Conceptualizing Disagreeable Rhetoric on the Supreme Court

As we conceptualize it, disagreeable rhetoric consists of language with harsh, unpleasant, or negative connotations. Harsh rhetoric is used as an integral part of a framing strategy to structure audience expectations in order to persuade (Tannen Reference Tannen1993; see also Murphy Reference Murphy1964; Entman Reference Entman1993). In the electoral context, emphasizing negative aspects primes emotional responses (Hainmueller and Hopkins Reference Hainmueller and Hopkins2014, 234) and allows the messenger to influence a variety of attitudes and behaviors (e.g., Knoll, Redlawsk, and Sanborn Reference Knoll, Redlawsk and Sanborn2011). Scholars have a more limited understanding of disagreeable rhetoric when it comes to the judicial branch, but research demonstrates that it is on the rise at the US Supreme Court and pervades multiple types of opinions (Bryan and Ringsmuth Reference Bryan and Ringsmuth2016; Wedeking and Zilis Reference Wedeking and Zilis2018), which suggests a need for more careful attention to how it comes about and the implications that result.

In order to add nuance, we consider in more detail the defining features of disagreeable rhetoric in the context of the Supreme Court. While justices may integrate harsh language at multiple stages, including oral argument and private deliberations, we are particularly interested in opinion language for a few reasons. First, opinions have a great deal of substantive importance from legal and policy-making perspectives. This is certainly true with respect to majority opinions of the Court, but even dissents and concurrences have the capacity to influence subsequent policy developments by constraining the impact of majority opinions and signaling conflict to other actors (Wahlbeck, Spriggs, and Maltzman Reference Wahlbeck, Spriggs and Maltzman1999; Urofsky Reference Urofsky2017). Relatedly, justices are very deliberate about their opinion language (Black, Owens, Wedeking, and Wohlfarth Reference Black, Owens, Wedeking and Wohlfarth2016) since legal and political elites pay close attention. Finally, the language of majority opinions in particular has been shown to influence media coverage and public responses, suggesting implications for the Court’s popular standing (Zilis, Wedeking, and Denison Reference Zilis, Wedeking and Denison2017).

With this in mind, disagreeable opinion language consists of a few key features. First, this language highlights unpleasant or negative characteristics for interested audiences (Nelson, Oxley, and Clawson Reference Nelson, Oxley and Clawson1997, 225–28), which is valuable because negative information has a higher likelihood of capturing audience attention and sticking in memory (Marcus, Neuman, and MacKuen Reference Marcus, Russell Neuman and MacKuen2000; Entman Reference Entman2004; Brader Reference Brader2005) and tends to be weighted more heavily in attitude formation (Lau Reference Lau1985). In the context of the Supreme Court, some noteworthy opinions adopt this tactic. For example, the recounting of the facts in Roper v. Simmons (543 U.S. 551 [2005]) describes a horrific chain of events, strongly emphasizing the brutal act of the crime, which ultimately sets the stage for the Court’s legal reasoning that follows. In the legal reasoning, the majority opinion repeatedly uses the term “death penalty” (over 80 times) while also taking a confrontational tone toward the petitioner’s arguments. This is far from the only example of the justices detailing unsettling facts to frame their arguments. In a recent opinion that attracted popular attention, Clarence Thomas went to “great lengths” to describe a murder and criticized his colleagues for leaving out these details (de Vogue and Cole Reference de Vogue and Cole2019). Such language fits within our conceptualization of disagreeable rhetoric by strongly emphasizing negative aspects as part of its persuasive approach.

In addition to emphasizing negative aspects, disagreeable rhetoric may also feature an overall unfavorable tone (see, specifically, Zilis et al. Reference Zilis, Wedeking and Denison2017 and, more generally, Young and Soroka Reference Young and Soroka2012). This particular aspect of opinions has garnered significant attention in recent years, with some commentators worried about the rise in harsh emotional language on the Court. Court observers often noted the language used by Justice Scalia, whose opinions characterize opponents in very unflattering terms (Chemerinsky Reference Chemerinsky2015). Unfavorable language also appears from different quarters of the Court, such as when Justice Alito suggested the “irredeemable corruption” of constitutional interpretation in the federal judiciary (Gerstein Reference Gerstein2015). But while popular attention tends to focus on isolated examples of harsh tone, we note that similarly heated rhetoric has worked its way into many other opinions over time (Wedeking and Zilis Reference Wedeking and Zilis2018).

