How Do Partisans Navigate Intra-Group Conflict? A Theory of Leadership-Driven Motivated Reasoning

Abstract

When faced with co-partisan politicians who disagree publicly, what side do partisan voters take? We draw on social identity theory to develop a theory of partisan leadership cues arguing that leaders have a key role in social groups and because of that centrality, and accounting for affect-based motivation, co-partisan voters resist ingroup dissent. We test this theory with a series of experiments focused on leaders who violate democratic norms and responses from within the party that reflect loyalty or dissent. Our findings show that co-partisan voters are loathe to punish misbehaving leaders, except when their action represents a major threat and the criticism comes from a high ranking party member. Ingroup critics of the leader risk their own reputation in the process. Importantly, leadership effects occur even in fictitious partisan contexts when partisans have no prior affect for a leader or critic. Our findings point to the power of party leaders in groups and raise questions about the prospects for democratic criticism and accountability.

When faced with politicians who disagree publicly, how do voters decide which side to take? In the U.S. polarized environment, theories rooted in partisan motivated reasoning suggest that partisan attachments will guide citizens’ evaluations (e.g., Druckman et al., 2013; cf., Groenendyk & Krupnikov, 2021). But what if there is public disagreement within the party? In such situations, where partisan cues are meaningless, co-partisan voters may use other criteria to take sides. Such criteria could be the gender or race of those in each of the camps (e.g., Holman et al., 2021). We offer another characteristic that may motivate co-partisan responses: leadership position within the party.

In recent decades, two presidents, one Democratic and one Republican, were impeached, the highest form of political criticism in the American system, both with some support from co-partisans. And criticism of leaders, outside the scope of impeachment, has been even more common. Criticism from within the party, and against the leader, is thus a relevant consideration for voters.

Extant perspectives provide conflicting answers to how partisans should navigate intra-group conflicts. Costly credibility theory argues that criticism of the leader from within the party—which is costly and credible—should lead to declines in leader favorability among co-partisans (Groeling, 2010). This perspective implies that co-partisan critics may benefit as well, or at least not be penalized. Yet, partisans in contemporary politics continue to support their party’s candidates even when the candidates violate democratic norms or the voter’s own moral convictions (Walter & Redlawsk, 2019) and may even increase their support in the face of negative information (Redlawsk et al., 2010). This suggests that shifting attitudes about party elites may be difficult, even if the impetus comes from within the party.

In a theory developed both inductively and deductively, we argue that partisans make decisions on intra-party conflicts by using the leader as a cue. According to our leadership-driven motivated reasoning theory, leaders play a special role within groups because they are yardsticks to judge distance from the out-group and one’s own group standing (e.g., Hogg 2001). Group members are motivated by in-group favoritism (e.g., Mason 2018) and a desire to maintain psychological distance from the other party (Iyengar et al., 2019). Attachment to the leader facilitates both goals so partisans have incentives to be more attached to the leader than other members. Thus, a voter’s response to the leader is not simply a reflection of perceptions of the leader qua individual but qua party leader.

Given the psychological centrality of leaders within parties, partisan voters may be motivated to use the structural position of the executive, not just prior affect, as their cue in deciphering intra-party conflicts. Our predictions extend the expectations of partisan motivated reasoning (Lodge & Taber, 2013). We argue that affect for the leader is, in part, endogenous to her structural position, and that when partisans are called on to choose sides in an intra-party conflict, they will generally side with the leader, even when the leader misbehaves. Contrary to the expectations of costly credibility theory, a co-partisan critic is likely to face penalties.

While we argue that a leader’s unique position in parties will drive partisans to not penalize them in the face of co-partisan dissent, we expect limits on the power of leaders. We inductively explore the potential boundaries of the partisan leadership effect by analyzing experiments that differ in the seriousness of the leader’s transgression; and/or the status of the dissenter within the party. We expect that it may take very serious transgressions, combined with dissent from high-status party members, for co-partisan voters to reduce their support for the executive.

We test the implications of our theory of partisan leadership cues with experiments focused on co-partisan voters’ responses to intra-party dissent against a norms-violating leader.¹ Some involve a real leader (Trump) while others focus on fictitious vignettes. The hypothetical vignette experiments allow us to more clearly establish a leadership-driven effect relative to an affect effect. The vignettes also allow us to test our theory with Democratic voters. Our results suggest that criticism of the norms-violating leader from a fellow partisan is often ineffective. Our comparison of two levels of seriousness suggests that dissent may produce strengthening rally-around the leader effects. Additional experiments tentatively suggest that only a very serious violation with a high status dissenter may reduce support for the leader. At the same time, the dissenting co-partisan appears to gain little from his criticism. These findings suggest that the power of leaders within parties may not simply be accounted for by affective ties to the individual but amplified by the position itself.

How Partisan Voters Respond to Conflict

Disagreement is central to politics. We focus on public intra-party disagreements. When partisan voters are confronted with intra-party division, how do they make sense of the disagreement, and whom do they reward or punish?

