LIWC to assess the mentioning of COVID‐19 preventive behaviors. We referenced CDC (April 2, 2020) and World Health Organization's (April 2, 2020) guidelines on prevention measures and derived 13 verbs and nouns that best represent them (e.g., mask, sanitize, clean). See Appendix A for a detailed description of the operationalization of the constructs and the operating basis of the content analysis programs.

We conducted MANCOVAs to explore polarization regarding COVID‐19 and US leadership in media coverage. Subsequent main‐effect analyses were conducted while controlling for inflation of Type I error with a Bonferroni correction (Tabachnick & Fidell, 2013). The total word count for each passage and source of media (TV/Radio vs. print) were included as covariates.

Results

An overview of descriptive statistics, mean differences between news sources on each construct, univariate F statistics, partial eta‐squared, and Wilks’ lambda are provided in Table 1.

TABLE 1

Study 1: Mean comparisons of variables by media sources trusted by democrats and republicans

	Democrats		Republicans
Variables	M	SE	M	SE	Univariate F		Wilks’ Λ
		SE	M	SE		ηp 2
Praise	4.33	.06	5.05	0.07	69.46	.02	.97
Ambivalence	13.65	.11	12.55	0.13	41.96	.01	.97
Positive Emotion	2.03	.02	2.20	0.02	54.12	.01	.95
Negative Emotion	1.69	.02	1.46	0.02	64.68	.02	.95
COVID‐Prevention	.28	.01	0.21	0.01	91.55	.02	.95

Note: Praise and ambivalence scores were generated with DICTION. Since DICTION produces a score for each 500‐word passage, one single news articles might be represented by multiple scores (if word count > 500). Thus, variables scores calculated by DICTION and LIWC are associated with different sample sizes. For DICTION: n Dem = 2,647; n Rep = 2,056; for LIWC 2015: n Dem = 2,643; n Rep = 1,677. Means (M) and Standard Error coefficients (SE) presented are adjusted for the covariates.

All F‐values are significant at p < .001.

Results from the DICTION analysis indicated that the content was significantly different between news outlets trusted by Democrats (coded as 0) and by Republicans (coded as 1), F(2, 4698) = 62.95, p < .001, Wilks’ Λ = .97, ηp 2 = .03. Main effect analyses suggested that news sources trusted by Republicans, on average, had a higher score on praise, F(1, 4699) = 69.46, p < .001, ηp 2 = .02, and a lower score on ambivalence, F(1, 4699) = 41.96, p < .001, ηp 2 = .01, compared to news sources trusted by Democrats. Our power analysis using G*Power 3 (Faul et al., 2007) indicated that our sample of 4701 provides 80% power to detect effect as small as f 2 = .0021, 90% power to detect f 2 = .0027, 95% power to detect f 2 = .0033.

The LIWC analyses also indicated that the content was significantly different between news from Republican‐ and Democrat‐trusted news outlets, F(2, 4314) = 70.38, p < .001, Wilks’ Λ = .95, ηp 2 = .05. For affective tone, news outlets trusted by Republicans had a higher positive affective tone, F(1,4316) = 54.12, p < .001, ηp 2 = .01, and a lower negative affective tone, F(1, 4316) = 64.68, p < .001, ηp 2 = .02, in their coverage of Trump and COVID‐19 compared to outlets trusted by Democrats. Media outlets trusted by Democrats also mentioned COVID‐19 preventive behaviors more frequently compared to those trusted by Republicans, F(1, 4316) = 91.55, p < .001, ηp 2 = .02. Our post‐hoc power analysis using G*Power 3 (Faul et al., 2007) indicated that our sample of 4318 provides 80% power to detect effect as small as f 2 = .0025, 90% power to detect f 2 = .0033, 95% power to detect f 2 = .0040.

These findings suggest that news outlets were significantly polarized and that those trusted by Republicans portrayed Trump as more effective, conveyed a stronger sense of certainty, but placed a lower emphasis on COVID‐19 prevention compared to news outlets trusted by Democrats. Although the effect sizes were small, it is important to note that this is expected for a context‐free content analysis of the media content with such a large sample size (see Bligh et al., 2004a, Study 2; M. C. Bligh, personal communication, July 1, 2020).

