The ways that judges justify their decisions have profound impacts on our lives. Notionally, at least, the reasoning that judges use should be rational, if not logical. One role of the attorney in litigation is to supply the judge an acceptable route to the outcome that the advocate’s client seeks—signposts in the form of arguments. These signposts, too, should be rational. An important way that lawyers and judges make arguments is by using prior court opinions that they cite in their arguments. This leads to an important question: How do lawyers and judges use cited cases in their legal arguments? Legal theorists have spilled a great deal of ink and a great many bytes of digital text over the question of what counts as a rational argument, which arguments have “rational force,” whether all legal arguments should be deductive in form, etc.[1] I call these questions ‘metanormative.’[2] They are normative questions—explored by legal philosophers and legal theorists—about what norms legal authors—judges and advocates—should embrace in their reasoning. Norms about norms. But empirical evidence about what norms legal authors do embrace in their reasoning should inform metanormative thinking.

I do not claim that merely because a certain way of arguing is prevalent among lawyers and judges it therefore ought to be. That argument would lose its head to Hume’s Guillotine.[3]

But the argument supporting the claim that legal practice should inform legal theory goes this way: Legal argumentation is practical argumentation. In that context, if an argument form can be rationally justified on the metanormative plane and lawyers and judges accept it in fact on the practical or plain-old normative plane, the argument form should be metanormatively acceptable.[4] If that perspective is correct, the argumentative practices of judges and lawyers should inform metanormative legal theory: If lawyers and judges use an argument form, metanormative theory needs to assess whether it is rational. If metanormative theory predicts or requires certain argument forms that lawyers don’t use, then the metanormative theory must at least account for its inability to model lawyers’ practices. In that case, we should expect the practices of judges and advocates to represent their normative preferences—and thus the practical norms of legal reasoning.[5] Part I develops this argument further.

Legal theory offers metanormative claims about how lawyers should argue by asserting, for example, that certain types of arguments should be the most rationally persuasive and that certain principles must guide the use of legal analogies.[6] There is practically very little research, however, about how lawyers do use cases. That research would take a form I would call ‘empirical legal theory’: the exploration of questions relating to norms of legal theory, as they are manifest in practice, using empirical materials and methods.[7]

‘How do judges and advocates use cases?’ is a big question.

The problem is that there is no extant study of how lawyers and judges use cases in their arguments—period.[8] Worse, there are no extant research methods for characterizing case uses in a way that is useful for answering the questions of empirical legal theory. There is no method for ‘coding’ them, as a qualitative researcher would say.[9] Available research methods are inadequate to the task of answering these questions because they are either too coarsely or too finely grained. Typically, the existing methods are coarsely grained and have involved assessing only whether an argument cites a particular case anywhere at all, without regard to how many times the argument cites it or how it uses it.[10] These methods are focused on establishing pecking orders: which opinions, articles, and other authorities courts cite most often and are therefore most influential. Empirical legal theory requires a finer grain of analysis than this. In the alternative, large-scale assessments of legal arguments sometimes approach them from a rhetorical or stylistic perspective, but these studies often dwell on minute stylistic details and do not tell us much about the practical norms of rational persuasion.[11] Empirical legal theory requires a coarser grain of analysis than this.

This study identified five common types of case use in a study of 199 artifacts—the textual documents that were the objects of study here. A case use consists of all citations to and discussion of a court opinion in a section of a legal argument to support the assertion of the author’s claim in that section of argument. The artifacts here were 199 advocates’ briefs and the court opinions they precipitated, which I hand-coded.[12]

In these artifacts, four types of case use made rational argumentative appeals: supporting assertion of a rule, supporting a policy statement, supporting a generalization about prior cases, and functioning as an example of how to apply a rule, policy, or standard (what I call a ‘legal analogy’). A fifth type consisted of citing a case because the author quoted it. And a sixth catch-all type existed for types of use that did not fall into these categories.[13] The list of four rational appeals proved robust for coding. Of 5638 case uses in the 199 document artifacts here, only 383 (or 6.8%) failed to fall into at least one of those categories, and more than 1500 fell into two or more of them.

When it comes to the ways that advocates and judges used these cases, there was a clear pecking order across all artifacts: Rule-based case uses were twice as common as legal analogies, which were twice as common as policy arguments. This article therefore provides support to one meta-normative theory, the hierarchy of rational appeals for which Professor Susan Provenzano and I have argued.[14] It may not be surprising that court opinions, which seek to project an air of authority, used cases significantly more often than advocates’ briefs to support rule statements and significantly less often to provide examples of the application of rules. This article also shows how the practices of prevailing advocates differed significantly from those of the losing attorneys, suggesting that so-called ‘losing briefs’ might have a certain look to them. It also notes that advocates of parties moving for relief made different uses of cases than those opposing such motions and that authors in some jurisdictions employed cases differently than those in others.

These findings should shape scholarship about legal theory, holding it to account for its (in)consistency with legal practice. Advocates may be interested to learn how the briefs of prevailing attorneys varied from those of their opponents. Teachers of legal theory and practice may wish to guide students toward practices that are typical of the professionals the students wish to become; on the other hand, they may wish to oppose the typical performances on display here and argue that there are better practices. This article concludes with a call to extend this research, binding legal theory and practice to each other through empirical research.

In Part II, this article develops a set of methods for studying empirical legal theory. That part provides a detailed explanation of methods used here to assist other researchers in reproducing or criticizing them. I ground these methods in contemporary argumentation theory and adapt them from methods in applied linguistics and writing studies. Part III then reports the principal findings. It answers two research questions, the first—using a pilot study to identify categories of case use—as a preliminary exploration of the second, a presentation of the relative frequency with which judges and advocates made these types of use of cases.[15] Part III also revisits the question of what kinds of use authors put cases to, considering whether case uses that did not fall into one of the four rational appeals identified here might exhibit some other pattern. Part IV then discusses the findings.

I. Theoretical & Empirical Contexts

This article provides partial answers to the very broad question: How do judges and advocates use cases in their arguments? No single study will answer that question completely. The question is what Professor Merton called an “originating question,” or what the researcher “wants to know.”[16] Fact-focused studies are necessary to lay a foundation for explanations, because law scholars are too ready “to assume that they know the facts about the workings of [law] without special investigation, because [law] is, after all, their native habitat.”[17] This part briefly explains one particular theory about how judges and advocates argue and identifies two resulting research questions that this study took on. Here “[t]he originating question must . . . be recast to indicate the observations that will provide a provisional answer to it,” which usually “requires a search for empirical materials through which the problem can be investigated to good advantage.”[18] We must explore the general problem in some particular instance(s).

The theory at issue here is the theory of legal topoi, which identifies three broad categories into which rational legal arguments must fall—empirical, conventional, and values-based—and places them in a hierarchy, with the first of them more rationally persuasive than the second, and the second more than the third.[19] Part I.A. describes that theory in more detail and grounds the research questions for this study in it. Part I.B. considers previous studies of legal argumentation and what they can tell us about these questions.

Before proceeding to those discussions, though, it is important to identify a basic framework for discussing legal argumentation and for motivating empirical study of argumentation within that framework. Argumentation is “a series of propositional sentences—called ‘premises’—arranged in a form that supports the truth or acceptability of another propositional sentence, called a ‘conclusion.’”[20] As a consequence of this definition, “any written or spoken legal analysis—whether it appears in a memorandum analyzing some aspect of the law, a lawyer’s brief written to persuade to a court, or a court’s opinion written to justify or explain a decision—contains argumentation.”[21]

Metanormatively, legal argumentation should be rational or “cogent”; that is, it should consist of “premises . . . acceptable to the audience to whom it is addressed, relevant to its conclusion, and sufficient to warrant belief in its conclusion.”[22] Legal argumentation is thus dialectical, in that it anticipates a verbal exchange where the parties subject their claims to critical assessment to “move from conjecture and opinion to more secure belief.”[23] In this article, I refer to a ‘rational appeal’ as a dialectical argumentative move that contributes to the cogency of the argument of which it is a part.

Lawyers, however, strive to win, so they are interested in constructing arguments that are persuasive. Legal argumentation is in this sense rhetorical, in that advocates and judges seek argumentation with an eye toward its “potential effectiveness . . . in convincing or persuading an audience in actual argumentative practice.”[24] Consequently, proponents of legal arguments often supplement rational appeals with what I will call ‘tactical appeals,’ argumentative moves that the proponent uses to make the argument more persuasive.[25] This could include an argumentative move that might be termed ‘purely rhetorical,’ in that it does nothing to increase the cogency of the argument, but rational and tactical appeals are not mutually exclusive. Only when a tactical appeal leads the audience away from a cogent argument is it deceptive or fallacious.[26]

This article is concerned principally with rational appeals but without derogation of tactical appeals, including applied legal storytelling and legal stylistics.[27]

A rational appeal on any legal issue follows the generalization for practical argumentation that philosopher Stephen Toulmin identified: Practical rational arguments consist of data or facts and a “warrant” that justifies a move from those facts to some conclusion.[28] Reframed in terms of the law, a rational legal appeal requires a set of operative facts and a legal standard—a rule, principle, policy or the like—that calls for a certain outcome in the presence of those facts.[29] In principle, we could see this pattern as a deduction:

Major premise: If operative facts, then normative consequence.

Minor premise: Operative facts.

Conclusion: Normative consequence.[30]

Of course, requiring a deductive rule here oversimplifies legal reasoning in important ways, because an argument may not articulate a major premise in terms of a clear deductive rule; the warrant may be a policy such as the best interest of a child or a principle such as judicial efficiency; and the argument may articulate the major premise only in the outcomes of precedent cases where no clear rule emerges from them.[31]

Toulmin’s insight was that practical argumentation is “field dependent”—that is, argumentation in different fields may have conclusions of different logical types, and “there are common standards applicable in [their] criticism . . . .”[32] So an argument from any field, for example, that would support an assertion about who should “be a member of the U.S. Davis Cup team . . . or for adopting Fröhlich’s theory of super-conductivity . . . is . . . tested against its own appropriate standard.”[33]

Deductive arguments, when they are available, are “field-invariant”—they work in any field—because they are “analytical,” that is, the conclusion of the argument does not convey any new information.[34] Consider this well-worn example:

Major premise: All men are mortal.

Minor premise: Socrates is a man.

It is hardly necessary to utter the conclusion that Socrates is a mortal, because it provides no new information. The two premises necessitate that conclusion.

There is no question that deductive reasoning is welcome among law judges and advocates. A legal rule is a premise in a deductive argument, like the major premise in a syllogism.[35] For example: ‘Under municipal ordinance, anyone who operates a vehicle in a municipal park is guilty of a misdemeanor.’ This states a rule in the form “[a]ccording to legal authority J, in every instance with features f1 . . . fn, legal category A applies,”[36] or just “‘[w]henever OF [operative fact(s)] then NC [normative consequence],’”[37] or ‘OF→NC.’ If the prosecutor combines this with the syllogism’s minor premise, the operative facts that the defendant did indeed operate a vehicle in a municipal park, then they may seem to compel the normative consequence that the defendant is guilty of a misdemeanor.

But law, like most human endeavors, permits non-deductive arguments. Consider this example:

In a previous case before the highest court in this jurisdiction, the defendant was found liable under circumstances similar to this case; consequently, the defendant in this case will likely be liable.

This seems like an entirely commonplace and acceptable argument in the law, though it is by no means deductively valid or sound. Would such an argument work in the science of epidemiology? How about this:

Recently in this community, an eight-year-old child contracted the COVID-19 virus but did not develop significant symptoms; the child before me is eight years old and has contracted COVID-19 and consequently will likely not develop significant symptoms.

Even one not trained in science can recognize this is not an acceptable scientific argument. Two reasons present themselves for the differing acceptability for these arguments. On the practical-normative level, lawyers are likely to find the first example acceptable and scientists are not likely to accept the second; in short, field-dependent norms dictate whether the form of the argument is acceptable. On the metanormative level, the reason that the argument form is reasonable in law and not in science may be that law and science reach conclusions of different logical forms. In law, the conclusions are about what legal consequences will flow from operative facts; the court’s judgment on an issue is a declaration that brings that very state of affairs into being. When scientists offer an observation about a scientific fact, they are making an assertion about how the world is.

A judge or lawyer’s argumentation about a legal issue generally consists in their assertions about what the law is, what the facts of the instant case are, and how the former apply to the latter. Conventionally, no U.S. judge should write an opinion to justify a decision, and no lawyer should write a brief to urge a particular outcome, without using previous court opinions, often just called ‘cases,’ to support the argumentation in them. Of course, other authorities affect the arguments: Statutes and regulations may supply some or all the legal rules, and secondary sources may supply interpretive frameworks or arguments that lawyers and judges find compelling, or at least moderately persuasive. This particular study focuses only on the use of cases as authorities, however.

This study gathers empirical evidence for one theory about how judges and advocates use cases—the legal topoi.

Professor Hanns Hohmann, an argumentation theorist and scholar of communication and the law, considered how to adapt the classical stases of Greek and Roman rhetoric to contemporary legal argument.[38] Hohmann argued that the rational arguments lawyers make in civil litigation fall into one or more of three “general dimensions of argument”—“operative, regulative, and optative.” Professor Susan Provenzano and I have argued that these dimensions are related to the classical stases in a slightly different way and that Hohmann’s dimensions are akin to the classical topoi, categories of arguments applicable in certain circumstances.[39]

Summarizing our perspective: A legal rule, standard, or principle—Toulmin’s warrant and similar to the major premise of the deduction—applied to legally relevant facts—Toulmin’s data and similar to the minor premise—resolves every legal issue. The topoi categorize the warrants and data that the law uses to resolve issues. We renamed them “empirical,” “conventional,” and “values-based” and defined them as follows:

[The] empirical topos refers to arguments grounded in observations about the world, including observations about factual events and legal texts, and rational inferences about them. The conventional topos refers to arguments that employ legal standards to categorize or interpret facts or law in a certain manner. The values-based topos covers arguments that appeal to underlying legal norms or real-world consequences, but within the constraints that our legal system imposes on such policy-oriented and emotionally appealing arguments.[40]

We proposed—“provisionally”—that normative appeals are more rationally compelling than conventional ones, and conventional more than values based.[41] We did not provide a metanormative justification for this hierarchy, and our empirical evidence for it consisted of references to selected cases, not a systematic study.