The costly credibility theory (CCT) argues that partisans respond to intra-party criticism of leaders by viewing them less favorably. Groeling (2010, 130) argues that because in-group criticism of the leader is costly, co-partisan voters view this behavior as credible and adapt their views accordingly. Credibility depends on whether the voter believes that she and the critic have common interests (i.e., protect the party), or if the critic is self-interested and opportunistic (Groeling, 2010, p. 127–128; also see our discussion of SIT below). CCT also implies the possibility that co-partisan critics may be viewed favorably in the eyes of co-partisan voters as truth-tellers. Compelling as CCT may be, there are several limitations to applying it to contemporary intra-party conflicts.

First, Groeling’s study captures the beginning of polarization, but not the recent hyperpolarization era in which increased party homogeneity among elites makes it easier for voters to differentiate between sides (McCarty, 2019).² This has encouraged voters to sort and align their partisanship and ideology (e.g., Levendusky 2009), and express stronger attachment to parties and negativity towards the opposition (Abramowitz & Webster, 2016; Mason, 2018).

Second, the study doesn’t tackle normative violations, which is our focus (explanation below). The two presidents most relevant for norms violations are Clinton and Trump. While their norms violations and the co-partisan critiques differ, together they illustrate how criticism, even from within the party, didn’t dramatically shift leader evaluations. Groeling includes Clinton but doesn’t analyze the case separately. Clinton’s troubles started in January 1998 when he testified about the Lewinsky affair; he was impeached in December 1998. As Fig. 1 shows, Gallup data suggest that Democratic support for Clinton didn’t decline in his second term.³ His co-partisan approval rose from around 70% to more than 80% across his presidency despite the impeachment. Similarly, Republican support for Trump trended upward throughout his term despite two impeachments. Yet, in both cases, the accusations of misbehavior included co-partisan critics. In Clinton’s case, 31 Democrats voted for an impeachment inquiry and five Democrats voted to impeach. While no House Republican voted to impeach Trump in 2019, ten Republicans did so in 2021. One Republican Senator in 2020 and seven in 2021 voted to convict Trump. Even among those who didn’t vote to impeach, several were critical. These cases are important because impeachment is the most serious challenge to a presidency. Yet, this costly and credible criticism didn’t move co-partisan public opinion against the offending leader in these two cases.⁴

Fig. 1

Presidential Favorability among Co-partisans (Gallup Data)

Third, aggregate patterns don’t speak to how movable individual opinion may be. Recent experiments using individual data suggest that partisans are likely to support leaders even when leaders offend the voter’s moral convictions (Walter & Redlawsk, 2019). Moreover, negative information about preferred candidates may lead to rally effects (Redlawsk et al. 2010). While this research does not speak to co-partisan criticism of leaders, it suggests that shifting co-partisan support for the leader may be harder than previously theorized.

Finally, contrary to the implicit expectations of CCT, co-partisan voters appear to punish rather than reward co-partisan critics of the leader. Consider the case of Senator Romney. In the 2018 Republican primary, Romney secured 71% of the Utah vote. However, in late 2019, Romney emerged as a key Trump critic. Poll data show that in March 2020, Romney’s support was 29% among strong Republicans and 53% among weak Republicans. By comparison, favorability for fellow Utah Republican Senator Mike Lee, who didn’t criticize Trump, was 64% among strong Republicans and 59% among weak Republicans. Romney continued his criticism of Trump and by October 2020, Romney’s favorability was 16% among strong Republicans and 45% among weak Republicans. By comparison, Lee’s favorability was 77% and 56% respectively. There are limits to the conclusions we can draw from polls, but the magnitude of the decline in Romney’s favorability is illustrative. These data suggest that partisan voters may punish rather than reward co-partisan critics of the leader.

How do we account for the contrast between the general expectations of CCT and these examples that suggest that in cases of co-partisan dissent, voters may penalize the critic and not the offending leader? One explanation is affective motivated reasoning: partisans penalized Romney’s criticism because they liked him less than Trump and they rationalized their response.⁵ While certainly important, we argue that affective attachments are not the only influencer. Another explanation is leadership-driven motivation: partisans gravitated to Trump because he was the leader; an effect that may combine with prior affect but is theoretically independent of it.

Leadership-Driven Motivated Reasoning

We argue that when party officials publicly disagree with each other, more often than not, partisan voters will side with the leader. In part, this may reflect stronger affect for the person, but it is not simply a story about partisan affect. As with most social organizations, political parties are organized hierarchically with visible leadership roles. As social groups, parties generate strong attachments among members and firm boundaries with out-groups (Steffens et al., 2015). This partisan attachment is rooted in greater sorting, and improved alignment of various other identities (Levendusky, 2009; Mason, 2018). This attachment is also reflected in disdain for the opposing party (emotional distancing) (Iyengar et al., 2012).

Leaders personify the essence of the party: partisans think of them as combining the positive traits and qualities of the group and, by extension, their own self. Because of this, there is an incentive for group members to identify strongly with group leaders as this ensures maximum psychological distance from out-groups (Steffens et al., 2015). Leaders become the yardstick against which group members determine their own and other group members’ group standing and normative distance from out-groups (Steffens et al., 2015; Hogg, 2001). The result is that members are loyal to leaders and perceive them as representative of the group identity (Giessner et al., 2009), which, in turn, increases group attachment (Steffens et al., 2014; Steffens et al., 2015).