The results demonstrate that the mass media plays a role in political polarization in the context of COVID‐19. Namely, news outlets trusted by Republicans portrayed US President Donald Trump as more effective, conveyed a stronger sense of certainty with less (more) negative (positive) affective tone, and had a lower emphasis on COVID‐19 prevention compared to outlets trusted by Democrats. Such differences in their rhetoric mirror the partisan differences regarding responses to COVID‐19 that have been captured in research (Funk & Tyson, 2021; van Kessel & Quinn, 2020). After establishing the active role of the mass media in polarizing perceptions of COVID‐19 and leadership, in the subsequent studies, we investigated how political polarization influences partisans’ responses to the COVID‐19 pandemic and the mediating role of perceived leader effectiveness.

STUDY 2

In Study 2, we employed a cross‐sectional survey to compare Republicans’ and Democrats’ responses to the COVID‐19 pandemic and tested Hypothesis 1 (political affiliation is related to perceived leader effectiveness), Hypothesis 2 (Democrats will consider Trump as less effective as a leader and experience higher levels of negative affect (H2a) and lower levels of positive affect (H2b) as compared to their Republican counterparts), and Hypothesis 3 (Republicans are more likely to consider Trump to be managing the COVID‐19 pandemic effectively, and in turn, are less likely to engage in COVID‐19 preventive behaviors compared to Democrats).

Participants and procedure

We recruited 296 participants on Amazon Mechanical Turk (MTurk, a crowdsourcing internet marketplace). Data were collected in April 2020. Participants filled out an online survey assessing perceived leader effectiveness, positive and negative affect, and engagement in COVID‐19 preventive behavior, along with demographic questions including age, gender, education, political affiliation, and socioeconomic status (SES). Respondents that did not identify as a Democrat or a Republican (n = 55), failed one or more of the attention‐check items (n = 17), or were multivariate outliers (n = 11) were excluded2. The final sample contains 214 participants (54.2% female; age: M = 41.30, SD = 12.95; 54.2% Democrats; 49.5% hold a bachelor's degree).

A part of this dataset was reported in Tse et al. (2021) to examine prosocial and hoarding behavior in different countries during the COVID‐19 pandemic (see Appendix B for data transparency table).

Measures

Study 2 survey items can be found in Appendix C.

Leader Effectiveness. We developed a 3‐item measure to assess perceptions of Trump's effectiveness in handling the COVID‐19 pandemic (α = .94). Participants indicated their evaluation of Trump's effectiveness on a 5‐point Likert scale ranging from 1 (Not at all) to 5 (A great deal). A sample item reads “To what extent do you believe that Donald Trump is effective in dealing with the COVID‐19 pandemic?”

Affect. Positive and negative affect were assessed with the Scale of Positive and Negative Experience (SPANE; Diener et al., 2010; α positive = .92; α negative = .92). This scale contains 12 items – six items to assess positive feelings (e.g., happy, contented) and six to assess negative feelings (e.g., sad, angry). Participants were asked to rate the frequency of experiencing these feelings in the past month on a 7‐point scale ranging from 1 (Never) to 7 (Always).

COVID‐19 Prevention Behavioral Intention. We also developed a novel measure that assesses respondents’ COVID‐19 prevention behavioral intentions (α = .84). We followed the same procedure as described in Study 1 and derived 11 behavioral items (e.g., wear a face mask/respirator, avoid going to public areas; CDC [April 2, 2020]; World Health Organization's [April 2, 2020]). Respondents were asked to rate their intention to engage in these behaviors during the pandemic on a 7‐point scale ranging from 1 (Extremely unlikely) to 7 (Extremely likely).