In short, our understanding of disagreeable rhetoric on the Supreme Court identifies a few important conceptual features, including the emphasis on unpleasant characteristics and use of an unfavorable tone. This language appears increasingly common on the Court and in multiple types of opinions. Indeed, in spite of the fact that commentators often highlight disagreeable rhetoric in dissents, there is nothing to prevent the majority from adopting this language as well, which is critical given that their opinions carry the force of law. We examine in more detail the differences across types of opinions in subsequent sections.

Strategic Considerations Relevant to Disagreeable Rhetoric

Supreme Court justices act strategically when crafting opinion language. With multiple audiences and goals in mind, justices in the majority aim to build public support, bring about implementation, and evade congressional oversight through adjustments in the clarity and complexity of their writing (Owens and Wedeking Reference Owens and Wedeking2011; Owens, Wedeking, and Wohlfarth Reference Owens, Wedeking and Wohlfarth2013; Black, Owens, Wedeking, and Wohlfarth 2016). Majority justices also make use of strategic citation patterns in order to signal stronger legal support for their opinions (Corley, Howard, and Nixon Reference Corley, Howard and Nixon2006; Hume Reference Hume2006) and, more germane to our purposes, adjust their level of disagreeable language depending on the salience of a case in order to safeguard popular support (Wedeking and Zilis Reference Wedeking and Zilis2018). For dissenters, aggressive language is common on issues about which justices feel strongly and for which they value policy outcomes over the legitimacy of the institution itself (Cross and Pennebaker Reference Cross and Pennebaker2014; Bryan and Ringsmuth Reference Bryan and Ringsmuth2016).

On the whole, existing work views disagreeable rhetoric as a product of the justices’ desire to take clear positions for interested audiences (Cross and Pennebaker Reference Cross and Pennebaker2014); justices have long been seen as concerned about their reputations among audiences (Baum Reference Baum1994, Reference Baum2006). Using disagreeable language is valuable not only as a framing device but also as a way to raise the salience of specific policy considerations that the justices seek to emphasize.

The literature supplies additional traction when we broaden the focus beyond opinion rhetoric and look at other forms of contentious or disagreeable behavior, which are also becoming increasingly common (e.g., Black, Owens, and Wedeking Reference Black, Owens, Wedeking, Danelski and Ward2016; Rice and Zorn Reference Rice, Zorn, Danelski and Ward2016). One study explains, “The use of anger may seem to be a lapse by the justices [but] the use of emotion may be a strategic tool that makes opinions more powerful” (Cross and Pennebaker Reference Cross and Pennebaker2014, 890; emphasis added). Consistent with this insight, Johnson, Black, and Ringsmuth (Reference Johnson, Black and Ringsmuth2009) show that judges become more likely to read oral dissents when a dissent’s ideological distance from the majority increases and the majority coalition is a minimum winning one. Conversely, interpersonal cooperation can dissuade justices from behaving in a disagreeable fashion. For example, the reading of oral dissents becomes less likely when one has worked with the majority opinion author in the past (Blake and Hacker Reference Blake and Hacker2010). This view squares with a broader literature showing that interpersonal relationships play a critical role for policy-minded justices (Haynie Reference Haynie1992; Epstein and Knight Reference Epstein and Knight1998; Maltzman, Spriggs, and Wahlbeck Reference Maltzman, Spriggs and Wahlbeck2000). In the short term, collegial accommodation allows justices to persuade colleagues to sign on to a specific opinion, while in the long term it fosters mutually beneficial interactions among the group.Footnote ² Again, however, noncollegiality is not just the province of dissenting justices (e.g., Black, Johnson, and Wedeking Reference Black, Johnson and Wedeking2012; Cross and Pennebaker Reference Cross and Pennebaker2014; Rice and Zorn Reference Rice, Zorn, Danelski and Ward2016; Smith Reference Smith2017), although it is an approach that dissenters frequently use (Johnson et al. Reference Johnson, Black and Ringsmuth2009).