In a political organization, the position of a person within the party signals their status and thus their representativeness. The media visibility of leaders further reinforces their centrality (Rosenthal, 2013). The leadership cue can work to activate motivated processes in two ways. First, the leadership signal can boost affective attachment to the leader. For example, studies show that views of representativeness and affect for party leaders changes as a function of electoral victory or loss (Alabastro et al., 2013). Second, the leadership cue can work independently of affective attachment. As partisan-motivated reasoning theory suggests, prior positive affect should nudge partisans to side with the leader in an intra-group conflict. But even if they don’t like the leader much, or have no priors (as for a fictitious leader) the need to protect the in-group will incentivize partisans to take the leader’s side over a co-partisan critic. By contrast, signals of loyalty from within the party cement the leader’s centrality to the group and may mitigate concerns about the leader’s behavior. Therefore, motivated reasoning and rationalization of misdeeds can be triggered not only by affect but also by hierarchical standing.

This may be especially the case in a polarized environment because polarization incentivizes harder group boundaries. Especially in the U.S., threats to the party may be magnified because of the zero-sum structure of electoral politics (Mason, 2018). Although parties continue to represent multiple viewpoints and moderate partisans may be less happy with the stance of co-partisan elites (Groenendyk et al. 2020), differences between the parties are the dominant political cleavage in elite discourse and legislative voting patterns (McCarty, 2019) and structure social identities among the mass public (Mason, 2018). Furthermore, if partisans follow partisan-leaning media (Jacobson, 2015) and executives receive “the lion’s share” of media attention (Rosenthal, 2013), this may reinforce both partisan attachments and leadership-driven motivated reasoning. Experimental evidence in polarized environments hints at the centrality of leaders: policy cues from Trump shifted Republican voters’ issue stances, but cues from congressional Republicans were not effective (Barber & Pope, 2019). This loyalty carries over to normative issues; partisans don’t update their beliefs about their party even when their leaders contravene important norms (Walter & Redlawsk, 2019). In fact, negative information about co-partisan leaders may motivate followers to rally around their standard-bearer (Redlawsk et al., 2010).

In this context, partisans may perceive in-party criticism of the leader as a threat, motivating them to reaffirm their own good standing by dismissing the critic. Thus, contrary to CCT, we expect that co-partisan dissent should be ineffective at reducing leader support, or could produce a rally effect, while the critic may be punished.

Boundary Conditions

Does this mean that once a politician captures a leadership position she can behave with complete electoral impunity as far as party voters are concerned, even when others in the party publicly criticize her actions? Partisan voters may tolerate and rationalize many contra-normative behaviors by leaders. However, taken to the extreme this would suggest that partisan voters will support a leader even if she destroys the party. We suggest that leaders have wide, but not unlimited, latitude for action. Yet, especially in a dynamic environment with deepening affective polarization, it is an empirical question where the tipping point may be. The tipping point could be affected by the seriousness of the norms violation and thus the threat to the party (our focus) and/or by the amount of negative information about the leader growing over time (Redlawsk et al., 2010). We consider two features that may limit the pull of partisan leaders in the context of norms violations: the nature of the transgression itself and the status of the co-partisan critic.⁶

When partisans perceive that the threat from the leader’s behavior is greater than the cost of withdrawing support, they may turn against a leader. Thus, when a norms violation by the leader is very serious, and the leader’s behavior constitutes a threat to the party, it may produce negative emotional responses among co-partisan voters. Dissonance could set into motion mechanisms (e.g., information seeking) that make the voter open to following the dissenter. Clearly, at the extreme, some violations may force many, if not most, co-partisans to turn against the leader. For example, conviction for murder should turn-off co-partisan voters. Or if the leader decides to disband the party and oust its key members, voters should penalize him. However, outside some very extreme cases, during a leader’s tenure in office, it is difficult to a priori determine the tipping point when the partisan public may withdraw support. Moreover, whether a norms violation is viewed as major is a subjective decision and likely shaped by co-partisan media coverage (DellaVigna & Kaplan, 2007). A behavior viewed as a serious norms-violation at one point in time, or in the abstract, may be cast differently by opinion leaders focused on inter-party conflict.

Although we focus on executive branch leaders as the primary party leader, the U.S. separation of powers structure creates additional partisan leadership positions within the legislative branch. Legislative leaders have higher status than individual rank-and-file legislators. Perceptions of legislative leaders’ high-status may be further reinforced by media (Vinson 2018).

This status differential may be important to how co-partisans respond to criticism of the executive leader. Restoring the positive image of the group and reinforcing boundaries are key motivations for members in how they perceive dissent (Hutchison et al., 2008), but this may be moderated by the status of the dissenter (Marques & Paez, 1994). Dissent by a prominent member on the grounds of protecting the group is more likely to be perceived positively. Moreover, a high-status critic is less likely than a low-status critic to be reclassified as “deviant” and thus be penalized (Jetten & Hornsey, 2014). Therefore, criticism from high-status co-partisans may be more effective at reducing leader support, particularly if the leader’s transgression is serious. Yet when the leader’s transgression is less serious, even high-status co-partisans may be limited in pulling voters away from the leader.