Method of analysis

Given the inclusion of newly developed items and the use of single‐sourced data, prior to hypothesis testing, we conducted a series of confirmatory factor analyses (CFAs) to address potential issues with reliability and single‐source bias (Podsakoff et al., 2012). First, we subjected the measurement model to a CFA including items for leader effectiveness, negative affect, positive affect, and preventive behavior. Corresponding scale items were loaded on perspective factors. The 4‐factor measurement model fit the data adequately (Kline, 2015), χ2 (293) = 640.96, p < .001, χ2/df = 2.19, CFI = .90, TLI = .89, RMSEA = .07, SRMR = .07 (standardized loading range: .41–.99; see Appendix C). Next, we compared our hypothesized measurement model to alternative models to address common method bias (Podsakoff et al., 2012). The hypothesized 4‐factor model demonstrated superior fit compared to the alternative models (model description and fit indices are detailed in Appendix D), providing some evidence that the data were not contaminated by single‐source bias (Podsakoff et al., 2012). Finally, we conducted a path analysis using the R package ‘lavaan’ to test Hypotheses 1 ‐ 3 (Rosseel, 2012; see Figure 1).

FIGURE 1

Study 2: Perceived leader effectiveness mediates the effect of political affiliation (Democrat coded as 0; Republican coded as 1) on negative affect, positive affect, and COVID‐19 prevention.

Values are standardized coefficients.

Model fit: χ2(5) = 12.95, p = .02; CFI = .97; TLI = .89; RMSEA = .09; SRMR = .07.

Note: Age and ratio of cases was included in the model as control variables but are not shown in the figure. Non‐significant paths (p > .05) are denoted by dashed line.

We dummy coded political affiliation (Democrat = 0; Republican = 1) and treated it as an exogenous variable. Age and ratio of new COVID‐19 cases in participants’ residing state on the date of study participation were included as covariates in the analysis.

The key variables’ bivariate correlations are presented in Table 2 and the path analysis results can be found in Appendix E. The results demonstrated that political affiliation was a significant predictor of perception of Trump's effectiveness, b = 1.94, 95% CI [1.68, 2.21]. Republicans (M = 3.49, SD = 1.08) perceived Trump as more effective in managing the pandemic compared to Democrats (M = 1.55, SD = .89). Thus, Hypothesis 1 was supported.

TABLE 2

Study 2: Zero‐order correlations for all variables (N = 214)

Variable	1	2	3	4	5	6	7	8	9
1. Political affiliation	––
2. Trump effectiveness	.70**	––
3. COVID‐19 prevention	−.24**	−.28**	––
4. Positive affect	.31**	.35**	−.08	––
5. Negative affect	−.20**	−.20**	.13	−.57**	––
6. New cases	−.01	−.03	.14*	−.04	.13	––
7. Gender	−.06	−.04	.18*	−.06	.15*	−.08	––
8. Age	.23**	.21**	.12	.07	−.19**	−.04	.17*	––
9. Education	−.10	−.11	.05	−.02	−.01	.05	−.07	−.02	––
10. SES	.22**	.24**	−.06	.31**	−.23**	.00	−.11	−.01	.35**

Note: Political affiliation: 0 = Democrat, 1 = Republican; COVID‐19 Prevention = COVID‐19 preventive behavior;.

new cases ratio = ratio of new COVID‐19 cases in the state in which the participant resides on the date of study.

participation to the population of that state. SES = socioeconomic status. Gender: 0 = men; 1 = women.

*p < .05, **p < .01.

Next, we examined the indirect effect between political affiliation and affect through Trump's effectiveness. The analysis showed that the indirect effect was non‐significant in explaining the relationship between political affiliation and negative affect (Republican: M = 3.08, SD = 1.50; Democrats: M = 3.63; SD = 1.20), b = −.20, 95% CI [−.55, .15]. Hypothesis 2a was not supported. However, the indirect effect was significant in predicting positive affect (Republicans: M = 4.56, SD = 1.15; Democrats: M = 3.80; SD = 1.20), b = .44, 95% CI [.13, .75]. Hypothesis 2b was supported. Democrats, compared to Republicans, perceived Trump as less effective in handling the pandemic, which was related to a lower level of positive affect, but not significantly related to their experience of negative affect.