Therefore, existing work suggests that strategic considerations influence both the majority and the dissent in terms of their willingness to employ disagreeable rhetoric. To be clear, this does not always mean that the coalitions write the same types of opinions but rather that both are guided by strategic considerations when they act. An important underlying goal for both is to build support for their positions at the expense of their opponents.

Finally, we note an additional drawback of disagreeability: its ability to weaken the Court’s popular authority (Epstein, Segal, and Spaeth Reference Epstein, Segal and Spaeth2001, 362–63; Baird and Gangl Reference Baird and Gangl2006; Zink, Spriggs, and Scott Reference Zink, Spriggs and Scott2009). Because media coverage tends to track closely the language used by the Court majority in reporting on decisions (Zilis Reference Zilis2015), an important cost associated with breaches of collegiality is sensationalist or divisive portrayals of rulings to the public, which hamper the Court’s support (Johnston and Bartels Reference Johnston and Bartels2010). Simply put, the Court’s rhetoric matters in shaping responses to decisions.Footnote ³

In sum, the literature suggests two primary insights. First, justices adjust the language of their opinions as part of a framing strategy for external audiences. When it comes specifically to disagreeable rhetoric, this language allows justices to make salient their positions for interested audiences and influence the reception that external actors give to rulings. Second, we note that disagreeable behavior harms the ability to bargain with one’s colleagues and may also have costs in terms of institutional support. Focusing on these costs and benefits, we next build our theory of disagreeable rhetoric with an application to Supreme Court opinions.

Theoretical Framework

Empirical Strategy

Our empirical approach proceeds in two steps. First, we explore our personal importance hypothesis, which predicts that opinion authors become more willing to integrate disagreeable rhetoric on issues that are personally important to them. While there are several avenues to study this, we focus our first analysis on the rhetoric in separate (nonmajority) opinions. This is advisable because the justices have significant control over the content of their concurring and dissenting opinions. Unlike for majority opinions, there is no incentive to bargain over opinion language in order to maintain a winning coalition, making for an appropriate test of our justice-specific personal importance hypothesis.

After showing that the importance increases the likelihood that justices will integrate disagreeable rhetoric in separate opinions, we then proceed to our second analysis, which involves analyzing the consequences of disagreeability, particularly as they concern the size of the Court’s majority coalition. We anticipate that disagreeable rhetoric in majority opinions will influence the size of the coalition. However, we are also cognizant of concerns with endogeneity or reverse causation, in which coalition size may influence opinion rhetoric. In other words, while harsh rhetoric may cause some justices to refuse to join, it may also be the case that coalition size influences disagreeable rhetoric in an opinion.Footnote ⁵

Given these concerns, we use two-stage least squares as an estimation strategy in our second analysis because it is designed to account for endogenous relationships. The first-stage equation explains disagreeable opinion rhetoric as a function of the issue’s personal importance to the opinion author as well as a series of other covariates. We then estimate a second-stage equation to explain coalition size as a function of disagreeable opinion rhetoric and a series of other covariates. Estimating a two-stage least squares model requires us to identify instrumental variables that are exogenous for each stage.

For the first-stage equation, we use the personal importance to author variable as an instrument for estimating disagreeable opinion rhetoric (i.e., personal importance to the author directly influences opinion rhetoric but not the size of the majority coalition). To evaluate the theoretical justification for this instrument, we draw on our first analysis, which demonstrates that personal importance influences the extent to which justices rely on disagreeable language in their opinions. Specifically, justices are more likely to compose a harsh opinion on a case of high personal importance than any other case.