By contrast, low-status dissenters may be dismissed or penalized by a co-partisan public that perceives them as “black sheep” who embarrass the party (Marques & Paez, 1994). Given a leader’s role in a social group, criticism by a low-status member can be interpreted as disloyal. Regardless of the seriousness of the leader’s transgression, dissent from a low-status co-partisan may be ineffective.

Case Selection and Justification

Our argument that party leaders provide a valuable cue to voters applies broadly to co-partisans’ responses to any type of conflict. However, our experiments focus on violations of democratic norms. This is an intentional choice. Although research suggests that partisans may adjust their policy preferences to match those of their leader (Barber & Pope, 2019), when it comes to intra-party disagreement over policy choices, it is difficult to determine whether a co-partisan voter’s response is the result of a motivation to support the leader, true policy preferences, or ambivalent policy preferences. Following the party leader may be a relatively easy cue for uninformed voters, and thus an insufficient test.

Focusing on norms violations allows us to examine the influence of leaders in a context where policy preferences should not drive behavior and where the public recognizes the norms violation as problematic. In principle, people support democracy, recognize core democratic norms, and agree that violating democratic norms is unacceptable (Carey et al., 2018). From this standpoint, evidence that co-partisan voters reject dissent from co-partisan critics and continue to support a (fictitious) norms-violating leader constitutes clear evidence of leadership-driven motivated reasoning because absent the psychological pull of leaders, voters should reject counter-normative behavior and reduce their evaluations of the leader in the face of criticism.

To test the implications of leadership-driven motivated reasoning and explore some of the limits, we considered multiple norms violations across several studies. As we started this project early in the Trump Administration, there was little systematic research on public responses to norms violations in the U.S. Examples from other countries (e.g., a military coup by a leader) sounded far-fetched, so we feared that experimental primes using such examples may produce low quality responses. So we used several cases of norms violations that were in the public mind at the time. We included some that focus on actions taken by Trump and others that extend beyond the specific actions of his presidency.

The issues we used were: (1) pardoning party insiders convicted of campaign finance violations (“pardon”); (2) political interference with the judiciary by threatening to investigate the personal lives of judges (“judges”); and, (3) conditioning foreign aid on another government’s provision of opposition research on the leader’s opponents (“oppo research”). We also reference a fourth study that focuses on publicly sharing sensitive intelligence information with foreign adversaries (Appendix G). Political pardons and divulging classified information have been salient issues across several administrations.

Since 2017, Bright Line Survey has been tracking U.S. performance on key indicators of democracy (Carey et al., 2019), and several of our cases correspond to these indicators. When exploring the limits of the partisan-leadership cue, we use results of a separate survey that asked respondents to rank order the seriousness of several norms violations as well as the Bright Line opinion surveys to make determinations of which norms violations are more or less serious (see Appendix A). This is a validation exercise to reassure us that the primes indeed reflect different levels of severity. For the norm violations discussed in the experiments below, divulging classified information is the most serious, followed by investigating judges, pardoning a party insider, and, lastly, conditioning aid on opposition research. Although the adjudged seriousness of a given norms violation is not temporally fixed, polls suggest that around the time of our experiments, some of the norms violations we explore were viewed as more serious than others.

Given our focus on executive leaders and the status differential of the misbehaving leader relative to the critic, we rely on formalized roles in the legislative branch to signal the higher or lower-status of the critic. We leverage experiments that examine both the state and the federal level. Although the nationalization of politics increasingly focuses attention on national leaders (Hopkins, 2018), party leaders are important actors in policymaking and inform group attachments at the state level as well (Rosenthal, 2013). Ranking exercises embedded in an online experiment confirmed that voters perceive a clear distance in status between the President and the Speaker/Majority Leader and a further distance to a senator, and similarly view state Governors as higher status than Assembly leaders (Appendix B).⁷

In the following analyses, we begin with a relatively less serious norms violation and a high-status dissenter—pardoning political donors (“pardon”). Study 1a leverages hypothetical state leaders, while study 1b, which is a replication of the structure in study 1a, uses both hypothetical state and real federal leaders. We then turn to experiments that provide suggestive evidence about the potential limits to the pull of party leaders. Study 2 focuses on a high serious norms violation and a high-status dissenter (“judges”). Study 3 focuses on a low seriousness norms violation and a low-status dissenter (“oppo research”). The intelligence sharing study in the appendix covers a high seriousness norms violation and both a high and low-status dissenter. Our aim with this range of studies is to explore the implications of our argument in a variety of realistic contexts, with hypothetical and real leaders, and in scenarios that allow us to explore the boundary conditions of partisan leadership cues.⁸