The same analysis revealed a significant indirect effect between political affiliation and COVID‐19 prevention behavioral intentions via Trump's effectiveness, b = −.23, 95% CI [−.42, −.04]. In other words, Republicans, as compared to Democrats, perceived Trump as more effective, and this perception was negatively related to their COVID‐19 prevention (Republican: M = 6.15, SD = .90; Democrats: M = 6.51, SD = .53). Thus, Hypothesis 3 was also supported. Our post‐hoc power analysis using software MedPower (Kenny, 2017) indicated that our sample of 214 was adequate to detect a standardized mediation effect as small as .09 (assuming path a = path b = path c = .30) with α = .05 and achieved power = .99 (power = .98, .96, .91 when the standardized mediation effect = .08, .07, .06, respectively).

The differential effects on positive and negative affect align with the affect literature – positive and negative affect are often independent of one another (Diener & Emmons, 1984), as reflected by the two‐factor measure for positive and negative affect (SPANE; Diener et al., 2010). The non‐significant effect on negative affect may be explained by third variables that were not captured in Study 2. For example, Larsen and Ketelaar (1991) have found that in an unpleasant situation, neuroticism significantly and positively predicts negative affect but not positive affect. In Study 2, it is possible that neither political affiliation nor perception of leader effectiveness were strong predictors of negative affect compared to dispositions such as one's personality trait (e.g., neuroticism).

Relatedly, previous research has differentiated between event‐specific, contextual‐specific (e.g., Sanford, 2012), and global affect (Diener et al., 2010). Our measure of affect, SPANE, captures the respondents’ “general feelings” (Diener et al., 2010; p.145), which are more susceptible to the influence of traits/dispositions than situation‐specific emotional responses (Cropanzano & Dasborough, 2015; Huppert & So, 2013). This further supports our speculation that a third variable (potentially trait/dispositions) may have obscured any effect on negative affect. Tracking affect that are event‐specific longitudinally, then, may provide us further insight into the effect of social identification and perceived leader effectiveness on affect. Study 3 was thus designed to address these limitations and to bolster our findings.

STUDY 3

To capture the evolving nature of affect and examine how follower responses unfold during the pandemic, for Study 3, we employed a longitudinal survey examining the same key variables as in Study 2 with samples of Republicans and Democrats.

We conducted a three‐wave longitudinal study, with data collection taken place in late March (Wave 1; N = 1001), late May (Wave 2; N = 511), and early October 2020 (Wave 3; N = 475). Wave 1 and Wave 2 data belong to a larger project that aims to track the rapid change in Americans’ and Chinese’ response to the COVID‐19 pandemic (see Chan, Wang, et al., 2021 and Appendix B for data transparency table). In total, 359 participants completed all three surveys, with an attrition rate of 31.31% from Wave 2 to Wave 3. For the purpose of the present study, we included participants who self‐identified as Democrats (n = 142) and Republicans (n = 109) only.

Similar to Study 2, all participants were from MTurk. Participants averaged the age of 39.55 years old (SD = 13.56). The majority were female (59.60%), and bachelor's degree holders (42.40%). At each time point, participants reported their perception of Trump's effectiveness, levels of negative and positive affect, and frequency of engagement in COVID‐19 preventive behavior. Additionally, at Wave 1, participants reported their political affiliation (Democrat, Republican, Independent, or none of the above) and demographic information (gender, age, education level, and annual household income).

We conducted an attrition analysis to examine differences between participants who completed all three waves and those who dropped out. Our analysis revealed that participants who responded in all three waves of data collection reported lower levels of negative affect (Hedge's g = −.21) and less frequent COVID‐19 preventive behavior (Hedge's g = −.13) compared to those who did not (Hedge's g = .22). There was no significant difference in negative affect and COVID‐19 preventive behavior between the two groups at Wave 2. There were also no significant differences in perceived effectiveness and positive affect between the final and excluded samples at both Wave 1 and Wave 2. Details of the attribution analysis can be found in Appendix F.

Leader Effectiveness. Two items were adopted from Chan et al. (2021b) to assess perceptions of Trump's effectiveness in handling the COVID‐19 pandemic (α ranged from .95 to .97). Participants indicated their evaluation of Trump's effectiveness in managing the COVID‐19 crisis on a 7‐point Likert scale ranging from 1 to 7. A sample item reads “Do you approve or disapprove of the way Donald Trump is handling the response to the COVID‐19 outbreak?”