This study extends that work. Before the present study, I had intuitions about how lawyers and judges used cited cases in their arguments, and I could point to artifacts where argument proponents did as I expected. Those intuitions are grounded in the cases that I have happened to read over the years and my felt sense of what is typical in them, not in a systematic collection and analysis of arguments. Based on this theory, if lawyers and judges share this metanorm as a practical norm, we would expect them to prefer to make more arguments using rules, fewer using case examples or legal analogies, and even fewer asserting policies. Before the present study, however, there were no data that showed the presence or magnitude of these differences or how those types of arguments work together. In fact, there were no data to support the claim that those are generally the only types of rational uses to which advocates and judges put case precedents. We also did not know the extent to which they cited cases for tactical appeals.

Thus, at the root of the present study is the question, ‘How do lawyers and judges make use of cited cases to support their arguments?’ It’s also a really big question. Even narrowing it to explore this one theoretical problem leaves a pretty broad range of options for exploration. Consequently, I developed these research questions:

Research Questions

RQ1. To what uses did the judges and attorneys put citations to previous cases in their opinions and briefs in the artifacts selected for study here?

RQ2. With what relative frequency did they use cases in these ways?

That the answers to these questions matter is on some level obvious: The ways that judges and lawyers make and justify their arguments and decisions have profound impacts on all our lives. Understanding those practices in light of theories of reasoning and argumentation is thus critical for understanding this country’s laws and its society.

There are practical and pedagogical implications as well. Understanding what counts as typical for legal arguments is practically important for lawyers making those arguments. If lawyers present their arguments in unexpected ways, they may be harder for the reader to comprehend, resulting in a failure of the communication and its argument.[42] Understanding what counts as typical for legal arguments is practically important for legal pedagogy as well. Legal textbooks and books on legal theory targeted at law students make assertions about how legal arguments do or should work, and those assertions should be consistent with what lawyers typically do, or they risk disadvantaging the students who believe in them. For example, Professor Fredrick Schauer’s Thinking Like a Lawyer purports “mostly . . . to introduce beginning and prospective law students to the nature of legal thinking.”[43] Regarding the use of cited cases, however, he asserts that one either must use an explicitly stated rule from the case or infer from the case what that rule would be before applying it to the current legal problem.[44] Some textbooks on legal analysis make similar claims.[45] If these views are not consistent with an empirical assessment, we should not teach them to law students; if empirical data support them, so much the better.

As Part I.B. shows, there have been no answers to these research questions from an empirical perspective. They remain unknown.

B. Relevant Literature

My review of bibliometric, legal, and argumentation scholarship did not reveal satisfactory answers to the research questions. The scholarship in applied linguistics and writing studies suggested some methodological perspectives but did not address legal argumentation. None of the existing research methods was adequate for answering these research questions

Numerous theories in bibliometrics exist for why scholars cite other works.[46] Studies exist, too, to explore how students cite works in school writing to introduce topics, support claims, and compare and contrast findings of cited works with each other and with students’ own projects, among other purposes.[47] There are nods, too, to a rhetorical motivation in each kind of writing.[48] Sociological and bibliometric theories of reference and citation explain them in functional, normative, and phenomenological terms.[49] These perspectives do not provide insights to the questions in this study because they do not address the litigation context—where the arguments function for different purposes than those in academic papers.

Within the legal scholarship, there is a rich theoretical discussion about legal analogy and legal deduction.[50] Law scholars have written a great deal about legal reasoning, precedents, etc.[51] But there is very little empirical work examining how lawyers and judges actually use cases in their arguments. The legal literature is replete, though, with citation-focused studies.[52] Nearly all of it relies entirely on noting whether an opinion has cited an authority anywhere at all, or not—a scale that I will explain below is inadequate for this project. Like many efforts in the legal academy, such studies focus on pecking orders: the “most influential” law reviews, law-review articles, judges, courts, opinions, etc.[53]

Some studies purport to focus more on how or why judges make choices about the authorities they cite. Professor John Merryman’s study was an early one, later updated.[54] Merryman offered numerous proposals for why judges cite authorities: shoring up the institutional legitimacy of the court by providing certainty or predictability; to recognize the parties’ reliance on precedent; supporting the veil of impartiality over the law or the judge that precedent provides;[55] maintaining “the fiction of an abstract law which is mechanically applied by its instrument, the judge”; satisfying society’s expectations or demands “that the judge decide cases in the traditional judicial manner”;[56] and satisfying natural drives “toward certainty and simplicity” with “law [that] appears to be stable, certain and ascertainable through consultation of the appropriate writings (authority).”[57] He claimed that citations to authorities advanced efficiency objectives.[58] Finally, he argued that judges have a habit or natural propensity to cite.[59] His evidence for this, however, was only the frequency with which the California Supreme Court and individual judges on it cited various authorities.[60] It is difficult to see how he could draw these conclusions from those data.

Other studies are interesting because they have used citation counts to argue about (a) whether courts use cited cases to decide the law rationally or as window-dressing to legitimize their decisions,[61] (b) how judges use scientific and social scientific research in opinions,[62] and (c) whether judges cite the same cases that parties’ briefs do.[63]

All these studies attempted to answer their research questions by counting associations between cited authorities and citing artifacts—court opinions or advocates’ briefs.[64] In other words, if an artifact cites an authority anywhere, one time or a dozen, to support one point or a dozen, it still just counts as one citation. Some of the scholars noted the limitations of this approach, acknowledging that actually examining the content of the opinions could answer more of their questions.[65]

Some studies of legal argumentation in the law do look at more granular data. For example, Professor Nina Varsava has explored the stylistic practices of Justices Kavanaugh and Gorsuch in opinions before their appointments to the Supreme Court.[66] Professors Lance Long and William Christensen explored whether the use of intensifiers, such as “clearly” and “obviously,” correlated with the success of appellate briefs[67] and whether an author “faced with an argument that a legal writer believes—or knows—she is likely to lose, the writer will tend to write in a style that uses more intensifiers.”[68] This focus on stylistic matters, however, addresses the tactical appeals that I described above and does not address the rational appeals in legal arguments.

Argumentation theory similarly has a rich tradition theoretically addressing the use of cited cases or precedents in legal argumentation.[69] The studies of actual court opinions there, however, tend to involve close reading of a very small number of opinions.[70] I hope, however, to answer the descriptive question—How do authors use cited cases in their arguments?—generally at some scale.

Outside the law and argumentation theory, there is work in the field of writing studies and applied linguistics, particularly in the area of genre theory, that accounts for the use of cited authorities. For example, applied linguist John Swales analyzed scholarly research articles to identify their genre characteristics.[71] He segmented the articles into typical sections, like introduction, methods, results, and discussion.[72] Within the sections, he analyzed certain rhetorical “moves.” A move “in genre analysis is a discoursal or rhetorical unit that performs a coherent communicative function in . . . discourse.”[73] This might be a group of sentences, a sentence, or even a clause.[74] Swales considered the structure of discussion sections in research articles and the moves there, including background information, statement of results, and references to previous research.[75] The last of these, reference to previous research, is most like citing cases in a legal text, and Swales asserted that authors use them “for purposes of comparison with present research and . . . for purposes of providing support for present research.”[76] Others have extended this work in the field of technical communication, studying the ways that students made use of cited sources to situate their own work in technical-report writing.[77] Neither of these genre-theoretic approaches subjected legal communication to analysis.[78]

Counting citations does not supply an adequate method for assessing how, functionally, authors use citations to previous opinions in their legal argumentation. Thus, answering the research questions identified above will not benefit from merely identifying those court opinions that a legal argument cites. Cases may also be used in diverse ways in distinct parts of an artifact. An extended example may help bring the categories alive and show concretely how they can overlap. Consider, Hosseinzadeh v. Klein,[79] artifact 17.03.00 in this study, where the court analyzed fair use. While discussing the first fair-use factor, the court cited the U.S. Supreme Court case of Campbell v. Acuff–Rose Music, Inc.[80]

“The central purpose of this investigation is to see . . . whether the new work merely supersedes the objects of the original creation, or instead adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message.” Campbell, 510 U.S. at 579, 114 S.Ct. 1164.[81]

Here, we can read the use of Campbell as an effort to offer authority for the rule or policy statement contained in the quoted passage. The citation is, of course, also necessary to explain the use of the quotation marks. But the same opinion cites Campbell again, in the section discussing the fourth fair-use factor:

The question is whether the allegedly infringing work serves as a “market substitute” for the allegedly infringed work, not merely whether the market for the allegedly infringed work was harmed. Campbell, 510 U.S. at 592, 114 S.Ct. 1164 (noting that critical parodies may legitimately aim at harming the market for a copyrighted work, and that “a lethal parody, like a scathing theater review, kills demand for the original [but] does not produce a harm cognizable under the Copyright Act.”).[82]

Here, it seems that the Hosseinzadeh court cites Campbell in support of its assertion of two other rules: One regarding the extent to which the court must consider market harm, and the other (in the explanatory parenthetical) the rule that the harm a parody does to the market of its original is not cognizable under the Copyright Act. A few lines later, the court continues: “Accordingly, ‘the role of the courts is to distinguish between biting criticism that merely suppresses demand and copyright infringement, which usurps it.’ Campbell, 510 U.S. at 592, 114 S.Ct. 1164.” This seems to support the previous rule about parodies and market effects.

But how should we characterize the function(s) of Campbell in this opinion? Hosseinzadeh cites one case; it uses that case to prove three legal rules and to support three quotations from the cited case; the rule and quotation functions co-occur, with two citations together supporting one rule and two supporting one rule each.

For the present study, therefore, I concluded that I would need to segment textual artifacts in some fashion to determine how the cited cases function in different parts of the instant argument, and I would need to determine some means for counting what phenomena are present. Analysis on the sentence-by-sentence level—or word-by-word, as some of the stylistic studies have done—was likely to produce too many units of analysis, particularly because many sentences in an artifact—opinion or brief—do not refer to cases at all.[83] Looking for Swalesian moves or steps would have been problematic in part because it is difficult to define move or step boundaries.[84] I describe the solution to that problem and the methods I used for coding the resulting argument segments in Part II.

II. Materials & Methods

It would be impossible to examine how all judges and advocates have used cases in all situations. Consequently, I selected a narrower set of materials. Part II.A. explains which artifacts I chose to study and why. I had to develop new methods to analyze these artifacts, because as Part I.B. showed, the methods of previous studies were not sufficient to answer my research questions. The methods I employed were first to segment the arguments to analyze, a process described in Part II.B., and then to assign coding categories to the case uses, a process described in Part II.C. Readers not interested in methodological details about this study may wish to jump to Part III for the findings of the study and proceed to Part IV for interpretations of them.

A. Data Selection & Collection

This section describes the strategy for selecting and method for collecting data for this study and the resulting corpus. I chose just one area of law because I suspect there are genre-theoretic[85] concerns about mixing case types. Each body of law may represent its own genre(s) in legal briefs and opinions. Consequently, lawyers arguing about family, criminal, contract, and other bodies of law may employ different practices to make their points. Variations across such genres might function as noise in analyzing any particular practice across them. Better would be to study one subject-matter area and then sample across other types of case to see whether patterns observed here are present there. I chose copyright fair use, mostly because it is interesting to me.

Fair use is a doctrine under copyright law that permits a secondary user of a copyright-protected work to make use of it in a way that would otherwise be copyright infringement.[86] It is an affirmative defense that defeats the rights holder’s infringement claim.[87] It requires a balancing of four factors on a case-by-case basis:

(1) the purpose and character of the [secondary] use, including whether such use is of a commercial nature or is for nonprofit educational purposes; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work. [88]

Fair use, and copyright generally, is also exclusively subject to the jurisdiction of federal courts.[89] The choice of fair use raises some questions or limitations relating to this study.[90] Though I address some of them in the findings and discussion below, only further research will resolve the others. I chose district courts as the source for these artifacts because in the federal hierarchy, trial courts provide more access to the day-to-day work of lawyers and judges.[91] In practice, what this means is that there is far more legal reasoning and argumentation at the trial-court level simply because there are more cases adjudicated there, and each case may require many stopping points or stases where the parties and the judges consider legal arguments.[92] Appellate courts, on the other hand, hear fewer cases, and it is possible that the kinds of arguments there will be different because of the different role of these courts in our system.[93] If we are exploring practical norms relating to the use of cases in arguments, we should consider those places where most of the arguments are happening.

The data for this study were 199 artifacts, consisting of portions of fifty-five court opinions and 144 of the parties’ briefs that led to them. The opinions in question are reported opinions of federal district courts resolving dispositive motions relating to copyright fair use.[94] I selected reported opinions randomly from the period January 1, 2012, through December 31, 2018, from those coded in Westlaw’s key numbering system with key 99, section 53.2, “Fair use and other permitted uses in general.”[95] For each opinion, I checked the docket and the opinion itself to identify which motion papers were related to it. Many opinions resolved more than one motion implicating fair use, as, for example, when the parties cross-moved for summary judgment on that question.[96] In each instance, I pulled the non-moving party’s opposition memo but did not pull further reply briefs.[97] I believed it was important to have a round (or two turns) of argument where possible on the fair-use issues. If a moving party used a court’s opinion in its brief in support of a motion (or in support of its fair-use defense), it might be important to see whether and how the non-moving party used the same opinion in its argument. For purposes of coding here, the artifacts consisted only of the portions of the briefs that addressed fair use.[98] I downloaded the opinions from Westlaw and the applicable briefs from Bloomberg Law. I followed the procedures in the coding guide.[99]

For each artifact, I recorded the forum court; the procedural posture of the case; the identity of the brief’s author and, in the case of a law-firm author, the approximate size of the firm;[100] whether the artifact was a judge’s opinion, a rights holder’s brief, or a secondary user’s brief;[101] whether the rights holder or secondary user prevailed in the opinion;[102] whether each brief was in support of a motion (usually for summary judgment or to dismiss for failure to state a claim), in opposition to it, both, or neither;[103] and whether the client of the brief’s author was the prevailing or non-prevailing party on the fair-use issues discussed in it. The Appendix describes the resulting corpus and data.