To evaluate the instrument’s validity for majority coalitions, we consider the concern that other (nonauthor) justices may vary in their likelihood of joining a coalition given the importance they place on the case. In other words, one threat to validity might occur when the importance of an issue to nonopinion authors influences their likelihood of joining a coalition, thus affecting the second-stage equation. This objection necessitates that we account for the aggregate importance of an issue to all other justices, and not just the majority opinion author, since this factor is likely to influence their voting behavior. To do so, we include an aggregate importance control variable in our first- and second-stage equations. This variable captures the importance that nonauthor justices place on a given case. The personal importance to author instrument, however, is included only in the first stage.Footnote ⁶

For the second-stage equation, whether the Court is operating with a missing member (i.e., if it has all nine members or not) is a theoretically appropriate instrument.Footnote ⁷ This factor should have a strong link to estimating the size of the majority coalition, since cases in which the Court is at less than full strength have fewer members available to sign on to a majority opinion, but it is not theoretically expected to influence the disagreeability of an opinion. To ensure the validity of this instrument, we have performed a number of robustness checks, omitting the most common missing members from our analysis to ensure that the recusals of highly disagreeable justices are not driving the results. Additionally, the results of a Sargen-Hansen test for overidentifying restrictions do not allow rejection of the null hypothesis that both the first- and second-stage instruments are valid.

To summarize our empirical approach, we begin by evaluating our personal importance hypothesis, showing how this factor increases the use of disagreeable language, particularly in separate opinions in which an author has significant control over content. Then, we model the endogenous relationship between disagreeable language and coalition size for majority opinions. We find that highly disagreeable majority opinions attract smaller coalitions of justices to sign on.

Measuring and Validating Disagreeable Rhetoric

For our dependent variable, we examine all US Supreme Court opinions from the 1946 to 2011 terms. While some work has explored other forms of disagreeability (e.g., Black et al. Reference Black, Treul, Johnson and Goldman2011), very little explores written opinions (but see Wedeking and Zilis Reference Wedeking and Zilis2018), which is unfortunate since these are the clearest means through which the justices communicate with interested audiences, policy makers, and other actors in the political and legal system. As such, the language used in written opinions has tremendous importance in shaping subsequent developments both inside and outside of the legal system. Furthermore, we anticipate that our approach is a conservative one, since justices are much less likely to employ inflammatory language in carefully thought-out written opinions than they would be in oral arguments or oral dissents.Footnote ⁸

We examine three different measures of disagreeable language from two different computer programs. We do so because it enables us to make more robust claims if our results are replicable across different data sets and coding schemes. The first computer program is the Dictionary of Affect in Language (DAL), which measures the emotional meaning of words and texts (Whissell et al. Reference Whissell, Fournier, Pelland and Makarec1986; Whissell Reference Whissell, Plutchik and Kellerman1989). The program observes the frequency with which a set of words, corresponding to a given emotional construct, occurs in a text, rating each word for its degree on three dimensions: activeness, pleasantness, and imagery. DAL has been widely used in psychology and was previously used in political science to study the Supreme Court (Black et al. Reference Black, Treul, Johnson and Goldman2011).

For our first two indicators that capture disagreeable language, we make use of the DAL subdictionaries of “nasty” and “very unpleasant” words. Nasty words are those that score in the top quartile for activation and the bottom quartile for pleasantness, while very unpleasant words are in the bottom decile for pleasantness; as such, both give us insight into the disagreeability of opinions.

A third measure of disagreeable rhetoric examines the negative emotion category from Linguistic Inquiry and Word Count (LIWC), a textual analysis software package that operates on the same dictionary-based principles as DAL. LIWC employs a word count strategy that searches whatever text is under review for over 2,300 words (or word stems) using specific, validated dictionaries (Pennebaker and King Reference Pennebaker and King1999). LIWC has also been employed to study the Court on several occasions (e.g., Owens and Wedeking Reference Owens and Wedeking2011; Corley and Wedeking Reference Corley and Wedeking2014; Cross and Pennebaker Reference Cross and Pennebaker2014) and has a specially designed dictionary of negative emotion words. LIWC’s dictionaries were designed specifically for capturing tone; their validity and reliability have been extensively tested by previous scholars (Pennebaker and King Reference Pennebaker and King1999; Tausczik and Pennebaker Reference Tausczik and Pennebaker2009).