Affect. To capture affect that are more likely to be impacted by the COVID‐19 pandemic (event‐specific), we referenced sentiment analyses regarding the pandemic on social media platforms (Khattar et al., 2020; Stella et al., 2020; Venigalla et al., 2020) and used three items (optimistic, calm, hopeful) to assess positive affect (α ranged from .87 to .89) and five items (afraid, anxious, threaten, tense, disgust) to assess negative affect (α ranged from .91 to .92; also see Chan et al., 2021b). Participants were asked to rate the frequency of experiencing these feelings in the past month on a 7‐point scale ranging from 1 (Not at all) to 7 (Extremely).

COVID‐19 Preventive Behavior. We adopted Chan, Wang, et al.'s (2021) measure of COVID‐19 Preventive Behavior (α ranged from .83 to .91). Their measure contains 15 COVID‐19 preventive behaviors that were generated based on the recommendation from CDC and those recommended in countries that had earlier outbreaks (e.g., China; see also Chan et al., 2021b for a description of this scale). Respondents were asked to report the frequency of their engagement in these behaviors in the past week on a 5‐point scale ranging from 1 (Never) to 5 (Always).

We followed the same procedure in Study 2 to determine the validity of our measurement model. See Appendix G for standardized factor loadings for the hypothesized 4‐factor measurement model and Appendix H for measurement models comparisons.

Similar to Study 2, we conducted path analysis to test the hypotheses (see Figure 2) using the R package “lavaan” (Rosseel, 2012). To test the effect of party affiliation, we created a dummy variable – party affiliation (Democrat = 0 and Republican = 1). We tested the longitudinal effect of party affiliation on perceived Trump's effectiveness at Wave 2 by controlling the effect of perceived Trump's effectiveness at Wave 1. Similarly, we tested the longitudinal effect of perceived Trump's effectiveness at Wave 2 on the three outcome variables at Wave 3 by controlling each outcome variable at Wave 2, respectively. We included demographic variables as covariates in predicting perceived Trump's effectiveness at Wave 2 and the three outcome variables at Wave 3.

FIGURE 2

Study 3: Illustration of the longitudinal relationship between Wave 1 political affiliation (Democrat = 0; Republican = 1), Wave 2 perceived leader effectiveness, and Wave 3 COVID‐19 related outcome variables.

Note. We controlled for the effect of Wave 1 perceived leader effectiveness in predicting Wave 2 perceived leader effectiveness. We also controlled for Wave 2 COVID‐19 related outcome variables in predicting each Wave 3 outcome variables respectively. Demographic variables were included as covariates in the model. Dotted lines denote paths that are not statistical significance (p > .05).

The results of path analysis (see Table 3) showed that political affiliation at Wave 1 predicted perceived Trump's effectiveness at Wave 2 (controlling for Wave 1 Trump effectiveness, b = 1.03, SE = .19, p < .001, 95% CI = [.65, 1.41]). Compared to Democrats, Republicans perceived Trump as more effective in managing the pandemic. Hypothesis 1 was supported.

TABLE 3

Study 3: Longitudinal relationship between political affiliation, Trump effectiveness, and COVID‐19 related outcome variables (N = 251)

	Wave 3 Disease‐preventive behavior			Wave 3 Positive affect			Wave 3 Negative affect
	b (SE)	p‐value	95% CI	b (SE)	p‐value	95% CI	b (SE)	p‐value	95% CI
Wave 1 variables
Republican	−.10 (.09)	.297	[−.28, .09]	.13 (.21)	.521	[−.26, .55]	−.42 (.21)	.045	[−.83, .01]
Wave 2 variables
Trump effectiveness	−.05 (.02)	.016	[−.09, −.01]	.14 (.05)	.005	[.05, .25]	−.05 (.05)	.305	[−.14, .05]
Disease‐preventive behavior	.80 (.05)	.000	[.71, .90]
Positive Affect				.62 (.05)	.000	[.52, .72]
Negative Affect							.71 (.05)	.000	[.62, .81]
Indirect effect via Wave 2 Trump effectiveness
Republican	−.05 (.02)	−	[−.10, −.01]	.14 (.06)	−	[.05, .29]	−.05 (.05)	−	[−.15, .06]

Note: Republican was positively related to Wave 2 Trump effectiveness (b = 1.03, SE = .19, p < .001, 95% CI = [.65, 1.41]; controlled for Wave 1 Trump effectiveness). Demographic variables (gender, age, education, and income) were included as covariates in the analysis. Indirect effects were estimated using bootstrapping method. Model fit: χ2 = 136.94, df = 43, p < .001; CFI = .944, TLI = .903; RMSEA = .087, SRMR = .113; AIC = 8385.74, BIC = 8576.39.