Analyses

Experiment 1a: Pardon of a Party Insider, Fictitious Leaders

This study was fielded with Lucid from Oct. 24–26, 2018 and included 734 partisan respondents. The sample was balanced to national demographics on gender, race, region, and age (prior to dropping independents). This is a 2(seriousness)x 2(loyalty v. dissent) design. Respondents read a story about a fictitious co-partisan governor who pardoned a party insider convicted of illegally bundling campaign donations. Respondents were randomly assigned to a lower or higher seriousness crime by the party insider. Our hypothesis was that the seriousness of the crime reflected on the seriousness of the leader’s misbehavior. Respondents were then randomly assigned to either a loyalty condition, where they read that the co-partisan state Assembly leader defended the Governor, or to a dissent condition, where the co-partisan Assembly leader criticized the Governor. Designating the co-partisan supporter or critic as the “leader” of the party in the legislature emphasizes his high-status. In the loyalty condition, the Assembly leader states that the party “stands by the Governor’s decision.” In the dissent condition, the Assembly leader criticizes the Governor’s behavior and provides a rationale, suggesting that the Governor’s behavior is an affront to the party and to the rule of law. While the strength of the treatments may be asymmetrical, this was a purposeful choice to ensure that a justification in the loyalty condition didn’t persuade respondents that there was no norms violation by the leader. As a result, any evidence that respondents rally around the Governor or punish the Assembly leader when he dissents is a conservative estimate of the power of leaders.

We use fictitious governors for several reasons. First, it allowed us to test our hypotheses with Democrats who didn’t control the White House at the time. Second, it allows us to control for respondents’ prior affective responses to a state and individuals.⁹ This helps us to disentangle prior affect from leadership effects. We expect that absent personal affect (given the fictitious names and no state mention), if partisans are influenced by leadership cues, they should gravitate to the governor not the Assembly leader. While state leaders may not have the same centrality in the party as federal leaders, the governor is the most public-facing officer and for many voters s/he “personifies” the government, while other state officials have less media visibility (Rosenthal, 2013, 28–30). To avoid confounders, the names for our fictitious Governor and Assembly leader were selected to evoke white, Anglo-Saxon males.¹⁰ The wording of the primes, descriptive statistics, and balance tables are in Appendix C.

After receiving both pieces of information, respondents were asked to evaluate the Governor and the Assembly Leader in terms of affect (feeling thermometer) and intent to vote. Both are scaled from 0 to 1, with 1 indicating greater support. We ask about voting in the primary rather than the general election because, given the pull of partisanship, partisans are unlikely to switch their vote to a different party. It is more realistic to expect that, if partisans are to punish a misbehaving leader, this would happen in the primary. Figure 2 presents results from the first experiment. Panels (a) and (b) show co-partisan responses to the norms-violating leader having read about both the violation and the Assembly leader’s response. Panels (c) and (d) in the second row show responses to the co-partisan Assembly leader based on his loyalty or dissent. For all our experiments, we follow the convention of first discussing responses to the norms-breaking leader and then discussing responses to the Assembly leader who is either a loyalist or a dissenter. The figures present the mean evaluation for each treatment condition, along with 95% CIs. We reference t-tests of whether the difference in means between the loyalty and dissent conditions is significant.

Fig. 2

Pardon of Party Insider

First, we did not find a consistently significant effect of the level of seriousness on co-partisans’ responses to the leader or the critic (see Appendix C5). As a result, we present pooled data. The results point to a significant rally effect in response to co-partisan criticism. Dissent induces co-partisan voters to affectively “rally around” the Governor (Fig. 2a, M_Loyalty = 0.350, M_Dissent = 0.643, p < 0.001). However, dissent has no effect on co-partisans’ intent to vote for the Governor in the primary (Fig. 2b, M_Loyalty = 0.333, M_Dissent = 0.337, p = 0.870).¹¹ The results are consistent with the leadership-cues expectations; co-partisan voters have no prior affect for either individual in this dispute, but the criticism prompts them to support the leader.

Moreover, evaluations of the Assembly leader suggest that co-partisans don’t reward dissent of the norms violation. Dissent has no statistically significant effect on co-partisans’ affect for the Assembly leader (Fig. 2c, M_Loyalty=0.316, M_Dissent = 0.309, p = 0.725) or on primary vote intention (Fig. 2d, M_Loyalty=0.514, M_Dissent = 0.501, p = 0.575).¹² The null rather than negative response to criticism in this vignette could be because the critic is high-status, perhaps protecting him from a negative response from co-partisans.

To explore how people viewed dissent to the leader’s norms violation, we analyzed three questions that tap into the mechanisms behind these relationships. This addresses, in part, perceptions of the dissenter’s motivation and the consequences or cost of the dissent as judged by respondents. The first question gauges the degree to which co-partisans judge the Assembly leader’s response as appropriate. Co-partisan respondents recognize dissent as appropriate (Fig. 3a, M_Loyalty=0.422, M_Dissent=0.791, p < 0.001). The next two questions asked partisans to identify who served the interests of the party and who embarrassed the party. The response options were: the leader, the Assembly leader, both, or neither. The results suggest that partisans recognize the value of dissent and the flaws in the leader’s behavior: 67% believe that the dissenter serves the party compared to 17% for the loyalist (Fig. 3b), while 62% in the dissent condition believe that the misbehaving leader embarrasses the party compared to 19% in the loyalist condition (Fig. 3c). This suggests that respondents did recognize the norms violation as problematic and dissent as appropriate. Yet they rallied around the fictitious party leader, independent of prior positive affect, highlighting the pull of party leaders in groups. More research is necessary to explore the motivated rationales that respondents use to reconcile recognition of the virtue of the dissent, but rallying to the leader nonetheless.