Factor analysis of these three indicators confirms that they have an underlying unidimensional structure. Cronbach’s alpha is 0.785, and the eigenvalue of the first extracted factor is 1.337 (the second is −0.139). The factor loadings range from 0.628 (negative emotional words) to 0.695 (very unpleasant words). Given this evidence, we create a factor variable using a simple regression method. Using the factor variable provides the added benefit that it focuses on the variance common to all three measures, which we assume to be disagreeable rhetoric, and reduces reliance on idiosyncratic effects of each individual measure. This added robustness protects against misleading false positives from words that may be used in one dictionary but not the other two. Importantly, in the following model estimates, if we use all three indicators individually (rather than the factor variable that combines all three), we get nearly identical results (see, available online). Written opinions range from −2.599 (mild) to 6.188 (harsh), with a mean of 0 and a standard deviation of 0.819. To provide a more concrete illustration of the variance, we can focus on one indicator that is a component of the factor score: the percentage of nasty words used by the justices. This ranges between 0% and 12.5% in a given opinion, with a mean of 3.657%. While this may seem like a small difference at first, remember we are talking about opinions that contain thousands of words (our mean majority opinion length is over 4,000 words).

To explore the properties and validity of our dependent variable, figure 1 plots three distributions of disagreeable rhetoric, one for each type of opinion (majority, concurrence, and dissent). As figure 1 demonstrates, the distribution for each opinion type is very similar. In addition, figure 1 highlights 10 opinions along the distribution to provide further context for how these opinions map onto our dependent variable.

Figure 1.

Distribution of disagreeable rhetoric, across opinion types. There are 10 opinions highlighted in the distribution to provide a greater description of the content of the measure. Each opinion lists the opinion author by type as well as the author’s initials. PS = Potter Stewart; WJB = William J. Brennan; WHR = William H. Rehnquist; RBG = Ruth Bader Ginsburg; TCC = Tom C. Clark; WEB = Warren E. Burger; AMK = Anthony M. Kennedy; HAB = Harry A. Blackmun; BRW = Byron R. White.

To provide more detail about some of the cases that exhibit high levels of disagreeable rhetoric, let us consider three cases that appear on figure 1. First, Justice White’s dissent in Booth v. Maryland (469 U.S. 221 [1985]) has an extremely high score. Consider these two quotes from the opinion:

The affront to humanity of a brutal murder such as petitioner committed is not limited to its impact on the victim or victims; a victim’s community is also injured, and in particular the victim’s family suffers shock and grief of a kind difficult even to imagine for those who have not shared a similar loss. …

The Court is “troubled by the implication that defendants whose victims were assets to their community are more deserving of punishment than those whose victims are perceived to be less worthy,” and declares that “our system of justice does not tolerate such distinctions.” Ante, at 506, n. 8. It is no doubt true that the State may not encourage the sentencer to rely on a factor such as the victim’s race in determining whether the death penalty is appropriate. Cf. McCleskey v. Kemp, 481 U.S. 279 (1987). But I fail to see why the State cannot, if it chooses, include as a sentencing consideration the particularized harm that an individual’s murder causes to the rest of society and in particular to his family.

(Booth, 483 U.S. at 515, 517; Justice White dissenting)

In addition, consider how Justice White’s majority opinion in Enmund v. Florida (458 U.S. 782 [1982]), which is also shown in figure 1, frames the legal question. Specifically, he writes, the case is about “the question whether death is a valid penalty under the Eighth and Fourteenth Amendments for one who neither took life, attempted to take life, nor intended to take life” (Enmund, 458 U.S. at 787). The opinion then goes on to use the word “death” as part of “death penalty” over 70 times. The words “killing” and “rape” are also used extensively. Yet the harsh tone does not stop there. It also criticizes the dissent: “The dissent criticizes these statistics on the ground that they do not reveal the percentage of homicides that were charged as felony murders or the percentage of cases where the State sought the death penalty for an accomplice guilty of felony murder. Post at 818–819. We doubt whether it is possible to gather such information” (Enmund, 458 U.S. at 796; Justice White in the majority). These sorts of attacks are becoming more commonplace. Consider Justice Blackmun’s concurrence in Dee Farmer v. Edward Brennan (511 U.S. 825 [1994]), also shown in figure 1. For example, Blackmun writes, “The Court’s analysis is fundamentally misguided; indeed it defies common sense” (Farmer, 511 U.S. at 854). In sum, there is a tone and hostility that the measure seems to capture.