As I noted in Part I.B., I needed to segment the argument texts analyzed here to determine how judges and advocates cited cases in different parts of an artifact. To address this issue, I used an idealized hierarchical text-segmentation strategy suggested by rhetorical structure theory (RST), with each proposition that constitutes a claim, called a “nucleus,” supported by other propositions, called “satellites.”[104] RST idealizes the textual argument as having a main claim (sometimes also called a conclusion or standpoint) and sees all the parts of the text as falling into collectively exhaustive and mutually exclusive subsidiary arguments. A statement that receives support is a nucleus, in RST-theoretic terms, and each statement that supports it is a satellite. One can think of it as a form of outline or tree view of an argument, with each level consisting of discrete sub-arguments supporting the conclusion immediately above it.[105] Legal opinions and briefs are well suited to hierarchical textual analysis because canonical perspectives in law-school teaching recommend hierarchical structure in legal writing.[106]

I identified coding spans, the sections of each argument we would analyze closely.[107] Second, we recorded the number of words and the raw number of citations to court opinions in the coding span.[108] I refer to these as ‘raw citations’ in the findings below.[109] Third, we broke the text below each nucleus or claim down into satellites or subpoints in support of that nucleus. We continued by treating each satellite or supporting claim as a nucleus or principal claim that was in turn supported by other satellites.[110] The lowest level of satellite (that is, the satellite that is not also the nucleus for segments of arguments below it) is the point at which the text identifies some legal rule, principle, policy, or label and then shows how the facts in the instant case relate to it, without recourse to analysis of some subpoint within it. This might be the lowest application of what legal-writing teachers call “CREAC,” “IRAC,” or some other acronym that indicates the structure of giving a legal rule, principle, or standard, explaining it, sometimes with case examples, and applying it to facts in the instant case to reach a conclusion.[111] That might involve application of a legal label, such as ‘the secondary user’s use is a parody’; or assessment of a legal factor or subfactor, such as ‘the secondary works are not transformative’ or ‘the secondary user’s commercial use does not support a finding of fair use.’ As a consequence of these efforts, it was possible in each case to extract a tree or outline of statements that the brief supports with arguments.[112]

These efforts resulted in a division of the 199 artifacts in the study into 1810 distinct argument segments, 459 of them in the opinion artifacts and 1351 of them in the brief artifacts.

After segmenting the arguments in the coding spans, we identified the first citation to each case in each argument segment. The result was a coding unit that I call a ‘case use.’ A case use is a correspondence between one segment of an artifact’s argument and a single cited case and encodes the argumentative purpose(s) of citing that case in support of the argument in that section. An argument segment may cite the case more than once, but there is only one case use per segment per case cited.[113]

C. Coding Categories

Given that I divided the artifacts in this study’s corpus into argument segments and that each case cited in an argument segment represents a case use, I needed to decide what categories to assign to these case uses. During a pilot study described in subpart 1, it became evident that judges and advocates were using cases for a relatively small number of purposes, and that they could use each case for more than one purpose in a single argument segment. The pilot study provided a tentative answer to Research Question 1, which asked to what purposes the judges and attorneys put citations to previous cases in their arguments.[114] I concluded that judges and advocates use cases to support assertions about legal rules, provide examples of application of legal rules, assert policy justifications for outcomes, generalize about prior cases, and support quotations from cited cases; I observed few other common uses. I formalized the results of the pilot study in the coding categories described in subsection 2, which I used for the broader study.

1. Pilot Study (answering Research Question 1)

I worked to develop a typology for the uses that authors of court opinions and legal briefs made of the opinions that they cited. I worked with a research assistant, and we began simply by reading dozens of opinions and briefs and asking ourselves, ‘What’s going on here?’ ‘What is this author trying to achieve by citing this opinion?’ We started with detailed descriptions of what we saw, we discussed them with each other, and gradually, we settled on categories into which these uses normally fell.

We immediately saw that an argument section within an artifact might use a single cited opinion in multiple ways. This is consistent with previous studies.[115] But we could have called most of the uses we saw rational appeals.[116] The authors might make them in the following ways: First, the author might assert the existence of a legal rule before applying that rule to the instant case. Citation to a prior opinion would support the claim that the rule exists in the form the author asserted. Second, the author might assert that courts in previous cases have generally taken a particular approach. This is akin to asserting a rule, but somewhat less of a commitment to its universal nature; it suggests a rule of probability or rule of thumb without asserting it universally. Third, the author might describe a previous case, including facts about it and its outcome, to permit the author then to argue that the instant case should come out the same (opposite) way because the cases were (dis)similar. Fourth, the author might cite a prior opinion to assert that a policy goal or consideration supports a particular outcome.

The four rational uses of cases that the pilot study identified correspond to the topoi described above. First, authors who cite a case for a legal rule are using an empirical topos, often quoting an opinion when asserting what a rule of law is. Citation to a prior opinion supports the claim that the rule exists in the form the authors assert. Second, the use of a case example, where the author describes a previous case, including facts about it and its outcome, then argues that the instant case should come out the same (opposite) way because the cases were (dis)similar, is a conventional topos.[117] Though that approach is common in legal reasoning, it does not compel a conclusion to the same extent that the first does. Third, the author uses a values-based topos when citing a prior opinion to assert that a policy goal or consideration supports a particular outcome. Though this is still a rational appeal, courts are reluctant to apply policies directly to facts, as that places them in the role of legislatures; the possible exception is policies arising in and from the rules of court themselves, where judges have broader discretion. A fourth category of case use, asserting that courts in previous cases have generally taken a particular approach, did not fit easily into the three legal topoi, but it was common enough to warrant its own category here. Such a rule of probability or rule of thumb requires an inferential step from ‘several have been decided in a certain way’ to ‘this case should be decided the same way.’ It is thus something like an empirical argument, but the number of cases cited to make a generalization was sometimes as small as one, and that does not satisfy any commonly accepted form of inductive reasoning; consequently, such generalizations seem to be conventional (or field-dependent) in the law.

This exhausts the list of commonly observed rational appeals.

Conventionally, an author should support any quotation from a case with a citation to that case. The status of a case quotation under the legal-topoi model is complex. These case uses often had rational functions, as every quotation qualifies as a de facto empirical argument about what some legal text says. It may nevertheless require conventional or values-based arguments to determine what the legal text means. Supporting rational arguments, however, was apparently not the only purpose for which authors quoted opinions. Authors often used quotations in ways that seemed less designed to appeal to the reader’s reason and more to the reader’s emotion or sense that the writer was speaking authoritatively—in short, tactical appeals.[118] We also identified some case uses that either did not fit into the categories above or that appeared to be doing something specific in addition to those typical uses. Consider the parting quote/cite and stitched application/quote, fairly common practices discussed below.[119]

2. Coding Categories & Process Used in this Study

Based on the pilot study, I developed the following approach to coding the case uses. Again, recall that ‘coding’ here refers to the process of assigning category codes as described here, not to writing computer code. Note that references in this section to artifacts in this study transition to being references to documents and coding units according to the numbering scheme in the coding guide rather than to typical case and case-document citations. Thus ‘14.01.00’ is the court’s opinion in case 14.01, and ‘14.01.10’ refers to a brief in the case 14.01 with the PACER/ECF identification number 10. References to case uses within an artifact use the ‘CU’ abbreviation—for ‘case use’ (or ‘coding unit’). So ‘14.01.00 CU01’ refers to the first case use in the opinion 14.01.00.[120] The coding guide further describes and illustrates the coding categories themselves.[121]

Two coders coded each artifact.[122] Coders reviewed an entire argument section in which a coding unit appeared because the case cited in the coding unit may have been cited elsewhere in the same argument section. Coders coded the case use for all the uses made within the same argument section. They indicated which, if any, of five identified uses the author was making of a case in each portion of the argument and marked all that applied. For example, if the author cited the case to support a quotation at one point in an argument section but as an example at another point in the same argument section, I instructed the coder to count both use types. There were five defined uses to which arguments put cases and a catch-all category for uses that did not fit in the five. Each case use thus received at least one code.

We coded a case use as rule if the author cited the case to support a legal rule that the author asserted in the text. A rule is a statement that could be applied to facts to derive a legal outcome. It usually comes in the form of operative facts lead to/compel/suggest normative consequence or normative consequence results from operative facts. For example, ‘If a secondary use is a fair use, there is no liability for infringement.’[123] The assertion of a rule did not need to be entirely clear about the direction a normative consequence will go based on operative facts.[124]

We coded author citations to a case to support an assertion about what courts often, usually, or generally do as generalization. Generalizations thus commonly began with certain phrases, such as ‘Courts have traditionally . . .’ or ‘Courts have routinely . . . .’ Such cue phrases were not necessary, though. It was common for a string cite to follow a generalization, and in that event, each of the cases in the string cite was usually supporting the generalization.

Where the author cited a case to support a claim that some policy underlies the law in this area, we coded it as policy. A policy statement tells us the why of a rule or legal principle. It might generally be understood to be the complex assertion that some consequence is good and that some rule or practice is justified because it brings about that consequence.[125]

Where the author cited a case to support a quotation from the case in a section of the argument, we coded it as quotation. Coders marked this option even if the quotation appeared only in a parenthetical after the case citation.

Where the author cited a case as an example of a court deciding a case in this body of law and for which the author provided some facts from the cited case, the cited case’s outcome, or both, or for which the author at least impliedly compared or contrasted facts of the case with the instant case, we coded it as example. Sentences referring to cases that begin with ‘In Case Name . . .’ often offered examples. The ‘In Case Name . . .’ heuristic did not always work, though.

If we believed the argument segment cited the case for some other purpose (whether or not we marked it for any of the five previous purposes), we marked the case use as other use and added a comment in the note field in the database indicating what purpose we thought it was. Sometimes it was unclear what use the author was making of a case. Many times, the addition of a parenthetical explanation to a citation could have helped a great deal. Other times, the use was clear, but it was not among those in the coding scheme.[126]

Part II has described the selection and collection of data for this study and described the text segmentation and coding strategies developed for it. Part III offers the findings.

III. Findings

In this study, I identified 1810 argument segments in the arguments relating to copyright fair use[127] in a total of 199 artifacts, trial-court opinions and lawyers’ briefs that gave rise to them. I then coded each case use in each section to categorize it according to whether it proved a rule, supported a generalization, provided an example of a previously decided case, asserted a policy justification for a rule, supported a quotation, or served some other purpose, with many case uses serving more than one of these purposes.

I have assessed the findings along several dimensions to answer two research questions:

RQ1. To what uses did the judges and attorneys put citations to previous cases in their opinions and briefs in the artifacts selected for study here?

RQ2. With what relative frequency did they use cases in these ways?

The research questions responded to theoretical questions raised in Part I, and particularly whether the legal-topoi theory Professor Provenzano and I put forward finds evidence in these data. Section II.C. provided a preliminary answer to RQ1 and the coding categories used to answer RQ2.

Part III.A first looks at the artifacts and the frequency of case uses within them to answer RQ2. There I note that all legal authors use cases to support rule assertions relatively frequently. They use them as examples of how the law is applied less frequently, and they use them to support policy statements less frequently still. This fits with the theoretical perspective of the legal topoi.[128] But practices vary significantly between advocates and judges, between advocates on the winning and losing side, and between advocates supporting and opposing motions.

Finally, Part III.B tracks back to the first research question. In Part I.C, I developed a list of coding categories based on the pilot study. But after coding more than 5600 case uses, it is possible to review those coded other to see if there are any common (or even uncommon but interesting) types of case use not in the original taxonomy. The uses that I found made rational and tactical appeals.[129]

Part IV interprets these findings.

This section reports the findings of this study regarding the practices of judges and advocates using cases in artifacts—opinions and briefs. It first provides an overview taking into account all the artifacts in the study. From a genre-theoretic perspective, it is also useful to consider whether different authors, responding to different social situations, may have made different choices about how to use cases.[130]

I categorized artifacts and report findings here based on several of these potential community divisions. For example, judges’ opinions (n = 55) have different audiences—the parties and courts of appeals—and different purposes—making new law, justifying a decision, discouraging appeals, withstanding appellate review—than do the briefs of the parties’ attorneys (n = 144). Attorneys writing briefs may face different challenges depending on whether their cases were strong (measured here by whether they prevailed on fair use (n = 62) or not (n = 54)).[131] It is also possible that authors writing briefs seeking summary disposition of fair-use issues through motions (n = 68) write with a different focus than those opposing such motions (n = 62).[132] And finally, the practices of attorneys and district court judges in a jurisdiction where copyright cases are more common—the Southern District of New York (n = 96)—might differ from those in other regions (n = 103).

I gathered the applicable data and examined these subgroups of artifacts to ascertain if there were statistically and practically significant differences among them along those dimensions.[133] The Appendix describes the statistical tests in some detail, but in simple terms, statistical significance measures whether the differences between subgroups are real, and practical significance measures whether they are large enough to be interesting.

1. Characteristics Across All Artifacts

There were a great many differences among the case files and artifacts in terms of their lengths and the number of raw citations in them.[134] Between 62% and 84% of raw citations were typically case uses. This means that most authors in these artifacts did not cite an opinion more than once in a single argument segment. To be able to compare the artifacts, I normalized most of the variables that I assessed by calculating their relative frequency, measured as frequency per thousand words. So, for example, if an artifact was 5,834 words long and had 33 case uses, it had 5.66 case uses per thousand words.[135] The mean number of case uses per thousand words of argument was 7.99 and the median was 7.88.[136]

The central objective of this study has been to characterize the uses lawyers and judges make of cases. Using the coding scheme developed above, we can see the relative frequency with which the writers of these artifacts made the use of opinions to support rules, make generalizations, provide examples, support policy statements, and support quotations.