The same analysis revealed that perceived Trump's effectiveness at Wave 2 was related to lower levels of negative affect and higher levels of positive affect at Wave 3, controlling for the effect of negative affect and positive affect at Wave 2, respectively. Importantly, we found a significant indirect effect between political affiliation at Wave 1 on positive affect (indirect effect = .14, 95% CI = [.05, .29]) but not negative affect (indirect effect = −.05, SE = .05, 95% CI = [−.15, .06]) at Wave 3 via perceived Trump effectiveness at Wave 2. Compared to Democrats, Republicans had a higher perception of Trump's effectiveness at Wave 2, which was linked to a higher level of positive affect at Wave 3. Hypothesis 2a was not supported but Hypothesis 2b was supported.

Lastly, our findings revealed that there was a significant indirect effect of political affiliation on COVID‐19 preventive behavior via perceived Trump's effectiveness (indirect effect = −.05, SE = .02, 95% CI = [−.10, −.01]). Compared to Democrats, Republicans were more likely to perceive Trump as effective at Wave 2, which negatively predicts their COVID‐19 preventive behavior at Wave 3. Hypothesis 3 was supported. The results from Study 3 replicate that of Study 2.

Our post‐hoc power analysis using R package “PowerUpR” (Bulus et al., 2021) revealed that the sample size was adequate to detect a 2‐1‐1 level‐1 (lower‐level) standardized mediation effect as small as .09 (assuming path a = path b = path c = √.09 = .30, path b1 = path b2 = (path b) / 2 = .15, intraclass correlations for level 1 variables = .20, proportion of variance in the outcome explained by level‐1 and level‐2 covariates = .50) with α = .05 and achieved power = .93 with a Monte Carlo interval test (power = .91, .86, and .79 when the standardized mediation effect = .08, .07, and .06, respectively).

GENERAL DISCUSSION

How do social identity, leadership, and context interact to shape follower response in a crisis? We turned to the American people during the COVID‐19 pandemic to seek an answer to this question. Across three studies, we demonstrated that amid a global health crisis, political affiliation undermined Americans’ ability to unite in terms of their attitudes regarding leadership, their affective responses, and their behaviors. Namely, echoing and extending recent research on political polarization in the COVID‐19 context (Chan et al., 2021b; Gollwitzer et al., 2020; Kerr et al., 2021; Peacock & Biernat, 2021), we found that Republicans held a stronger belief in Trump's ability to handle the pandemic effectively and, in turn, were less likely to engage in COVID‐19 preventive behaviors but maintained a higher (lower) level of positive (negative) affect compared to Democrats. Contrary to previous research on crisis and leadership (Bligh et al., 2004a, 2004b; Boin & Hart, 2003; Drury et al., 2009; Flade et al., 2019; Gaertner et al., 2000; Pearson & Clair, 1998), we found that followers of different subgroups did not unite in a time of uncertainty and threat. Such a discrepancy may be fueled by contextual factors such as the media (e.g., Lau et al., 2020; Rosenfeld & Tomiyama, 2021) and the intergroup leader's agenda and leadership (Pittinsky & Simon, 2007), further supporting the use of intergroup relations lens in the study of leadership.