Additionally, when we analyze temporal trends using our measure, we see a persistent rise in disagreeable rhetoric over time, including the fact that five of the most disagreeable justices in the entire sample serve on the current Court. This fact is worthy of attention in its own right, but it also suggests a measure of face validity for our dependent variable, since recent studies note that various forms of disagreeable rhetoric appear to be on the rise at the Court (e.g., Long and Christensen Reference Long and Christensen2012).

Analysis 1: Does Personal Importance Influence Disagreeable Rhetoric in Opinions?

We begin by analyzing our personal importance hypothesis, using the empirical models presented in table 1. We focus this analysis on the level of disagreeable rhetoric in separate opinions—dissents and concurrences—because of the control that an author exerts over the content. Put differently, separate opinion authors do not need to negotiate with colleagues in order to maintain a majority, making these opinions an appropriate avenue for studying how the personal importance of an issue affects opinion language.

Table 1.

Estimates of Disagreeable Rhetoric in Separate Opinions

Note.—Ordinary least squares regression coefficients with standard errors (in parentheses).

* p < .05.

** p < .01.

We find substantively consistent results across our separate models for concurrences and dissents when it comes to how disagreeable rhetoric comes about. Most notably, we find support in both types of opinions for the hypothesis that justices integrate disagreeable language more readily on matters of high personal importance to them. The effect is quite prominent in the case of dissenting opinions and significant at $p<.001$ . The model estimates that a 1 unit increase in an issue’s importance to a justice leads to a 2.10 unit increase in disagreeable rhetoric, a change of just under 3 standard deviations. From a substantive perspective, we can better understand this effect by isolating the influence of importance on any one of our three dictionary-based indicators of disagreeability (which we have done in the appendix). So, for example, the influence of a unit change in importance is to increase nasty words by about 3.97%. This equates to about 112 additional nasty words in the average dissenting opinion. The comparable figures for very unpleasant and negative emotion language are, respectively, an increase of about 65 and 48 words in the average dissent.

Turning to the model for concurring opinions, we also find support for the hypothesized effect of personal importance on disagreeability. The estimated coefficient demonstrates that a 1 unit increase in personal importance ramps up disagreeable rhetoric by about 1.85 units. Again, translating this to specific subdictionary word counts, the effects equate to an increase of about 37 nasty words, 30 very unpleasant words, and 20 negative emotion words in the average concurrence. While this is a substantively meaningful difference, representing dozens of words in a typical opinion, we note that it is a bit smaller than what we observed for dissents, which may be due to a few factors, including the fact dissenters might have even stronger incentives to emphasize harsh language given their losing positions in the case. But dissenting opinions also tend to be much longer on average than concurrences (in fact, more than double the length), which means the number of disagreeable words is naturally higher.

Turning briefly to our control variables, we see that a few of them have a consistent effect across both models. First, we have controlled for the collective importance of an issue to all justices on the Court using an aggregate measure, and the significant coefficient on this variable indicates that it is associated with increases in disagreeable rhetoric in both opinion types. Also, rights and liberties cases have a similar effect in bringing about more disagreeable language. Both of these results make some sense. They indicate that on collectively important and legally significant issues, separate opinions tend to be more disagreeable overall. Finally, we also see that lengthier opinions are more disagreeable on average, as shown by the significant coefficient on the word count variable.

Overall, the results from our first analysis tell us something important about the determinants of disagreeable language in separate opinions, in which justices have significant control over the content. We find support for our main hypothesis, indicating that in cases personally important to them, separate-opinion authors become more willing to go the disagreeable route. Our results also show that the substantive change in opinion language that results is quite meaningful.

Analysis 2: What Effects Does Disagreeable Rhetoric Have on Coalition Size?

Having demonstrated that disagreeable language becomes more common in dissents and concurrences when the justices resolve cases that they find personally important, we next turn to exploring the implications of disagreeable opinion language. This is important because even majority opinions integrate a significant amount of disagreeable rhetoric, yet the effects are poorly understood. At the same time, we note that the bargaining environment is likely to have a greater effect on the language of majority (as opposed to separate) opinions, which suggests the potential for an endogenous relationship between disagreeable rhetoric and majority coalition size. Our modeling approach enables us to take this into account. Table 2 contains the estimates from the two-stage least squares model, with the first stage using disagreeable rhetoric as the dependent variable and the second stage using majority coalition size.Footnote ¹³

Table 2.