Figure 1
Figure 1.Frequency of Case-Use Types per Artifact (N = 199)

Figure 1 depicts the findings in a box plot.[137] It shows the significant difference in relative frequencies between rule and quotation case uses and the others. Indeed, if we compare the inter-quartile range (IQR) or typical range of rule case uses (3.26 – 6.90) and quotation case uses (3.82 – 6.99) per 1000 words, we can see that they overlap very substantially, occurring with similar frequency across the sample. Typical ranges of case uses coded as policy (0.31 – 1.85 / 1000 words) and generalization (0.00 – 1.36) were the least common, again overlapping very substantially with each other. Occupying the gap between the most and least common case uses is the typical range of case uses coded as example (1.45 – 3.73 words). I interpret these findings in Part IV.

2. Opinion Artifacts vs. Brief Artifacts

Because judges’ opinions have different audiences and purposes than parties’ briefs, it was likely that they might exhibit different case-use characteristics. As it happens, lawyers and judges made use of cases as rule and example with significantly different frequencies. The frequency of case uses to support a rule is greater in opinions than in briefs; the opposite is true for example case uses, where opinions use them less frequently. Figure 2 presents the data.

Figure 2
Figure 2.Case Uses in Briefs (n = 144) & Opinions (n = 55)

Consequently, there are practically significant differences in the frequency of use between briefs and opinions in two of the case-use categories: rule and example. The apparent difference between briefs and opinions on the other three categories is more modest and not, in my view, practically significant.[138] Taking the rule category first, the medians were noticeably different between briefs and opinions. There is also a difference in dispersion of the values in that the relative frequency of rule units was typically in the range of 2.99 – 6.37 / 1000 words, while in an opinion, the typical range was 4.68 – 7.772.[139] Overall, then, it is fair to say that the typical relative frequency of case uses to support a rule is greater in opinions than in briefs. The opposite is true for example case uses, where opinions use them less frequently; the typical range for briefs being 1.70 – 3.84, and for opinions 0.89 – 2.50. I discuss possible interpretations of these differences in Part IV.

3. Briefs from Prevailing vs. Non-Prevailing Attorneys

Lawyers with winning cases—that is, cases where the law and facts favor them—may have a tendency to use different arguments than lawyers who have losing cases. Figure 3 shows that there are a few significant differences in these data. It is pretty easy to see that the briefs of prevailing parties included more case uses of every kind. As Table 6 shows, that difference was the only one between subgroups on case uses overall that was practically and statistically different. The differences are practically significant for quotation (prevailing IQR 4.79 – 7.30 / 1000 words; non-prevailing IQR 2.85 – 6.28) and somewhat less so for example (prevailing IQR 2.17 – 4.48; non-prevailing IQR 1.14 – 3.81) and policy (prevailing IQR 0.35 – 2.03; non-prevailing IQR 0 – 1.45).[140]

Figure 3
Figure 3.Briefs of Prevailing (n = 62) & Non-Prevailing (n = 54) Parties

Of course, this correlation tells us nothing about causation. We cannot say that a brief with more case uses is more likely to cause a court to rule in the party’s favor. Neither can we claim, based on these data, that having a losing case makes it harder for advocates to find cases to use in their arguments. And we cannot rule out other factors that might cause both a party’s victory and a propensity to use more cases in briefs. I discuss the implications of these findings in Part IV.

4. Briefs from Moving vs. Non-Moving Attorneys

It is possible that advocates in an offensive stance—those making motions—use cases differently than advocates in a defensive stance—those opposing motions. As it happens, there were practically significant differences between these two kinds of advocates in two types of case use. Moving parties use rule case uses more often than non-moving parties (moving-party IQR 4.15 – 6.58 / 1000 words; non-moving, 2.71 – 5.90). They also use quotation more often (moving-party IQR 4.33 – 7.34; non-moving, 2.92 – 6.22). I interpret these findings further in Part IV.

Figure 4
Figure 4.Briefs Making (n = 68) or Opposing (n = 62) Motions

5. S.D.N.Y. Artifacts vs. Other Jurisdictions

Recall that fully ninety-six of the 199 artifacts in this study come from the United States District Court for the Southern District of New York. This is a function of that court sitting in the traditional publishing center of the country. A random selection of copyright cases is much more likely to present examples from there than from any other district. Is it possible that artifacts from this copyright-rich jurisdiction would exhibit case-use characteristics different than those from other jurisdictions?

Figure 5
Figure 5.Artifacts from S.D.N.Y. (n = 96) or Other Courts (n = 103)

I used the Southern District as the basis for my comparison, as only six artifacts from this study that were from the Second Circuit were not from that district, and I expected that other districts courts in the Second Circuit might be more like district courts in general than they are like the Southern District. I examined whether the practices of advocates and judges in this court differed from those in other parts of the country.[141] Figure 5 displays my findings.

As it happens, there are some significant differences between authors writing in the Southern District of New York and those writing elsewhere. The differences in quotation (S.D.N.Y. IQR 4.33 – 7.95; other courts, 3.60 – 6.19) and policy (S.D.N.Y. IQR 0.49 – 1.98; other courts, 0.11 – 1.55) appear practically significant.[142] Because this study focused on one kind of argument before these courts—fair use in copyright cases—we cannot say from these data whether the propensity to use cases more as quotation and policy in the Southern District of New York is restricted only to cases of this kind or runs the gamut of cases before that court. I interpret these findings in Part IV.

B. other Case Uses

Everything old is new again, they say. After tackling Research Question 2 in the previous section, I want to return to Research Question 1. Based on the pilot study described in Part II, I chose the coding categories that I have been discussing in Part III. It would be fair for the reader to ask, however, how many of the units later coded in the study did not fall into any of the categories I previously identified in the pilot study. Perhaps we overlooked some category in the pilot study that we should have had from the beginning? Recall that coders applied the other code to indicate that the author appeared to be using a case for something other than, or in addition to, the five categories I identified in the pilot study: rule, generalization, policy, example and quotation. As the quotation code commonly appeared in all case uses, regardless of the rational or tactical appeal the author appeared to be making, I decided to review instances of other that appeared in conjunction with the four rational appeals, and those of other that appeared alone or only with quotation.

To prepare this section, I reviewed all the case uses coded as other and the coder notes associated with them to see if any common (or uncommon but still interesting) uses appeared among them. Of the 5638 coding units, 637 or 11.3% received an other code, while only 383 or 6.8% were coded either as quotation, as other, or as both without also being coded as rule, generalization, policy, or example. This examination is less systematic and more preliminary than the previous section, and so I will avoid making many quantitative or comparative assessments.[143]

First, the most common other use, present in nearly a quarter of them and commonly associated with the one or more of the four rational appeals, was what I call the ‘stitched application’ or ‘stitched quote.’ Next in frequency, at around 6%, was what I call the ‘parting quote’ or ‘parting cite,’ a tactical appeal. Four rational appeals appeared in between 2% and 4.5% of the other units: distinguishing a case, citation to an earlier opinion in the instant case, use of a case to interpret another case’s holding or a statute, and use of a case to assert some non-legal fact about the world. I discuss these uses briefly below.

1. The Stitched Quote or Application

One other case use stood out as being very frequent, marking some 24% of all case uses in the study. I call it the ‘stitched quote’ or ‘stitched application.’ It was widely dispersed, appearing at least once in ninety-three different artifacts. In the stitched quotation form, the advocate would make a claim about the instant case, weaving it together with a quotation, again usually from an authoritative case. In this situation, the case use was also coded with one of the other rational appeals. An example will be helpful.

In the quoted material below, the author, a secondary user and defendant opposing a motion for preliminary injunction made by Fox Broadcasting Company, the rights holder, discussed the fourth fair-use factor. The suit was before the District Court for the Central District of California. The second sentence of the excerpt shows a stitched quote/application, where the author asserted that Fox had a burden but finished the sentence with a quotation of a Supreme Court case, Sony.

12.07.71 CU14–CU16[144]

The fourth factor requires consideration of “the effect of the [defendant’s] use upon the potential market for or value of the copyrighted work.” 17 U.S.C. § 107; CU14 Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 590 (1994). Because the consumer’s use is noncommercial, Fox bears the burden to show “by a preponderance of the evidence that some meaningful likelihood of future harm exists.” CU15 Sony, 464 U.S. at 451; CU16 Hustler Magazine Inc. v. Moral Majority Inc., 796 F.2d 1148, 1155 (9th Cir. 1986) (“when the use is noncommercial, the copyright owner must demonstrate by a preponderance of the evidence that there is ‘some meaningful likelihood of future harm’”).

Just as in Sony, however, Fox “fail[s] to carry [its] burden . . .” 464 U.S. at 451. In Sony, the content owners argued, in much the same vein as Fox here, that their economic advertising model was in severe danger and that a huge threat existed that, if the VCR were permitted, there would be an end to all television programming as we know it. As set forth in the Rapp Declaration, the entertainment industry has a habit of claiming that the sky is falling when new technology is introduced. Rapp Decl. ¶¶41-57. As history now shows, the frantic claims (“Boston Strangler”) of doom were wrong. The Court in Sony considered all such arguments and concluded that Universal had failed to carry its burden. It endorsed the district court’s view that Universal’s evidence of harm in that case was “speculative” and/or “minimal.” 464 U.S. at 454.

From the stitched quote, the reader could not tell whether Sony stated a rule to the effect that a copyright owner bears the burden of proving market harm when the secondary user shows its use was non-commercial. The coders here might have inferred that it stood for that rule based on the citation to and quotation from a different opinion, Hustler Magazine, but that opinion was not binding upon this court. The coders therefore did not code Sony as a rule. They did however, code it as an example, based on the author’s use of it in the second paragraph.[145]

An author might use the same pithy phrase in a stitched application as a rational appeal by couching conclusions about the instant case in the actual words of a prior, often authoritative opinion; and as a parting quote, where only the value as a tactical appeal is evident. Compare these two excerpts:

18.03.37 CU64

Moreover, Defendant’s copying of the Photograph was unnecessary to achieve Hearst’s news reporting purpose. As noted by Judge Easterbrook in CU64 Kienitz v. Sconnie Nation LLC, 766 F.3d 756, 759 (7th Cir. 2014), “[t]he fair-use privilege under § 107 is not designed to protect lazy appropriators.” In this case, there were at least four legal alternatives available to Defendant which did not include the “lazy appropriation” of Otto’s work: (1) Hearst could have published its informative story without any photo whatsoever; (2) Hearst could have published the Article using other photographs of President Trump…[146]

This excerpt is from a rights holder’s brief, the part of the argument addressing the third fair-use factor, how much of the original work the secondary user took. The citation to the Kienitz case (which was not binding before this New York district court) stopped a bit short of being a rule, as it did not define what counts as ‘lazy,’ and this was the first reference to that case in the brief. Nevertheless, the author proceeded to present instant facts that the author believed would paint the secondary user as a lazy appropriator. Compare the second excerpt.

15.05.00 CU14

The net result of this Court’s first factor analysis is unavailing. On the one hand, the transformative purpose and non-commercial nature of the Flyer weigh in favor of Defendants. On the other hand, Defendants did not need to use the Photograph in order to effectuate their criticism, and the fair use privilege “is not designed to protect lazy appropriators.” Id. [citing Kienitz]

In this excerpt, from a court’s opinion in the Northern District of Illinois, where Kienitz is binding, the court had previously cited the case for a couple points unrelated to the one made here. But here, at the close of the argument section, the judge stitched the quote from Kienitz, the first time we see this language in the opinion, into a statement about the instant case. Perhaps the judge wished only to grab a pithy phrase from a cited case to wrap up the argument section.

In fact, uses of cases as stitched quotes and stitched applications accounted for more of the other uses in judges’ opinions (around 29%) than in advocates’ briefs (around 23%).

2. The Parting Quote or Citation

A common technique both among judges and advocates was what I call the ‘parting quote’ or ‘parting citation.’ These were quotations or citations to authorities that were either binding (and perhaps well known) or offered pithy expressions with which an argument’s proponent could end their argument. Making up more than 6% of the other case uses (and more than 1% of total case uses), this practice was fairly widely dispersed, appearing in forty-five different artifacts.

Consider this example from the Southern District of New York, which is the rights holder’s entire argument regarding the second fair-use factor:

17.06.115 CU 16–18

“Jimmy Smith Rap” is a Vivid and Personal Recollection by Jimmy Smith of the Album. Defendants claim that “Jimmy Smith Rap” is a “thin” copyright because it is “a short statement of a purportedly factual nature with a significant amount of unprotected material.” (Def. Memo. pp. 14, 17). Pound Cake does not simply copy the facts that the record company provided champagne in the studio, or that the A&R men told them what to record, but Jimmy Smith’s subjective recollection and comments. As the Supreme Court stated in CU16 Feist Publications, Inc. v. Rural Tel. Serv. Co., 499 U.S. 344, 348 (1991): “Others may copy the underlying facts from the publication, not the precise words used to present them”. Defendant’s reliance upon CU17 Swatch Grp. Mgmt. Serv. Ltd. is misplaced because the Court there was faced with the publication of a recording of a financial meeting at which non-copyrightable financial facts and figures were announced. There was no personal expression or comment involved.

Jimmy Smith Rap is clearly a creative expression which lies “close to the core of copyright protective purposes.” CU18 Campbell supra, at 586. The second factor therefor favors Plaintiffs.

Here, CU16 supports a rule statement from a Supreme Court case, which the brief’s authors implied is inconsistent with the secondary user’s conduct described in the previous sentence. CU17 is an example used to distinguish a Second Circuit precedent previously cited by the secondary user. But what is the purpose of citing Campbell, another Supreme Court case, in the second paragraph? The brief’s authors did not assert any rule, policy, or generalization from Campbell; nor did they supply facts from Campbell to use it as an example. This parting quote can only function to wrap up a discussion of the instant case, apparently cloaking it in authoritative language from a high court. It is difficult to see the rational value of the use of this quotation, but it is not difficult to see the tactical or purely persuasive value: Citing an authoritative case as a parting shot in an argument segment may appear to give more weight to the writer’s conclusion.