Previous research on social identity and leadership suggests that social categorization predicts perceived leader effectiveness (Duck & Fielding 1993, 2003; Platow & van Knippenberg, 2001; Platow et al., 2006). In the current studies, we found that this effect persists even in the presence of a crisis. Its prevalence is likely owed to the hyper political polarization in the United States (Boxell et al., 2020). According to SITL, when group identity is particularly salient, we can expect a strengthened effect of social identity on the leadership process (Hogg, 2001; van Knippenberg & Hogg, 2003). In an intergroup context, the effect of social identity may be exacerbated or attenuated by leadership (Gleibs & Haslam, 2016). As Trump focused on reaffirming the identity of the Republican party and promoting his representation of the group (Reicher & Haslam, 2017), we found that his Democratic followers continued to consider him as a nonprototypical leader and were less likely to trust his leadership in handling the pandemic. In alignment with research on intergroup as well as political leadership (Bligh et al., 2004a; Lau et al., 2020; Oberschall, 2000), results from Study 1 also confirm the role of the mass media in potentially furthering polarization and shaping partisans’ differential perceptions of leadership and response to COVID‐19.

Our findings provide partial support to propositions regarding social identity and follower affect put forth by Tee et al. (2013). We reasoned that Republicans tend to perceive COVID‐19 as a lesser threat and Trump mirrored the group's sentiment. This may lead the Republicans to consider Trump as effective and have different affective experiences compared to the Democrats. Our findings revealed that perceptions of leader effectiveness partly explain the differential positive (but not negative) affective response between Republicans and Democrats.

People's behavioral responses are correlated with COVID‐19 contraction rates (Eikenberry et al., 2020). In the current studies, we found that when Trump is perceived as effective, his followers are more likely to engage in behaviors that he endorses, albeit against CDC's COVID‐19 guidelines. This aligns with previous research that examines the relationship between leader effectiveness and follower behaviors (De Cremer & van Knippenberg, 2002; Den Hartog et al., 2007; van Knippenberg & van Knippenberg, 2005). Evidently, although Democrats’ COVID‐19 contraction rates and death rates were higher than that of Republicans at the beginning of the crisis, this pattern shifted as the pandemic progresses (Jones & Kiley, 2020). In this case, follower deviance of the leader's direction may be considered as courageous followership (Chaleff, 2010). In these scenarios, followers who consider the leader as unethical (Carsten & Uhl‐Bien, 2013) or in this case, ineffective, take it upon themselves to strive toward their own goal. This observed effect demonstrates the agency of followers in the leadership process (Uhl‐Bien et al., 2014).

Our work contributes to the literature on social identification, crisis leadership, and follower‐centric perspectives of leadership. First, our study demonstrates that social identity links to perceptions of leaders in extreme circumstances such as a pandemic, thus contextualizing leadership processes and increasing our understanding of follower reactions toward leaders during crises. Beyond leader evaluation (van Knippenberg & Hogg, 2003; Zhu et al., 2013), our results show that social identification also associates with followers’ behaviors and affect during crises. This enriches our understanding of not just the process of leadership but also its outcomes (Duck & Fielding, 2003, 2007; Uhl‐Bien et al., 2014) during crises. Second, situating our study in the US political context provides a realistic account of the complexities of intergroup membership. As intergroup leadership is a ubiquitous phenomenon in organizations (Hogg et al., 2012a; Pittinsky & Simon, 2007), our results highlight the importance of accounting for subgroup membership in the study and practice of leadership as well as crisis management. Third, our findings demonstrate followers’ behaviors and their relationships with the leadership process and outcomes in a crisis situation, extending the scant research in this area (Carsten et al., 2019; Uhl‐Bien et al., 2014). Finally, the current studies demonstrate the value of content analysis in uncovering the leadership process and as an alternative and under‐utilized research method to understand followers’ affect and cognition. The employment of this naturalistic method may strengthen the generalizability and validity of findings of leadership research.

Limitations and future directions

This study, like any other, is subject to limitations. Given the non‐experimental nature of the current studies, definitive insights into the directionality of the relationships identified are not possible. In Study 1, we were not able to showcase the causal relationship between media coverage and followers’ responses. In Studies 2 and 3, we did not capture potential confounding effects of third variables, such as personality traits, on followers’ affect and behaviors. Further testing with experimental studies controlling for dispositional factors is needed to replicate the current findings.