Estimates of Disagreeable Rhetoric and Coalition Size in Majority Opinions

Note.—Two-stage least squares regression coefficients with standard errors (in parentheses). DV = dependent variable. N = 6,642.

* p < .05.

** p < .01.

First, we focus briefly on the first-stage results presented in table 2. Examining the coefficient estimates, we again see that author personal importance is significant and in the expected direction, suggesting that as a justice feels more strongly about an issue, this results in a significantly more disagreeable opinion. The estimated coefficient implies that a 1 unit increase in an issue’s importance to a justice leads to a 2.081 unit increase in disagreeable rhetoric, a change of just under 3 standard deviations. These results echo our earlier findings for separate opinions, providing more support for the idea that in cases personally important to them, opinion authors become more willing to go the disagreeable route. We have observed this effect for dissenting, concurring, and even majority opinions. However, we note that the size of the effect is slightly smaller in the majority versus dissenting opinion model, perhaps suggesting that negotiation over these documents among the coalition can blunt the impact of the issue’s personal importance to the author a bit.

More importantly, we are interested in the implications that disagreeable rhetoric has for the size of the Court’s majority coalition. Since our model estimates simultaneously speak to meaningful variation in the content of written opinions and their consequences for coalition size, we turn to the results of the second stage in table 2. In other words, we still want to know whether the justices might harm the institution when they deploy harsh rhetoric. Can it be of any particular consequence if the Court is apparently becoming a more disagreeable place over time?

To answer this, our primary focus is on the role of opinion disagreeability in explaining coalition size in the second stage. We find evidence of a significant negative association between opinion disagreeability and coalition size, with an increase in the use of disagreeable rhetoric accompanied by the predicted decrease in the size of the Court’s majority coalition. This finding suggests that there are real consequences for using disagreeable language.

But how much of an impact does disagreeable rhetoric have on the size of a majority coalition? Figure 2 displays the estimated linear prediction for coalition size across the range of disagreeable rhetoric. As it illustrates, a small increase in disagreeability has a small but meaningful effect on the size of the coalition. For instance, as a majority opinion becomes harsher, moving from 2 standard deviations below to slightly above the mean, coalition size is predicted to decrease by about one member, meaning that an ambivalent justice may choose not to sign on to a significantly more disagreeable opinion. Furthermore, a sizable injection of disagreeable language, moving across the range of the scale, will sharply diminish the size of the majority. Our model predicts that, all else equal, the least disagreeable majority opinions garner majorities of about eight members, while the most disagreeable ones generate a minimum winning coalition.

Figure 2.

Linear predicted majority coalition size as a function of opinion disagreeability. Estimates are derived from the second stage in table 2, with 95% confidence intervals. The line at bottom depicts the kernel density estimation of the opinion disagreeability variable.

We took a variety of steps to verify the robustness of these results (details in the appendix). First, to ensure that a subset of the justices did not drive our findings, we reran the analyses excluding the justices who were most commonly missing from the Court when cases were decided, which did not alter our results. We also ran our models both with and without justice-fixed effects. To ensure that our findings were not driven solely by any one of our three dictionary-based measures of disagreeable language, we reestimated our models with each individual dictionary measure from either LIWC or DAL as our dependent variable. Next, we reran our majority opinion analysis as two separate equations with errors clustered by natural court, estimating the effects of personal importance on opinion disagreeability and those of opinion disagreeability on majority votes. Finally, we considered other explanations that may account for variations in opinion language, including the tone used by dissenters in their written opinions. These explanations, in addition to the ones for which we have already controlled, help capture the idea that opinion language is the product of interactions between the majority and dissenting justices in a case. Across all alternative specifications, our results hold, demonstrating that justices write harsher opinions in issues personally important to them and that harsh opinions decrease the size of the majority coalition in a case.