Again, use of cases as parting quotes and parting cites accounted for more of the other uses in opinions (around 8%) than in briefs (around 6%).

3. Other Rational Appeals

The four miscellaneous rational appeals that occasionally appeared in the other units were distinguishing a case, citation to an earlier opinion in the instant case, use of a case to interpret a statute or another case’s holding, and use of a case to assert some non-legal fact about the world. The first of these, distinguishing a case, typically co-occurred with the example rational appeal for the simple reason that they were typically used as counterexamples. Consider artifact 17.02.46, supporting the secondary user’s motion for judgment on the pleadings on grounds of fair use, arguing that its work was a parody of The Grinch Who Stole Christmas.

17.02.46 CU30–31[147]

[The secondary user] did not merely use elements of Grinch to get attention or to avoid the drudgery in working up something fresh. CU30 Campbell, 510 U.S. at 580. Therefore, the case that [the secondary user] anticipates [the rights holder] will rely upon, CU31 Dr. Seuss Enterprises, L.P. v. Penguin Books USA, Inc., 109 F.3d 1394, 1396 (9th Cir. 1997), is distinguishable. That [ . . .] decision, which affirmed a preliminary injunction in favor of [the rights holder], concerned a poetic account of the O.J. Simpson double murder trial entitled The Cat NOT in the Hat! A Parody by Dr. Juice, which the [secondary user] alleged was a fair use of the well-known The Cat in the Hat by Dr. Seuss. Id., at 1396. The Ninth Circuit rejected the [secondary user’s] argument that the work at issue was a parody, finding that “the substance and content of The Cat in the Hat is not conjured up by the focus on the Brown-Goldman murders or the O.J. Simpson trial.” 109 F.3d at 1401. Here, in contrast, [the secondary user’s] writing not only mimics Grinch’s rhyming style, but the text has a critical bearing on the substance and style of Grinch. [The secondary user] did not randomly select any-old [sic] popular work to tell a story about a topical news event, like the OJ Simpson trial. The [secondary work] is a Christmas-themed work that conjures up Grinch and comments on Grinch as part of a newly created original work of authorship.

Here, the author anticipated that the rights holder’s opposition to this motion would cite Dr. Seuss and distinguished it even before the rights holder could strike. We coded this case use as an example, because the author provided facts about Dr. Seuss and its outcome and then explicitly contrasted it with the instant case. The vast majority of case uses coded as other that I found were distinguishing were also coded as example.

An exception to that rule, and a poorly executed version of this appeal, appears in artifact 17.07.32, where the entire discussion of the second fair-use factor consists of these two sentences:

17.07.32 CU10

The second factor (the nature of the copyrighted work as creative and expressive and not factual) indisputably weighs in [the rights holder’s] favor. The Work is expressive and does not merely recite facts like the names and addresses in a phone book. CU10 See, e.g., Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, 344-45 (1991).

This author had not cited Feist previously and never cited it directly again. Probably the only way the coders would have recognized this as an example is if one or both of them had known that Feist was a case about copyrights in a telephone book. That did not happen, and they coded it only as other. But because I am familiar with Feist, I believe the author intended this paragraph to distinguish the instant case from Feist.[148] As a consequence of the co-occurrence of example with this kind of other, I consider the latter a subcategory of the former. In fact, given my sense that drawing case distinctions is very common and expected, it’s possible that the coders in many instances just coded them as example and never noted an other use in the file. That might explain the relative sparsity of this type of other, which appeared in only nineteen of the artifacts (less than 10%) in this study.

One rational appeal, evident in about 4% of the other case uses (and less than 1% of total case uses) was citation to an earlier opinion in the instant case. So, for example, in 15.09.00 CU27 the judge cited her own earlier opinion in the same case for a proposition upon which she relied in her present opinion. This phenomenon occurred only where a case had a sufficiently long history for there to be prior published opinions, and only in nine (less than 5%) of the artifacts in this study.[149] Were I to revise the coding guide for this study, I would likely give this type of case use its own category tag.

A third miscellaneous rational appeal is the use of a case to interpret another case’s holding. This appeal marked less than 4% of the other case uses and was dispersed across only twenty-seven (less than 14%) of the artifacts in this study. A good example is a footnote in artifact 18.03.52, where the author attempted to discount cases their opponent cited or rules their opponent derived from them.

18.03.52 CU76–79

Plaintiff’s citation to CU76 Richard Feiner & Co., Inc. v. H.R. Industries, Inc., 10 F. Supp. 2d 310, 314 (S.D.N.Y. 1998) is inapposite because the image at issue in that case was used to introduce a section for special effects advertisers and not for any of the preamble uses in § 107 of the Copyright Act, and in any event that decision was vacated by CU77 Richard Feiner & Co., Inc. v. H.R. Industries, Inc., 182 F.3d 901 (2d Cir. 1999).

Here, CU76 cited the trial-court opinion in Feiner and immediately distinguished it on its facts. We coded it as example and other (distinguished). The author then cited the appellate opinion in Feiner in CU77 to explain to the instant court the weight it should give to the trial-court opinion: none. We coded it as example—because the facts and outcome were known—and other (interpretation). Anecdotally I observed that advocates often distinguished their opponents’ cases in footnotes, though I did not try to observe the practice systematically. This is perhaps to downplay the presence of those cases in the author’s brief, but it may be risky to commit one’s effort to distinguish the opponent’s case to a part of the authors brief the reader may never read. Other instances of interpretation include ones where the author would comment on the history or importance of some inquiry.[150]

The final miscellaneous rational appeal was the use of a case to assert some non-legal fact about the world. For example, several artifacts cited Campbell v. Acuff-Rose Music, Inc.,[151] for the proposition that paradigmatic fair uses like news reporting and criticism are carried out for profit in the U.S., supporting the argument that such activities do not automatically fail on the first fair-use factor.[152] This type of other appeared in only twenty case uses across the whole study (less than 0.5%).

The six appeals discussed in this section accounted for more than 500 of the case uses that we marked as other. There were a few other peculiarities, including points where we simply could not figure out why an author was citing a case, but they did not make up a substantial portion of the other case uses or of the total case uses.

IV. Discussion

This study set out to begin answering the question: ‘How do judges and advocates use cases in their arguments?’ It did so by asking two specific research questions relating to a corpus of 199 artifacts, the copyright fair-use arguments in fifty-five court opinions between 2012 and 2018 and the 144 advocates’ briefs that gave rise to them:

RQ1. To what uses did the judges and attorneys put citations to previous cases in their opinions and briefs in the artifacts selected for study here?

RQ2. With what relative frequency did they use cases in these ways?

This Part first discusses the answer to RQ1, then generally discusses the extent to which the data here were consistent with the theory of legal topoi Professor Provenzano and I advanced, described in Part I. This Part also discusses the differences in relative frequencies of certain types of case uses between opinions and briefs and between briefs written by authors in different socio-legal positions. Finally, it considers limitations of the study and suggests further research.

A. Uses Authors Made of Cases

The answer to Research Question 1 is that judges and advocates in these artifacts tended to cite previous court opinions to make rational appeals, the most common being to support the assertion of a rule, policy, or generalization or to illustrate the application of a rule with an example.

The most common use of a citation, though, was to support a quotation from a cited case. This type of case use can be mixed in terms of being a rational or tactical appeal.[153] In many instances, this was an empirical rational appeal, as that term is defined above.[154] When authors cited cases to support their characterization of a rule, generalization, or policy, for example, they commonly quoted some language from the cited case in that support—that observation about the cited text is empirical. A quotation offers rational advantages over a brief or opinion writer’s paraphrase of the underlying authority: Assuming the language is correctly quoted, the reader need not trust the argument proponent’s interpretation of the language, as a paraphrase would require them to do. The purely tactical (or purely persuasive) value of language from an authoritative source was not lost on authors, however, who sometimes appeared to use quotations for no other reason than to cloak a statement about the instant case in language and a citation from an authoritative source—what I have described as the ‘parting quote.’[155]

The only purely tactical appeal commonly observed was in fact this same parting quote or parting cite, where an author cited a case when wrapping up an argument section. Often the case cited or quoted was binding and influential, and often the author chose to quote language that was pithy and stitched together with some statement about the instant case. This type of case use does not contribute to the cogency of an argument, but it may contribute to its success. Of course, authors of these artifacts had many ways other than citing cases to make tactical appeals, including using thoughtful legal stylistic choices, storytelling, and other techniques.[156] The findings of this study do not suggest that the authors were not making tactical appeals, just that they tended not to use citations to previous opinions to do so. An extension of this study’s work could examine all the tactical appeals in a set of artifacts.

Among artifacts in this study, uses of cases to support rule statements, generalizations, policy arguments, examples, and quotations varied in relation to their strength in the theory of legal topoi.[157] Uses with higher rational appeal according to that theory—empirical appeals like rule and quotation—were more common than those with modest or lesser appeal—conventional appeals like example and values-based appeals like policy.

I propose that this hierarchy is based on the argument proponent’s goal of appearing to mobilize fewer interpretive resources in making the argument. In the terms of linguistic and philosophical pragmatics, this is the extent to which they must account for a variety of possible meanings—or possible implicatures—and rule out all those that will not apply.[158]

When a writer presents an argument based on the actual words of a rule from an authoritative text (like a binding court opinion), the writer apparently interprets the text less than they would to apply some prior case precedent without a rule. As an empirical matter, the rule statement either is or is not in the authoritative text. The words of the rule statement are usually relatively small in number—often just a sentence; they are usually couched in the language of the law, using applicable terms of art; and they usually provide a fairly clear statement in the form of operative factsnormative consequence. Alternative hypotheses about the meaning of the rule—possible implicatures—are relatively small in number.[159] The use of rule is an empirical rational appeal, in that characterization of a rule as some set of operative facts leading to a normative consequence permits authors (after offering evidence of the operative facts) to move inexorably to the conclusion they desire. It requires no further interpretation. This analysis explains the relative frequency both of rule and quotation case uses.

Applying an authoritative case precedent as a legal analogy using the common-law method, however, is considerably more complex. The example case use follows the pattern of legal analogy: A previous case had similar facts and came out a certain way; this case should thus come out the same way. This form of reasoning is conventional in law, though it is not in all fields.[160] It is empirical in the sense that it contains an assertion that the law and facts cited are indeed present in an authoritative text. But here, the argument’s proponent must draw from among all the facts in the cited and instant cases those comparisons that are relevant to application of the law and conclude that any differences are not so relevant as to distinguish the precedent. The proponent may or may not feel the need to apply what Professor Schauer refers to as “categories of assimilation”[161]—that is, they may or may not feel the need to propose a covering rule that brings the instant case into alignment with the cited case.[162] If they do so, however, they have the option to draw the category of assimilation at various levels of abstraction, all of which the precedent case may equally strongly implicate.[163] Alternative hypotheses about the meaning of the text are considerably more numerous, and the need to choose imposes higher interpretive costs on the author and calls on the reader to trust the author more. The authors of the artifacts in this study perhaps recognized, therefore, that conventional appeals based on legal analogy inspire less confidence than those based on bright-line rules. Case uses coded as example appeared less frequently than rule, but more frequently than other types.

Making a policy argument moves even deeper into interpretive territory. On the model discussed above, policy arguments in the law are values-based arguments, and they require the greatest interpretive effort on the parts of authors and readers to connect the arguments about the instant case.[164] They are not so much arguments about what the law is, but rather why it is or what it should be. The argument’s proponent may choose from a large number of texts, including ones not normally authoritative in law, such as literary and religious texts, may search for motivations behind statements of rules and decisions in previous cases, and so on. The possible implicatures the author could draw from such materials are vast in number, making it very difficult to reject most meaning determinations and settle on one. Accepting a policy argument requires the reader to repose considerable trust in the author’s interpretive performance. Uses of cases to support arguments coded as policy were substantially less common than the other types so far discussed.

Finally, we have the category of generalization, the least frequently used, which is complicated by a couple problems. First, the coders on this project found it difficult to agree when to assign this code.[165] The category should capture argumentative uses of cited cases that “support an assertion about what courts often, usually, or generally do.”[166] It was common, however, for one coder to see an expression like “commercial use generally weighs against fair use” as a generalization while another might see it as a rule (possibly subject to exception or countervailing considerations). Second, the generalization is something like a cross between an example and a rule, making its rational argumentative appeal less certain. Given enough examples of cases similar in a certain way to the instant case that come out a certain way, the author or reader might attempt to universalize the similarity as the operative facts of a rule statement. Here, the generalizations appeared to happen in cases where the author was not quite prepared to claim that the statement was a rule. On the other hand, generalization sometimes appeared in case uses where the author cited only one case; it’s difficult to see how a single example supports a generalization across cases.

In summary, uses by authors of cases in this study were generally consistent with the hierarchy of rational appeals that Professor Provenzano and I previously described, with rule and quotation units, which make empirical appeals, each about twice as common as example units; example units, which make conventional appeals, about twice as common as policy units; and policy units, which make values-based arguments, quite rare. The generalization units were rare even by those standards, though, probably because they are neither fish nor fowl (that is, neither rule nor example) from writers’ and readers’ perspectives.

In short, assuming in each case that the argument’s proponent has accurately represented the authorities on which the argument relies, the reader must trust the proponent more to have correctly interpreted those authorities if the argument is values-based or conventional than if the argument is empirical. It may be, however, that advocates and judges choose to use cases, at least in part, due to considerations other than persuasive effect.[167] In other words, perhaps the most frequent is not necessarily the most effective. This question would benefit from further research of the kind described below.

B. Variations Across Genres or Communities

This study showed that there were practically and statistically significant differences in relative frequencies of the types of case uses described here depending on whether the author was a judge or advocate, prevailing or non-prevailing advocate, party moving for or opposing a motion, and writing in the Southern District of New York or elsewhere. As Toulmin noted, argumentation forms are field-dependent.[168] Genre theory suggests that the typical communicative or argumentative response to a typical social situation will vary with the discourse community in which it occurs.[169] Swales offered some defining characteristics of a discourse community.[170] Considerably more research may help to explain the difference, but I offer some speculations here to set the agenda for that research.