There are some limitations with the measures that we adopted for the current studies. First, while our measure of preventive behavior mostly maps to the self‐regulation behavior of avoidance/prevention (e.g., “avoid going to public areas,” and “avoid physical contact with others [e.g., handshakes]), we did not measure more approach/promotion” oriented behaviors related to COVID‐19 (i.e., movement towards healthier states vs. avoiding contagion). However, it is possible that different identities may be associated with different types of approach/promotion and avoidance/prevention health behaviors. Previous research found that political conservatism is associated with avoidance motives and political liberalism with approach motives (Janoff‐Bulman et al., 2008), thus there may be another approach/avoidance theme underlying the identity‐based constructs and relationships in our model, presenting an exciting opportunity for future research. Second, we used different affective measures in Studies 2 and 3. While the examination of different positive and negative affective outcomes allowed us to hone‐in on threat‐related affect, it limits our ability to compare the results from the two studies. Third, to capture political leadership in the context of the COVID‐19 crisis, we adapted, rather than applied directly, measures of leader effectiveness from studies that use the SITL framework (Pierro et al., 2005; van Knippenberg & van Knippenberg, 2005). While our measure for leader effectiveness is ecologically valid and suitable for use in this unique context, we acknowledge that the adaptation may pose a question about the comparability of our study findings to previous SITL research,

Our studies treat follower behavior as a dependent variable. Nonetheless, according to Uhl‐Bien et al. (2014), followers’ behaviors and characteristics can impact leaders’ perceptions and behaviors in dynamic and cyclical ways (also see Carsten & Uhl‐Bien, 2017). Future research may investigate how differences in subgroup members’ attitudes toward their leader and their behaviors influence the leader's subsequent decisions during exogenous shocks, resulting in feedback loops and spirals.

Leadership involves the interplay between “specific leadership strategies, prevailing intergroup relations, and follower reactions” (Gleibs & Haslam, 2016, p. 558). Given different leadership and group characteristics, the leadership process shall thus unfold differently. For future research, scholars may investigate the interplay of these factors in other countries with different political contexts, cultures, and leaders, as recent studies demonstrated the differential response to COVID‐19 across countries and leadership (Chan et al., 2021b; Tse et al., 2021). For example, compared to countries with an individualistic culture, people of highly collectivistic cultures might be more likely to unite in times of crisis regardless of partisanship.

Practical implications

Whereas our studies focus on political affiliation, the implications can be applied to different intergroup contexts such as interdisciplinary teams and cross‐cultural organizations. Our findings illustrate that followers do not unconditionally look to their leaders even in times of crisis. To effectively manage crises, leaders, thus, need to have a good understanding of followers’ identifications, subgroup patterns, and how to construct a crisis interpretation and contingency plan that can mobilize members of different subgroups. Research on intergroup leadership presents some strategies to manage subgroups and promote intergroup relations. In addition to embracing transformational leadership approach, Hogg et al. (2012a) have discussed the importance for leaders to construct intergroup relational identity in a multi‐group context such that subgroup members are acknowledged as distinct but valued parts “entwined in an essential mutually collaborative relationship” (p. 240; also see Rast et al., 2018). Thus, leaders managing subgroups shall engage in tactics that highlight the uniqueness of the groups, their similarities with each group, and accentuate the interdependency of these groups.

Our findings also suggest that rhetorical construction of crisis is associated with how followers interpret and respond to crises, and that polarized perceptions of leadership may associated with negative consequences. We found that Republicans had lower engagement in crisis response behaviors compared to Democrats; they experienced, however, a high level of positive affect, which is important for resilience during an organizational crisis (Sommer et al., 2016). This suggests that despite their lack of responsiveness in terms of preventive behavior, they experienced a sense of positivity which is critical for survival. Whereas further evidence is needed for this claim, we suggest that leaders should, in their rhetoric, focus on both the behavior and feelings of their followers. For example, they may convey the severity and urgency of the crisis to mobilize their followers. As this fight against the pandemic continues, it is crucial for leaders to mobilize followers to engage in appropriate, responsive behaviors. In doing so, they may provide factual and up‐to‐date information to minimize anxiety elicited by the uncertainty of the situation. At the same time, however, leaders also need to instill a sense of optimism in order to provide guidance and a sense of hope for followers in times of crisis.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.

Online Appendix

Click here for additional data file.