This decrease in coalition size becomes even more important when one considers that the size of the majority coalition can often be seen as an indicator of the strength of the Court’s voice. For example, many Court watchers are often quick to point out one of the most important signals that the Court sent in its Brown v. Board of Education (347 U.S. 483 [1954]) decision was having it be a unanimous opinion. Imagine the amount of resistance that might have been encountered if that decision had been handed down with a 5–4 or a 6–3 decision. Recent empirical evidence verifies just such an effect: when citizens learn about division on the Court in a given case, they become less likely to agree (Zilis Reference Zilis2015) and comply with a ruling (Zink et al. Reference Zink, Spriggs and Scott2009). When we consider this alongside the finding from our earlier results that showed disagreeableness to be on the rise, it is no mystery as to why we often find recent Courts with more noticeable vote splits. Our exploration of disagreeability’s repercussions here may hint at further peril ahead for an institution that depends on popular support to carry out its democratic function.

Conclusion

In this article we make several important contributions. First, we provide the first large-scale attempt to explain the level of disagreeableness in multiple opinion types. We do so, importantly, by highlighting significant variation in disagreeable rhetoric even after policy disputes have been resolved. Our findings emphasize how issue importance influences opinion content and, in turn, affects majority coalition size. We find that the personal importance of an issue to an opinion author is a strong systematic predictor of the level of disagreeableness in opinion language.

Second, we show that this disagreeableness has real consequences for the institution. Harsh rhetoric is associated with smaller majority coalitions on the Court, which has multiple implications. There is an ongoing debate over the costs and benefits of Supreme Court dissent (Gibson, Caldeira, and Spence Reference Gibson, Caldeira and Kenyatta Spence2005; Sunstein Reference Sunstein2014; Urofsky Reference Urofsky2017), and one recent study suggests that it can increase perceptions of procedural fairness and acceptance of rulings on the part of the public (Salamone Reference Salamone2014). Other studies, however, find that division leads to more negative media coverage (Zilis Reference Zilis2015) and depresses popular support for rulings (Baird and Gangl Reference Baird and Gangl2006; Zink et al. Reference Zink, Spriggs and Scott2009), indicating that the institution has reason for concern when it produces disagreeable opinions. But beyond attitudes toward specific rulings, our findings also have implications when it comes to institutional support. The size of the majority coalition is often viewed as an indicator of the strength of the voice with which the Court is speaking, and “signals suggesting disagreement on the Court possibly weaken the standing of the Court as an institution” (Wahlbeck et al. Reference Wahlbeck, Spriggs and Maltzman1999, 491). While others have suggested that disagreeability may imperil the Court’s legitimacy (Bryan and Ringsmuth Reference Bryan and Ringsmuth2016), we offer one of the first studies to identify the mechanism of declining majority coalition size through which it may do so.

Third, we offer multiple tests of our constructs, indicating a robustness that cannot be dismissed as being weak or cherry-picking the best results. Our findings, as a consequence, have implications for the study of political rhetoric. They suggest that when an issue becomes important enough, justices use language they know may harm the institution. This modifies the existing understanding of the lengths to which the justices will go to maintain institutional support. Rather than scrub their opinions of harsh rhetoric once judicial coalitions have formed, these actors leave a considerable degree of disagreeable rhetoric in place in their final written opinions. We suggest that in doing so, they may have an eye toward winning the rhetorical struggle of the day. We note that our work is but a first step in suggesting that political actors may employ such rhetoric with this short-term goal in mind. Additionally, our results have implications for the opinion assignment literature in suggesting that authors who write disagreeable opinions may have a harder time holding together robust majority coalitions, something of which the assigning justice may be aware. If so, this fact may limit the ability of the Court’s most pugnacious justices to influence majority opinions.

Finally, our results also raise important questions for the future. Specifically, among others, we note that there is a decline in collegiality with respect to opinion writing over time. Why is this the case? Our results suggest a need for a closer look at the changing nature of the political environment and the shape of the Court’s docket. More specifically, they raise the question as to whether rhetorical norms may have played a role in the demise of the Court’s norm of consensus. While our results do not definitively tell us the answers, they point us in that direction for future research.