First, why might judges use rule more frequently and example less frequently than advocates? When judges write opinions to justify their decisions, they are in a considerably different position than the advocates who argue that the judges should take one position or another. The judge’s decision has binding effect on the parties, until and unless one or more of them appeal the decision. But most appeals fail, probably because within the options available, judges usually provide plausible justifications for the outcomes they select. Advocates, on the other hand, petition the judge to select their arguments over those of their opponents. They may feel it is necessary to provide a wider range of arguments to buttress their positions.

Judges also represent institutional interests, and at least in this sample, they are appointed for life tenure. As a consequence, they may wish for their decisions to be perceived as logically compelled rather than a judgment call (so to speak) between the positions of the parties. Judges in this situation may seek to appear as a matter of institutional credibility to engage less in interpretation of the law and more in its application. As I explained above, the quotation and rule case uses arise in the empirical legal topos, which is to say that they require least interpretation. Judges use cases to set out legal rules more often than advocates do.

Advocates, on the other hand, offer more examples, certainly in an effort to guide the judge in interpreting the rules that the court must apply. In addition to seeking to make decisions that appear logically compelled, judges may be more reluctant to rely on examples than advocates because the legal effect of a precedent case may be arguable enough to result in a higher likelihood of reversal on appeal. So, for example, a trial-court judge may be concerned that the court of appeals will distinguish a case upon which the trial-court judge relies for an example. It is an easy matter for the appeals court to say that the cited case is not relevantly similar, or is relevantly dissimilar, to the instant, so that the holding in the cited case is not compelled in the instant case. For the appeals court instead to contradict the trial court’s statement of a rule might require the appeals court to offer an exception and a rationale for it.

One reviewer of this article suggested that an advocate trying to persuade the judge might use examples as a way of showing the judge that their peers have already ruled the advocate’s way in other cases, making the judge feel safer in making that determination in the instant case.[171] Judges, on the other hand, are not motivated to show that any herd mentality governs them but rather wish it to appear that legal rules compel their decisions.

Prevailing advocates used quotation, example, and policy significantly more often than non-prevailing advocates. I have described these differences as both statistically and practically significant. The former is a conventional measure and the latter a statistical threshold that I selected for this study. One question that immediately presents itself is whether readers would note such differences. That requires further study. If these differences are visible to the skilled legal reader (like a judge or other advocate), it may be true to say that a brief with certain characteristics looks like what could fairly be called a ‘losing brief,’ potentially giving it diminished credibility. But this study does not provide evidence that losing briefs cause litigation losses. It may just as plausibly be that weak cases make only certain types of arguments practicable. There may also be some other cause(s) mediating both the form of the briefs and the outcomes of the cases.

Moving parties’ advocates used rule and quotation more often than non-moving parties’ advocates. These differences may well be a function of the burden placed on the moving party: to offer sufficient legal authority for its position and show the court that it is entitled to relief. The non-moving party need only show that the moving party’s argument is sufficiently infirm so as to make the relief it seeks unavailable. As a consequence, the non-moving parties may be focusing their arguments on attacking the arguments of the moving briefs rather than attempting to make independent arguments relying on other cited cases. Nevertheless, we might expect that the non-moving party would need to call on authorities for legal rules just as frequently as the moving party, and the difference observed here prompts inquiry into their practices.

Finally, the question of geography is an interesting one. The authors in the Southern District of New York—a jurisdiction commonly associated with copyright actions—used more quotation and policy than authors in other jurisdictions. Their greater use of example was also statistically but not practically significant. There is an extent to which we might expect practice to be fairly consistent across federal districts, but I suspect too that no one would be surprised to hear that lawyers in Texas, for example, write differently than lawyers in California. The community of lawyers practicing before the federal bar may well have subcommunities, and as genres are properties of (sub)communities, it may well be that certain practices become commonplace in some but not others. It may also be that copyright cases in SDNY draw judges and advocates who are more experienced or interested in copyright matters and that this conditions in some way the kinds of arguments they make.

C. Limitations & Future Research

This study is a start toward answering the big question: How do judges and advocates use cited cases in their arguments? But it is also subject to some limitations. First, it looked only at artifacts relating to one subject matter in the law, focused on one level of the federal judiciary. This was a deliberate choice, because other legal subject matter and other levels of court might exhibit systematically different practices. The only way to see that is to study them separately. The choice of copyright—a fact-intensive field with roots in judge-made policy—might have inflated the extent to which advocates and judges made use of example and policy here. Study of a more rule-based area of the law would help to assess the magnitude of any such difference. I suggest too that further study should examine artifacts from appellate cases, initially at least, in the same field—copyright law—but also more broadly. Appellate judges have different roles and are subject to different constraints than district-court judges. They may also tend to have different kinds of experience before reaching the bench.

I have described some differences noted above as practically and statistically different. Others can check my math on statistical and practical significance by examining the data.[172] It is impossible to say now, however, whether the typically trained legal reader would detect those differences, whether I have categorized them as practically significant or not. The question matters: If readers recognize the differences between briefs written in the winning and losing styles suggested by these findings, lawyers would be well-advised, whatever the quality of their cases, to emulate the briefs of prevailing advocates. If the distinction is invisible to readers, it may still be of theoretical interest without being something that advocates lose sleep over. Experimental study—exposing readers to different densities of case uses in short texts—would be valuable to answer that question.

As for the other potentially different communities of authors—opinions vs. briefs, moving vs. non-moving parties, SDNY authors vs. others—and the differences in their practices those communities mediate, further research might explore judges’ and advocates’ practices using social-scientific methods such as qualitative interviews, surveys, and ethnography. These efforts could provide clearer insight into the differences.

The data from this study also have further insights to offer. The corpus and data here provide an opportunity to examine both the arguments advocates make and the reactions of their opponents and judges. For example, some scholars debate whether using a legal analogy without asserting a deductive rule in its application can result in a cogent argument. This study has indeed identified instances where authors have used legal analogies (examples here) without rules. A deeper qualitative examination of these data will show whether such arguments exhibit apparent flaws in reasoning and whether they generate normative censure from opposing advocates and judges.[173] Similar qualitative deep dives are possible regarding the use of rule and policy arguments more generally. Case uses also illustrate questions of legal theory by allowing the researcher to show how a case the moving party cites elicits a response from the opposing party and then the judge. This would meaningfully extend a previous study that looked at the uptake of cases cited in briefs—their ‘stickiness’—by examining whether judges cited the same cases.[174] The findings would be important for legal theory and practice given the role of adversarial argumentation as a truth-finding mechanism and its relation to the dialectic nature of legal argumentation.[175] Other avenues open to exploration with the data already collected for this study are whether the patterns are different for cited cases that are binding vs. non-binding and for court opinions that are appealable vs. those that are not.


This article answered two research questions, puzzle pieces in the larger originating question of how judges and advocates make use of cases in their arguments. Though this is the first published systematic, empirical study that explores how lawyers and judges use cases in their arguments, it should not be the last. Legal theorists can make productive use of the corpus of artifacts and data from this study, and they can conduct their own studies with the methods developed here, or with their own extensions of them.

All these findings are significant in that they characterize the practices of these authors in such a way that they may inform metanormative legal theory. This study showed that the advocates and judges here exhibited the preferences that the theory of legal topoi expected: a preference for the empirical topos, followed by the conventional, and finally the values-based topos. And the findings here showed the magnitude of those preferences.

These findings have significance for lawyers, particularly those findings that suggest the brief of a party with a good case might look different than the brief of a party with a bad case. I would caution advocates not to change their writing practices until we can see whether these apparent differences appear in other types of cases and whether they are detectable by legal readers.

Finally, these findings have significance for legal pedagogy. To the extent that teachers of legal theory and writing emphasize a group of approaches on the metanormative grounds that they are best practices, those teachers should know the practical norms to which the profession will expect students to conform once they get out of law school. To the extent that law teachers’ pet methods and metanormative theories differ from the realities of practice, they should at least give students a heads up that what they must do in law school classes may look different—perhaps very different—from what practical norms expect them to do.

The findings in this study invite further research in at least three areas. The first relates to the methods described particularly in this article. I might call the second area the study of ‘reader reception’ of the kinds of texts studied here. And the third takes the rhetorical perspective, seeking to understand how authors think their construction of legal arguments will persuade their audiences.

First, scholars should use the methods described here to study arguments in other parts of the law: substantive law other than copyright, courts other than federal courts, levels other than the trial court, etc. It may also be interesting to see how the methods apply to analyses outside the litigation context: For example, genres like the office memo, to the extent it still exists.[176] If the patterns observed here are borne out there, they point to broad practical norms across American legal practice. If they are not, the findings here may describe only the practices of the types of authors studied here, and we should be able to account for the differences in practical norms based on differences in the relevant communities of argument proponents.[177]

Second, scholars should examine whether the differences in argumentative practice—say between advocates who prevail before the court and those who do not—are noticeable to legal readers. A researcher could begin this work using surveys, presenting skilled legal readers with texts that vary in the relative frequency of use of the various types of case use. It should be possible experimentally to determine whether they even notice the differences described above. Qualitative interviews might also be useful for understanding legal readers’ perceptions of these differences, if any.

Finally, scholars should explore the decisions that proponents of legal arguments make while deciding what is the best means to persuade their audiences, whether it is the court persuading the parties that they should not appeal its decision or the appeals court that it should not reverse it, or it is the advocate seeking or opposing the grant of some relief from the court. Legal argumentation is dialogic, and understanding how these arguments’ proponents anticipate responses from their audiences has important implications for legal theory, practice, and pedagogy.


Section A explains how to obtain the coding guide and dataset associated with this article. Section B offers a simple statistical description of the dataset. Section C explains the use of statistical tests and the data presentation in this article. Section D supplies the tables upon which the figures above are based.

A. Data Repository

“Coding Guide & Replication Data for ‘Precedent as Rational Persuasion’” is a dataset in the Texas Data Repository including the coding guide discussed in this article and all the original textual artifacts and data from which the findings in this study derive.[178] Those materials are available for other researchers, who may use them to replicate, criticize, or extend this study.

Any researcher can obtain these materials by navigating to, selecting all the files, and clicking the “Download” button. The data repository will generate a .ZIP file that the researcher can download. The downloaded file is between 124 and 125 megabytes in size. Uncompressing the .ZIP file places the applicable files into a file structure that makes them easier to navigate than they are on the data repository website.

B. Description of Dataset for this Study

The corpus for this study included sections of arguments from and related to fifty-five court opinions, and for each opinion, between zero and six briefs.[179] I called the opinion and its related briefs a ‘case file.’ In all, the fifty-five case files included a total of 199 artifacts. The histogram in Figure 6 shows how many case files had each number of artifacts associated with it.[180]

Of the opinions, twenty-one (38%) ruled on fair-use issues for the rights holders; twenty-five (45%) for secondary users; and nine (16%) went partially for one side and partially for the other.[181]

Because many of the businesses engaged in producing and distributing creative works in the United States are based in New York and California, federal districts sitting in those states were sources of a majority of the opinions, while several other districts were also represented, as Table 1 shows. While evaluating the findings, I considered whether artifacts from these courts reflected differences from those from elsewhere.[182]

Figure 6
Figure 6.Number of Artifacts Coded per Case File (N = 55)

The portions of those artifacts that we coded—all related to the question of fair use—totaled 767,980 words and represented 8225 raw citations to court opinions. They cited a total of 529 different court opinions. Table 2 summarizes statistics relating to case files and shows the mean number of words per case file was 13,963. The mean number of raw citations per case file was 150.

C. Statistics & Data Presentation

I have avoided tables of data in the text of this article, preferring to reserve them for the Appendix. Most of the tables show interquartile ranges of the variables they present. I present the data this way because the IQR provides a better impression of the typical range of values for a variable. The interquartile range is the range within which the central 50% of the values fall; below the lower value are roughly 25% of the observations, and above the higher value are roughly 25% of the observations.[183] I use these boundaries in this article when describing a value as ‘unusually low’ or ‘unusually high.’ My rationale is that it is helpful for the reader to know more than just whether a value is above average—around half the values in a non-skewed distribution with few or no outliers will be—or below average. In this article, if I say a value is ‘unusually high (low),’ I mean it is outside the IQR and therefore higher (lower) than 75% of the values. Similarly, when I refer to a value as ‘typical,’ I mean only to say that it falls within the IQR. Calculating IQRs also permits visualization of data with a box-and-whisker plot, something I use in the article and explain upon its first use.[184]

Some of the findings presented in Part III show differences between subsets or categories of the data: for example, between briefs and opinions, between texts from the Southern District of New York and from other districts, etc. I focus the presentation of comparative findings in Part III and the discussion in Part IV on differences that I describe as both statistically and practically significant.

Statistical significance really just measures the probability that the difference observed in two samples accurately measures a difference that exists in the broader population. In short, it measures how likely it is that the difference between the two sample values is real. The result is usually expressed with a p-value, with the threshold for significance (called alpha) established by the researcher and commonly accepted to be 0.05 in much social science research. Parametric tests of significance require that the sample have a Gaussian—sometimes called ‘normal’—distribution. I used the Shapiro-Wilk test to assess whether the categories here appeared in a Gaussian distribution.[185] Case uses overall and quotation case uses were in Gaussian distributions, but the coding categories rule, generalization, policy, and example were not. For categories in Gaussian distribution, I used the parametric Welch Two Sample t-test for statistical significance, and for non-Gaussian distributions the Wilcoxon rank sum test with continuity correction, each with alpha = 0.05.