Fig. 3

Pardon of Party Insider Mechanisms

Experiment 1b. Pardon of a Party Insider, President Trump

Our next study replicated the design of the previous study but focused on Republican respondents where we could extend the conditions to include Trump. This study was fielded on MTurk from October 24–25, 2018 and included 500 Republican respondents. Respondents were randomly assigned to the same low seriousness pardon story as in study 1a, except that we randomized whether the behavior was attributed to the fictitious state Governor or to Trump. After reading the story, respondents were randomly assigned to a loyalty or dissent condition. Here, the responding leader in the Governor condition was the Republican Assembly leader while in the Trump condition it was Senate Majority Leader Mitch McConnell, identified with his full title to signal his high status. The pardoned party insider was described identically across conditions. This study allows us to replicate study 1a and extend it to violations by real leaders for whom respondents harbor feelings. The primes, descriptive statistics, and balance tables are in Appendix D.

Dissent against the norms-violating leader leads Republicans to rally around the leader (Fig. 4a and b). Similar to experiment 1a, the results suggest an affective rally effect in response to dissent not only for the hypothetical Governor (M_Loyalty = 0.384, M_Dissent = 0.716, p < 0.001) but also for Trump (M_Loyalty = 0.533, M_Dissent = 0.623, p = 0.004). In contrast to experiment 1a, however, dissent also leads to a rally around the leader in vote intention for the partisan Governor (M_Loyalty=0.406, M_Dissent=0.714; p < 0.001) and for Trump (M_Loyalty=0.473, M_Dissent=0.620; p < 0.001). For both the feeling thermometer and vote intention, the boost the leader receives when he faces intra-party dissent is significantly smaller for Trump than the hypothetical leader (p < 0.001 for affect and p < 0.01 for vote), suggesting that the leadership cue and not affect is driving the rally.

Fig. 4

Pardon of Party Insider (GOP only)

When we turn to evaluations of the Assembly leader, similar to experiment 1a, we see no benefit for the dissenter (Fig. 4c and d). In both the hypothetical and Trump conditions, there is no difference in affect between the loyalist and the dissenter (Governor: M_Loyalty=0.334, M_Dissent = 0.321, p = 0.730; Trump: M_Loyalty=0.661, M_Dissent = 0.661, p = 0.997) and no difference in primary vote intent (Governor: M_Loyalty=0.439, M_Dissent = 0.499, p = 0.126; Trump: M_Loyalty=0.520, M_Dissent = 0.488, p = 0.396). Although the dissenter is not penalized, perhaps because the leader’s offense is not grave and the dissenter is high-status, neither is he rewarded for this costly criticism. The questions that tap into the mechanisms of protecting the party again yield patterns that suggest that people recognized the value of dissent to the leader’s transgression (see Appendix D).

Thus far, studies 1a and 1b provide support for our leadership-driven motivated reasoning argument. When faced with intra-party dissent against the leader, partisans boost their support for the leader. And even though they recognize that the leader’s behavior does not serve the party well and that dissent against the leader is appropriate, the high status co-partisan is not rewarded for his criticism of the leader. Moreover, this finding holds with both hypothetical and real leaders, suggesting that leadership-driven motivated rationalization and not only motivated reasoning rooted in affective attachments or partisanship is driving the results. Study 1a also suggests that the level of seriousness does not play a role. However, this is limited to the specific context of a type of misbehavior that partisans consider minor even though they distinguish in seriousness between the two. The remaining two studies explore the potential boundary conditions on the leadership effect—what happens when the norms violation is very serious or the critic is lower-status?

Experiment 2: Violation of Judicial Independence

This study focuses on a very high seriousness norms violation surrounding judicial independence and dissent from high-status party members. The study was fielded with Lucid from May 29–31, 2019 and included 867 partisan respondents. The quota-based sample was balanced to national census demographics on gender, race, region, and age (prior to dropping pure independents). This study was structured similarly to the “pardon” experiment but focused on a more seriousness norms violation. Respondents read a story about a fictitious co-partisan governor who sought to interfere with the judicial system. The story informed respondents that in response to a state Supreme Court decision that went against him, the Governor ordered that law enforcement investigate the personal lives and finances of the judges who opposed him. The Court is implicitly assumed to have operated within its bounds and made a decision on the merits. As such, it brings us closer to behaviors that emerge in authoritarian regime contexts and which are seen as serious threats in democratic politics. After reading the story, respondents were randomly assigned to either a loyalty or a dissent condition, similar to experiment 1a. We used the same prime design as in experiment 1a and the same character names, but altered the nature of the misbehavior. The primes, descriptive statistics, and balance tables are in Appendix E.