‘Practical significance,’ as I use the term, just means that a statistically significant difference in the measurement of two samples may have some practical importance. The traditional statistical measure for the latter is a test for the effect size of the difference between the two samples. There are many possible tests for effect size. Unfortunately, they are often represented in arbitrary numerical ranges that are hard for the layperson to interpret. Perhaps the easiest to make sense of is Vargha and Delany’s A,[186] which is what I have chosen to use here.[187] In principle, this statistic expresses the probability that a randomly selected member of one sample will be higher than a randomly selected member of the other sample. The higher the value of A, the greater the effect size. I’ve selected a threshold value of 0.62 (rounded to two decimal points) as the cut-off for practical significance.[188] All of the differences reported here that are practically significant are also statistically significant.

As it happens, the threshold I have chosen for practical significance (A ≥ 0.62) corresponds closely to an informal visual test that the reader without statistical training can apply to the box plots in the figures in Part III. Generally, a difference is practically significant when the median in one category is very near or beyond the end of the IQR in another category. Take the example of the rule category in Figure 2. There, the median value for briefs is 4.85, just above the lower end of the IQR for opinions (4.68). In other words, above that threshold are about 75% of the opinions and only 50% of the briefs.[189] Even though the median for briefs is just above the lower end of the IQR for opinions, I would count it practically significant because of the apparent overlap in Figure 2. This eyeballing of the differences generally corresponds to the statistical test of practical significance.[190]

D. Tables

Table 1.Source Courts of Opinion Artifacts in this Study (N = 55)
Court No. of Opinions
C.D. Cal. 7
D. Az. 1
D. Colo. 1
D. Mass. 1
D. Md. 2
D. Mont. 1
D.P.R. 1
E.D. Mich. 1
E.D. Mo. 1
E.D. Tex. 1
M.D. Fla. 1
M.D.N.C. 1
N.D. Cal. 1
N.D. Ga. 2
N.D. Ill. 3
N.D. Miss. 1
N.D. Ohio 1
N.D.N.Y. 1
S.D.N.Y. 24
W.D. Va. 1
W.D. Wash. 1
W.D. Wis. 1
Total 55
Table 2.Statistical Characteristics of Case Files (N = 55)
Mean (std dev) Median Min / Max Interquartile Range
No. of Artifacts 3.62 (1.02) 3 1 / 7 3 – 4
No. of Words 13,963.27 (13,259.41) 8,951 1,457 / 68,551 6,206 – 16,132
No. Raw Citations 149.55 (111.09) 101 14 / 521 71 – 210
No. Case Uses 102.71 (79.51) 72 12 / 405 52 – 134
Table 3.Statistical Characteristics of Artifacts (N = 199)
Mean (std dev) Median Min / Max Interquartile Range
No. of Words 3,859.20 (3,739.10) 2,934 220 / 35,002 1,724.5 – 4,991
No. Raw Citations 41.33 (32.36) 33 0 / 185 18 – 52.5
No. Case Uses 28.39 (22.70) 23 0 / 150 13 – 38
Ratio: Case Uses / Raw Cites 0.72 (0.17) 0.72 0.14 / 1.00 0.63– 0.84
Table 4.Normalized Raw Citations and Case Uses per Artifact (N = 199)
Mean (std dev) Median Min / Max Interquartile Range
Raw Citations / 1000 words 11.35 (3.90) 11.33 0 / 20.33 9.47 – 14.10
Case Uses / 1000 words 7.99 (3.13) 7.88 0 / 17.70 5.97 – 10.16
Rule CUs /
1000 words
5.28 (2.69) 5.25 0 / 15.33 3.26 – 6.90
Generalization CUs / 1000 words 0.91 (1.10) 0.51 0 / 5.70 0.00 – 1.36
Policy CUs /
1000 words
1.19 (1.10) 0.92 0 / 6.34 0.31 – 1.85
Example CUs /
1000 words
2.63 (1.71) 2.41 0 / 8.92 1.45 – 3.73
Quotation CUs / 1000 words 5.55 (2.57) 5.36 0 / 15.33 3.82 – 6.99
Table 5.Briefs (n = 144) vs. Opinions (n = 55)
(std dev)
Median Min / Max Interquartile Range
Total CUs / 1000 words
Briefs 7.85 (3.31)
Opinions 8.33 (2.55)
Rule / 1000 words **
Briefs 4.92 (2.77) 4.85 0.00 / 15.33 2.99 – 6.37
Opinions 6.22 (2.20) 6.23 0.94 / 11.09 4.68 – 7.77
Generalization / 1000 words *
Briefs 0.81 (1.03) 0.38 0.00 / 4.68 0.00 – 1.23
Opinions 1.16 (1.24) 0.76 0.00 / 5.70 0.06 – 1.73
Policy / 1000 words *
Briefs 1.06 (0.98) 0.85 0.00 / 4.69 0.20 – 1.75
Opinions 1.51 (1.29) 1.21 0.00 / 6.34 0.60 – 2.28
Example / 1000 words **
Briefs 2.93 (1.79) 2.85 0.00 / 8.92 1.70 – 3.91
Opinions 1.86 (1.15) 1.76 0.00 / 5.15 0.98 – 2.46
Quotation / 1000 words
Briefs 5.35 (2.59) 5.18 0.00 / 15.33 3.57 – 6.89
Opinions 6.09 (2.42) 6.01 0.74 / 11.49 4.27 – 7.56

Mean frequencies of case uses and case-use types for which the difference between briefs and opinions is statistically significant at p < 0.05 are designated with a single asterisk (*) and those for which the difference is practically significant at A ≥ 0.62 with two asterisks (**). See supra, Part C of the Appendix, for details regarding assessment of statistical and practical significance.

Table 6.Prevailing (n = 62) vs. Non-prevailing Briefs (n = 54)
(std dev)
Median Min / Max Interquartile Range
Total CUs / 1000 words **
Prevailing 8.71 (2.86) 8.67 2.22 / 17.35 6.80 – 10.49
Non-prevailing 7.12 (3.53) 6.93 0.00 / 17/70 4.81 – 8.25
Rule / 1000 words *
Prevailing 5.44 (2.46) 5.52 1.08 / 14.61 3.84 – 6.43
Non-prevailing 4.68 (3.04) 4.43 0 / 15.33 2.83 – 6.24
Generalization / 1000 words
Prevailing 0.89 (1.05) 0.51 0.00 / 4.68 0.00 – 1.41
Non-prevailing 0.85 (1.14) 0.36 0.00 / 4.42 0.00 – 1.29
Policy / 1000 words **
Prevailing 1.31 (1.12) 0.97 0.00 / 4.69 0.35 – 2.03
Non-prevailing 0.85 (0.83) 0.60 0.00 / 2.88 0.00 – 1.45
Example / 1000 words **
Prevailing 3.26 (1.72) 3.05 0.00 / 8.92 2.17 – 4.48
Non-prevailing 2.58 (1.97) 2.33 0.00 / 8.11 1.14 – 3.81
Quotation / 1000 words **
Prevailing 6.05 (2.18) 5.88 1.91 / 12.94 4.79 – 7.30
Non-prevailing 4.71 (2.81) 4.30 0.00 / 15.33 2.85 – 6.28

Mean frequencies of case uses and case-use types for which the difference between briefs of prevailing and non-prevailing parties is statistically significant at p < 0.05 are designated with a single asterisk (*) and those for which the difference is practically significant at A ≥ 0.62 with two asterisks (**). See supra, Part C of the Appendix, for details regarding assessment of statistical and practical significance.

Table 7.Briefs Making (n = 68) or Opposing (n = 62) Motions
Mean (std dev) Median Min / Max Interquartile Range
Total CUs / 1000 words
Moving 8.35 (3.24) 7.99 0 / 17.35 6.43 – 10.25
Opposing 7.24 (3.20) 7.37 1.32 / 17.70 4.81 – 9.31
Rule / 1000 words **
Moving 5.63 (2.89) 5.42 0 / 15.33 4.15 – 6.58
Opposing 4.25 (2.33) 3.85 0 / 10.55 2.71 – 5.90
Generalization / 1000 words
Moving 0.87 (0.97) 0.55 0 / 3.83 0.00 – 1.46
Opposing 0.73 (1.02) 0.27 0 / 4.42 0.00 – 1.16
Policy / 1000 words
Moving 1.27 (1.11) 0.92 0 / 4.69 0.32 – 1.94
Opposing 0.87 (0.81) 0.73 0 / 2.88 0.23 – 1.56
Example / 1000 words
Moving 3.03 (1.80) 2.99 0 / 8.11 1.71 – 3.86
Opposing 2.72 (1.86) 2.47 0 / 8.92 1.44 – 3.90
Quotation / 1000 words **
Moving 6.00 (2.72) 5.36 0 / 15.33 4.33 – 7.34
Opposing 4.64 (2.22) 4.73 0.45 / 9.78 2.92 – 6.22

Mean frequencies of case uses and case-use types for which the difference between briefs of moving and non-moving parties is statistically significant at p < 0.05 are designated with a single asterisk (*) and those for which the difference is practically significant at A ≥ 0.62 with two asterisks (**). See supra, Part C of the Appendix, for details regarding assessment of statistical and practical significance.

Table 8.Artifacts from S.D.N.Y. (n = 96) or Other Courts (n = 103)
Mean (std dev) Median Min / Max Interquartile Range
Total CUs / 1000 words
S.D.N.Y. 8.35 (3.42) 7.97 0.68 / 17.70 6.03 – 10.54
Other courts 7.64 (2.78) 7.74 0.00 / 17.35 5.97 – 9.29
Rule / 1000 words
S.D.N.Y. 5.25 (2.85) 5.42 0.00 / 15.33 3.17 – 6.87
Other courts 5.30 (2.53) 5.10 0.00 / 14.61 3.34 – 6.90
Generalization / 1000 words
S.D.N.Y. 1.02 (1.20) 0.70 0.00 / 5.70 0.00 – 1.43
Other courts 0.80 (0.99) 0.36 0.00 / 3.71 0.00 – 1.34
Policy / 1000 words **
S.D.N.Y. 1.38 (1.12) 1.22 0.00 / 6.34 0.49 – 1.98
Other courts 1.00 (1.04) 0.76 0.00 / 5.86 0.11 – 1.55
Example / 1000 words *
S.D.N.Y. 2.92 (1.76) 2.60 0.00 / 8.92 1.75 – 4.03
Other courts 2.37 (1.60) 2.12 0.00 / 8.11 1.31 – 3.09
Quotation / 1000 words **
S.D.N.Y. 6.22 (2.78) 6.05 0.00 / 15.33 4.33 – 7.95
Other courts 4.93 (2.18) 4.98 0.00 / 10.74 3.60 – 6.19

Mean frequencies of case uses and case-use types for which the difference between from S.D.N.Y. and those from other courts is statistically significant at p < 0.05 are designated with a single asterisk (*) and those for which the difference is practically significant at A ≥ 0.62 with two asterisks (**). See supra, Part C of the Appendix, for details regarding assessment of statistical and practical significance.

  1. See Richard A. Posner, The Problems of Jurisprudence 90 (1990); Neil MacCormick, Rhetoric and the Rule of Law: A Theory of Legal Reasoning 53–54 (2005). See generally Frederick Schauer, Precedent, 39 Stan. L. Rev. 571 (1987); Cass Sunstein, On Analogical Reasoning, 106 Harv. L. Rev. 741 (1993); Larry Alexander, Bad Beginnings, 145 U. Pa. L. Rev. 57 (1996); Scott Brewer, Exemplary Reasoning: Semantics, Pragmatics, and the Rational Force of Legal Argument by Analogy, 109 Harv. L. Rev. 923 (1996); Lloyd L. Weinreb, Legal Reason: The Use of Analogy in Legal Argument (2d ed. 2016); Brian N. Larson, Law’s Enterprise: Argumentation Schemes and Legal Analogy, 7 U. Cin. L. Rev. 663 (2018).

  2. I use this convention: I use double quotation marks only for material I quote from other authorities. References to the name of a concept or any word or phrase not quoted from other authorities appear in single quotation marks. Italics indicate emphasis. Cf. Brewer, supra note 1, at 931 n.14. So I might write:

    While Smith called these theories “norms about norms,” I call them ‘metanormative’ and acknowledge that neither usage is widely accepted.

    I also adopt the convention of using third-person plural pronouns (‘they,’ ‘them,’ etc.) when referring to individual persons of unknown gender.

  3. ‘Hume’s Guillotine’ refers to the mistake of inferring normative principles solely from empirical observations of what is. David Hume, A Treatise of Human Nature 469 (L.A. Selby-Bigge ed. 1896) (“[T]he author proceeds for some time in the ordinary way of reasoning, and . . . makes observations concerning human affairs; when of a sudden I am surpriz’d to find, that instead of the usual copulations of propositions, is, and is not, I meet with no proposition that is not connected with an ought, or an ought not. This change is imperceptible; but is, however, of the last consequence.”).

  4. It can still be a quite-bad argument, especially if its premises are not true or if it is presented incoherently by its proponent.

  5. Weinreb, supra note 1, at 10 (making the same argument by implication); Larson, supra note 1, at 692–701 (making the argument expressly); Susan E. Provenzano & Brian N. Larson, Civil Procedure as a Critical Discussion, 20 Nev. L.J. 967, 978–79 (2020) (same).

  6. Part I also explores those claims.

  7. This is a broader term than “experimental jurisprudence,” which appears to be the application of experimental empirical methods to legal theory. See, e.g., Kevin P. Tobia, Testing Ordinary Meaning: An Experimental Assessment of What Dictionary Definitions and Linguistic Usage Data Tell Legal Interpreters, 134 Harv. L. Rev. 726 (2020) (examining laypersons’, lawyers’, and judges’ responses to definitions of terms—“ordinary meaning”—under experimental conditions).

  8. See the review of literature in Part II.B.

  9. This ‘coding’ means the assignment of category labels, not authoring computer code. “A code in qualitative inquiry is most often a word or short phrase that symbolically assigns a summative . . . attribute for a portion of language-based . . . data.” Johnny Saldaña, The Coding Manual for Qualitative Researchers 4 (3d ed. 2016).