The results suggest that there may be limits to the pull of party leaders, especially when not compounded by prior affect. While studies 1a and 1b showed a rally around the leader, this study shows the first evidence of dissent producing a penalty for the leader. Dissent produces a significant affective penalty (Fig. 5a, M_Loyalty = 0.429, M_Dissent = 0.375, p = 0.005), suggesting that high-status dissent to a very serious violation relative to loyalty is effective at reducing support for the leader. However, this penalty still does not extend to intent to vote in the primary, although the effect is negative (Fig. 5b, M_Loyalty = 0.417, M_Dissent = 0.386, p = 0.102).

Fig. 5

Investigation of Judges

Turning to evaluations of the Assembly leader, high-status dissent to a serious norms violation boosts evaluations of the critic. Co-partisans show greater affect for the dissenting Assembly leader than the loyalist (Fig. 5c, M_Loyalty=0.449, M_Dissent=0.618, p < 0.001) as well as greater willingness to vote for him in the primary (Fig. 5d, M_Loyalty = 0.476, M_Dissent = 0.617, p < 0.001).

To understand what prompted this rally around the dissenter, we return to the questions about the underlying mechanisms that we used in study 1a—appropriateness of the response and whose behavior better serves (or embarrasses) the party. Co-partisans view dissent by the high-status member as more appropriate than loyalty to the misbehaving leader (Fig. 6a, M_Loyalty=0.570, M_Dissent = 0.788, p < 0.001). Moreover, 57% of respondents in the dissent condition indicate that the Assembly leader best serves the party compared to 21% in the loyalty condition (Fig. 6b), and conversely 51% say that the misbehaving leader embarrasses the party in the dissent condition compared to 18% in the loyalty condition (Fig. 6c). This suggests that co-partisans recognize the leader’s behavior as a serious threat to the integrity of the party (it “embarrasses” the party) and the dissenter’s action as an effort to protect the party’s reputation (it “serves” the party). In the loyalty condition, the plurality of respondents think that neither actor serves the party (43%) and both embarrass it (44%). Taken together, these results support our contention that with very serious norms violations by leaders, dissent by high-status party members may be rewarded by co-partisan voters because such voters recognize the dissent as protective of the party. In this case, the tendency to rationalize the leader’s behavior may be overcome by the magnitude of the offense when it is pointed out from within the party. Yet, even with a high-seriousness norms violation, moving vote intent is more difficult, suggesting that shifting electoral support from the leader may require even more egregious behavior that draws even more vigorous and sustained co-partisan dissent.

Fig. 6

Investigation of Judges Mechanisms

Experiment 3: Foreign Opposition Research

The final experiment uses another relatively low-seriousness norms violation—requesting that a foreign government provide information against a political opponent—but where the critic is an individual senator (lower-status). Although this norms violation led to Trump’s impeachment (and was thus a high-seriousness norms violation to Democratic political elites), our sample didn’t view the behavior - tying aid to opposition research—as a very serious offense (see Appendix A). This study was fielded January 2–4, 2020 on Lucid and included 190 Republican participants. Study participants were asked to read a story related to the impeachment trial of Donald Trump, which was starting in the Senate. After this, respondents were randomly assigned to a loyalty or a dissent condition in which a fictitious Republican senator either sides with Trump or criticizes Trump’s action and proclaims he will vote in favor of impeachment.¹³ An individual senator represents a lower-status member compared to a legislative leader (e.g. studies 1–2). Because the senator is fictitious, respondents don’t have prior feelings about him.¹⁴ The primes, descriptive statistics, and balance tables are in Appendix F.

Republicans’ views of Trump are not moved by the response of the senator; there is no difference in affect for Trump between the two conditions (Fig. 7a, M_Loyalty=0.772; M_Dissent=0.805, p = 0.423). However, the low-status dissenter is penalized for his criticism. Affect for the dissenting senator is significantly lower than for the loyalist (Fig. 7b, M_Loyalty=0.579; M_Dissent=0.308, p < 0.001) as is approval (Fig. 7c, M_Loyalty=0.726; M_Dissent=0.360, p < 0.001). Similarly, Republicans are less likely to vote for the dissenting senator in the primary (Fig. 7d, M_Loyalty=0.670; M_Dissent=0.363, p < 0.001).

Fig. 7

Opposition Research and Impeachment (GOP only)

Next, we look at perceptions of loyalty as a potential mechanism for why dissent is penalized. The survey included a question that asked “In your view, how well does each of the following describe Senator Conroy?” The items included: loyalty to the country, loyalty to the party, loyalty to democratic institutions, and being a good American. All items were scored on 4-pt scales ranging from “describes very well” to “doesn’t describe well at all.” These items were combined in an additive index (alpha = 0.748) with higher values indicating greater loyalty. We find that Republicans were less likely to view the senator as loyal when he dissents against Trump (Fig. 7e, M_Loyalty=0.671; M_Dissent=0.484, p < 0.001), suggesting that he is not seen as acting in the interests of the party. As the boundary conditions of our partisan leadership theory suggest, co-partisan voters penalize dissent by a low-ranking member both in terms of affect and in terms of vote.

The results of study 3 suggest that a low-status critic is not only ineffective at lowering evaluations of the misbehaving leader but that the critic faces a “disloyalty” penalty. We find a similar effect in a separate study that examines the response of the Republican Party leadership and an individual senator to Trump’s sharing of sensitive intelligence information with Russia (see Appendix G). Even though this norms violation is viewed as highly serious (see Appendix A), dissent has no effect on evaluations of Trump or the party leadership (a high-status critic) but significantly reduces support for the individual senator (a low-status critic). Although this study, like study 3, is limited in the ability to disentangle leadership-driven motivated reasoning from broader affective motivated reasoning, the results are suggestive that lower-status critics may face a penalty for criticizing the party leader while higher status critics may have more leeway.

Discussion

Our partisan leadership-driven motivated reasoning theory predicted that intra-party dissent against a norms-violating leader would be ineffective at reducing support for the leader and may even lead to a rally. By contrast, the dissenting co-partisan may damage his reputation. These expectations are at odds with CCT. Our theory draws on motivated reasoning but suggests that positive prior feelings may not be a necessary condition to motivate partisans to defend their leader: leadership can induce motivated processes even without prior affect.

Our survey experiments test some implications of the theory and lend support to the argument, particularly with respect to evaluations of the leader. Co-partisan dissent against a norms-violating leader produces a rally around the leader across two levels of seriousness (studies 1a and 1b) or no change in evaluations (study 3). Because we use fictitious leaders, we conclude that leadership cues, not just prior affect, can produce this response. As expected, co-partisan critics gain little. Despite co-partisans’ recognition that dissent is the appropriate response, dissent has no effect on evaluations across two levels of seriousness (studies 1a and 1b) or reduces support for the critic (study 3). This suggests that dissent has no payoff and may be costly for backbenchers.

We offer tentative conclusions as to the importance of boundary conditions. Study 1a which included a direct comparison of two levels of seriousness showed no effect of seriousness on the observed rally effect. The leader benefited from criticism in both cases. But study 2b suggests that the pull of leaders may not extend to very serious norms violations—although we cannot pinpoint a tipping point and our results do not have the benefit of a direct comparison. The status of the dissenter may also matter for how co-partisans evaluate both the leader and the critic; a high-status critic (studies 1a and 1b) may be less likely to face negative consequences than a low-status critic (study 3). Again, the lack of a direct comparison makes this conclusion tentative. We view these findings as preliminary and recognize than more research is needed to establish boundary conditions.

While our theoretical perspective does not differentiate between Democrats and Republicans, extant research points to some reasons that we might expect partisan differences in the power of in-group party leaders (Grossmann & Hopkins, 2016). Studies 1a and 2 include Democrats and Republicans, allowing us to explore if the effects of dissent against the leader differ across parties. In study 1a and study 2, Republicans respond less favorably to dissent (see Appendix C6). These results offer suggestive evidence that Republicans are more critical of dissent against the party leader.

Limitations

We acknowledge several limitations to our studies. First, since they were conducted in a hyperpolarized climate, our work cannot assess the effect of polarization on the evaluation of partisan criticism. Second, we do not have a theory of what types of offenses the public may consider more or less serious. We approach this issue inductively, relying on survey tests. It is possible that people’s perceptions of seriousness may not be linear or it may be context-dependent. We view our study as an opening salvo, introducing these questions, rather than a final answer. Another weakness of the study is that only one experiment provides a direct comparison of seriousness. Therefore, we cannot preclude the possibility that the leader penalty we identify in Study 2 may be an artifact of the sample or time effects. Future research should vary both the level of seriousness, the status of the critic, and (ideally) include real and fictitious leaders within the same study.

Limitations notwithstanding, our findings speak to the power of party leaders in groups qua leaders and not simply to the affective attachments that partisans have for current political leaders. Although not every experiment we present can tease out the role of leadership and affective attachment, the combined patterns (especially with hypothetical leaders) point to the importance of leadership in how partisans navigate intra-party conflicts. Our experiments allow an examination of in-group responses to co-partisan criticism of the leader across a range of scenarios, different levels of government, and across hypothetical and real leaders.

Conclusions

Our findings have important implications for party unity and democratic accountability. On the one hand, leaders can use this power afforded by partisan attachments to overcome internal opposition and forge broad bipartisan change to strengthen democracy. On the other hand, leaders may take advantage of the power afforded by partisan attachments and behave in anti-democratic ways, endangering institutions. The role of partisan leadership cues contributes to explaining recent assertions that in highly polarized settings, democracy is often subverted by incumbents (Svolik, 2020). The self-censorship of potential co-partisan critics also has implications for democracy since it suggests that in polarized systems, political opinions voiced in the public square may not be representative of the full range of views espoused by elected officials. Moreover, co-partisan members fearing electoral penalties may allow violations of democratic norms to go unpunished, further contributing to the erosion of democracy.