  10. See infra text accompanying notes 50 through 53.

  11. See infra text accompanying notes 66 through 68.

  12. Typically, I will use the first-person singular pronoun throughout this paper when referring to efforts on this study, as I am its sole author. As the detailed coding guide in the data repository notes, however, most of the text segmenting and coding tasks undertaken here involved two coders working together, usually one of my excellent research assistants on the one hand and me on the other. Brian N. Larson, Coding Guide & Replication Data for "Precedent as Rational Persuasion," Texas Data Repository Dataverse (Oct. 23, 2020), All the data files used in preparing this article and all the textual artifacts coded in creation of those data also appear in the data repository. Id.

  13. See infra Part II.C.

  14. Provenzano & Larson, supra note 5, at 978–79.

  15. See infra Part I.A.

  16. Robert K. Merton, Introduction: Notes on Problem-Finding in Sociology, in Sociology Today: Problems and Prospects ix, xiii (Robert K. Merton et al., eds., 1959).

  17. Id. at xv (describing fact-focused studies in the context of sociology).

  18. Id. at xxvi.

  19. Provenzano & Larson, supra note 5, at 1025. As used here, ‘topoi’ are merely categories of arguments appropriate in certain circumstances, a term that derives from classical rhetoric. Id. at 1005–06.

  20. Larson, supra note 1, at 668 (citations omitted). I consider the argumentation to include both the premises and the conclusion.

  21. Id. Some scholars distinguish the arguments of lawyers from the “justification” of judges. Id. n.21 (citing Schauer, supra note 1, at 571 n.2). As both satisfy my definition of “argumentation,” I cannot support the distinction.

  22. Trudy Govier, The Philosophy of Argument 119 (1999).

  23. Frans H. van Eemeren & Peter Houtlosser, Strategic Maneuvering: A Synthetic Recapitulation, 20 Argumentation 381, 382–83 (2006) (“[D]ialectic is defined pragmatically as a method for dealing systematically with critical exchanges in verbal communication and interaction ‘that amounts to the pragmatic application of logic, a collaborative method of putting logic into use so as to move from conjecture and opinion to more secure belief.’”) I avoid the term ‘logical’ in terms of describing legal arguments and argumentative moves because of various senses in which that term is used. Larson, supra note 1, at 674–75.

  24. van Eeemeren & Houtlosser, supra note 23, at 383.

  25. This concept is adapted loosely from the argumentation-theoretic concept of strategic maneuvering. Id. (“Strategic maneuvering refers to the efforts arguers make in argumentative discourse to reconcile aiming for rhetorical effectiveness with maintaining dialectical standards of reasonableness.”). “Strategic maneuvering manifests itself in argumentative discourse in the choices that are made from the ‘topical potential’ available at a certain stage in the discourse, in ‘audience-directed framing’ of the argumentative moves, and in the purposive use of ‘presentational devices.’” Id. The presentational devices of their definition come closest to my notion of the tactical appeal. See also van Eemeren et al., Handbook of Argumentation Theory 552–55 (2014).

  26. van Eeemeren & Houtlosser, supra note 23, at 387 (“[F]allacies are violations of [rational] critical discussion rules that come about as derailments of strategic maneuvering.”).

  27. See Ruth Anne Robbins, An Introduction to Applied Storytelling and to This Symposium, 14 Legal Writing 3, 3 (2008) (describing storytelling as “the backbone of the all-important theory of the case, which is the essence of all client-centered lawyering”); Steven J. Johansen, Was Colonel Sanders a Terrorist? An Essay on the Ethical Limits of Applied Legal Storytelling, 7 J. ALWD 63, 64 (exploring the “concern that [legal storytelling] may be too powerful or, perhaps, inappropriately powerful”); Shaun B. Spencer & Adam Feldman, Words Count: The Empirical Relationship Between Brief Writing and Summary Judgment Success, 22 Legal Writing 61, 61–62 (2018) (“measur[ing] the statistical relationship between brief readability and case outcomes”); Lance N. Long & William F. Christensen, Clearly, Using Intensifiers is Very Bad—Or Is It?, 45 Idaho L. Rev. 171, 173, 181 (2008) (exploring use of intensifiers “very,” “obviously,” “clearly,” “patently,” “absolutely,” “really,” “plainly,” “undoubtedly,” “certainly,” “totally,” “simply,” and “wholly” and showing “excessive intensifier use in appellate briefs is associated with a statistically significant increase in adverse outcomes for the ‘offending’ party, but in other situations . . . excessive intensifier use was associated with a significant increase in favorable appellate outcomes”).

  28. Stephen E. Toulmin, The Uses of Argument 90–91 (updated ed. 2003).

  29. See MacCormick, supra note 1, at 24.

  30. Id.

  31. See generally Larson, supra note 1 (proposing defeasible argumentation schemes as a model for non-deductive argumentation in law).

  32. Toulmin, supra note 28, at 15.

  33. Id. at 35.

  34. Id. at 116.

  35. Note that I only reluctantly refer to legal reasoning by deduction as a ‘syllogism,’ for reasons I have explained elsewhere. Larson, supra note 1, at 676 n.64.

  36. Id. at 698.

  37. Id. at 676 (quoting MacCormick, supra note 1, at 24).

  38. Hanns Hohmann, The Dynamics of Stasis: Classical Rhetorical Theory and Modern Legal Argumentation, 34 Am. J. Juris. 171 (1989); see also Provenzano & Larson, supra note 5, at 988–1001 (providing a treatment of classical stasis theory, applicable definitions and conventions, etc.).

  39. Provenzano & Larson, supra note 5, at 1005–06.

  40. Id. at 1025. This hierarchy is similar to the hierarchy that Professors MacCormick and Summers proposed for statutory interpretation in their collection of essays taking a comparative approach to that question. See D. Neil MacCormick & Robert S. Summers, Interpretation and Justification, in Interpreting Statutes: A Comparative Study 511, 512–14 (D. Neil MacCormick & Robert S. Summers eds., 1991). In decreasing order of preference, “[l]inguistic arguments” are those based on the meanings of the actual words used in a statute; “systemic arguments” include arguments based on “contextual harmonization,” precedent, and analogy; and “teleological/evaluative arguments” include arguments about legislative purpose. Id. at 512–14; see also Eveline Feteris, Prototypical Argumentative Patterns in a Legal Context: The Role of Pragmatic Argumentation in the Justification of Judicial Decisions, 30 Argumentation 61, 63 (2016).

  41. Provenzano & Larson, supra note 5, at 1025.

  42. See Brian N. Larson, Gender/Genre: The Lack of Gendered Register in Texts Requiring Genre Knowledge, 33 Written Communication 360, 364 (2016). “Genre knowledge comprises components of the communicator’s cognitive environment: her assumptions about communicative behaviors she expects to have a particular effect or effects on a reader based on knowledge about a typified situation in the writer’s cognitive environment.” Id. at 364. If the author of a legal brief, for example, varies from genre conventions, she imposes higher cognitive effort on her reader; the unexpected performance of the author is harder for the reader to comprehend. See id. at 362–65.

  43. Frederick Schauer, Thinking Like a Lawyer: A New Introduction to Legal Reasoning xi (2009).

  44. See generally id. ch. 3.

  45. See, e.g., Christine Coughlin, Joan Malmud Rocklin & Sandy Patrick, A Lawyer Writes: A Practical Guide to Legal Analysis 116 (3d ed. 2018) (counseling argument proponents to precede each use of a case as an example with a “hook” a statement “of the legal principle that the case illustration will clarify and prove to be true”); Richard K. Neumann, Jr. & Sheila Simon, Legal Writing 53 (2008) (framing case analogies with the description of “determinative facts” that unite the cited and instant cases). But see Michael D. Murray & Christy H. DeSanctis, Legal Writing and Analysis 145 (2d ed. 2015) (counseling argument proponents to use “explanatory synthesis”—statement of a principle synthesized from the example case(s) but allowing for merely case-by-case comparison of examples).

  46. Emanuela Riviera, Scientific Communities as Autopoietic Systems: The Reproductive Function of Citations, 64 J. Am. Soc’y for Info. Sci. & Tech. 1442, 1450 (2013) (Citations “pay intellectual debts; . . . are considered as reward devices; they work as concept symbols, standing for shared knowledge; they work as codes and medium of communication.” (citations omitted))

  47. See generally Lee-Ann Kastman Breuch & Brian N. Larson, Research and Rhetorical Purpose: Using Genre Analysis to Understand Source Use in Technical and Professional Writing, in Points of Departure: Rethinking Student Source Use and Writing Studies Research Methods for the Study of Student Writing (Tricia Serviss & Sandra Jamieson eds., 2017).

  48. See, e.g., Bruno Latour, Science in Action 30–44 (1987) (describing the rhetorical functions of citations in the scientific article).

  49. Blaise Cronin, Metatheorizing Citation, 43 Scientometrics 45, 46–48 (1998). Functionalists explain citations as the citing author’s effort to “provide supplementary evidence, to support or refute an hypothesis, to furnish historical context.” Id. at 46. Normative theorists describe the social norms, including “dispensing of [rewards to cited authors] within the scholarly communication system . . . . [and] providing . . . socially appropriate cues and reinforcers.” Id. at 47. Phenomenologists explore how “social-psychological variables [shape] an author’s referencing behavior.” Id. The perspectives are not mutually exclusive.

  50. See generally Larson, supra note 1, and sources cited therein.

  51. See generally Edward H. Levi, An Introduction to Legal Reasoning (1949); Ruggero J. Aldisert, Logic for Lawyers (3d ed. 1997); Larry Alexander & Emily Sherwin, Demystifying Legal Reasoning (2008).

  52. For lengthy, though by no means exhaustive, summaries, see Dietrich Fausten, Ingrid Nielsen & Russell Smyth, A Century of Citation Practice on the Supreme Court of Victoria, 31 Melbourne U. L. Rev. 733, 734–36 (2007); Kevin Bennardo & Alexa Z. Chew, Citation Stickiness, 20 J. App. Prac. & Process 61, 67–75 (2019). Note that in many ways, the study of citational bibliometrics in law much pre-dates that in the broader academy. Fred R. Shapiro, Origins of Bibliometrics, Citation Indexing, and Citation Analysis: The Neglected Legal Literature, 43 J. Am. Soc’y for Info. Sci. 337, 337 (1992).

  53. For example, Sirico and Margulies assessed the frequency with which the U.S. Supreme Court cited legal periodicals. Louis J. Sirico, Jr., & Jeffrey B. Margulies, The Citing of Law Reviews by the Supreme Court: An Empirical Study, 34 UCLA L. Rev. 131, 131–32 (1986). Sirico followed up on this study later. Louis J. Sirico, Jr., The Citing of Law Reviews by the Supreme Court: 1971-1999, 75 Ind. L.J. 1009 (2000). It is unclear whether the researchers there counted a one-to-one relationship between opinion and cited journal, where a journal article would count as one if it was cited anywhere in the opinion, one time or many; or a many-to-one relationship, where a journal article was counted as one each time an opinion cited it. The first study is silent on this question, Sirico & Margulies, supra at 132 n.3, while the latter seems to suggest the many-to-one approach, Sirico supra at 1010 n.9 (“For purposes of consistency, we counted a citation only when the citation included the name of the law journal. For example, we would not count an 'id.'”). For other pecking-order studies, see Fred R. Shapiro, The Most-Cited Articles from the Yale Law Journal, 100 Yale L.J. 1449 (1991) (identifying the most-cited articles from the first 100 years of the Yale Law Journal); William M. Landes, Lawrence Lessig & Michael E. Solimine, Judicial Influence: A Citation Analysis of Federal Courts of Appeals Judges, 27 J. Legal Stud. 271, 271 (1998) (using “the number of citations to the published opinions of judges on the federal courts of appeals to measure the influence of individual judges”).

  54. John Henry Merryman, The Authority of Authority: What the California Supreme Court Cited in 1950, 6 Stan. L. Rev. 613 (1954) [hereinafter Merryman, Authority]; John Henry Merryman, Toward a Theory of Citations: An Empirical Study of the Citation Practice of the California Supreme Court in 1950, 1960, and 1970, 50 S. Cal. L. Rev. 381 (1978) [hereinafter Merryman, Theory].

  55. Merryman, Authority, supra note 54, at 613–22.

  56. Id. at 625.

  57. Id. at 626.

  58. Id. at 622 (satisfying the parties’ reliance on precedent); id. at 624 (reviewing prior decisions for possible solutions to the instant case rather than reinventing the wheel); id. (valuing the “accumulated wisdom and experience” of the previous judges); id. (“disposing of legal problems with relative finality, rather than allowing them to be relitigated to no purpose”); id. (simplifying the judge’s argument, “letting [the judge] dispose of the problem on the basis of conclusions reached in earlier decisions”).

  59. Id. at 625 (transferring “some of the responsibility for [the judge’s] decision to other shoulders than [the judge’s] own”); id. (recognizing the “inertia” of the judicial process); id. (the influence of habit and training on lawyers and judges); id. (satisfaction of the urge “toward symmetry, toward system-building”); id. at 626 (a natural tendency to “[r]espect . . . the opinions of those who have gone before, especially the dead ones”); id. (satisfying natural drives “toward certainty and simplicity” with “law [that] appears to be stable, certain and ascertainable through consultation of the appropriate writings (authority)”).

  60. Id. at 650–72; see also Merryman, Theory, supra note 54 (replete with tables of tallies throughout).

  61. Frank B. Cross, James F. Spriggs II, Timothy R. Johnson & Paul J. Wahlbeck, Citations in the U.S. Supreme Court: An Empirical Study of Their Use and Significance, 2010 U. Ill. L. Rev. 489. They concluded that the answer is not black and white. Id. at 573.

  62. In 1990, Acker considered the extent to which the U.S. Supreme Court used “social science research evidence” in its opinions, and the extent to which the Court found those citations in briefs of the parties or amici or on the Court’s own initiative. James R. Acker, Social Science in Supreme Court Criminal Cases and Briefs: The Actual and Potential Contribution of Social Scientists as Amici Curiae, 14 Law & Hum. Behav. 25, 28 (1990).

  63. Bennardo & Chew, supra note 52.

  64. Cross et al., supra note 61, at 522–25, is interesting in that it also used network analysis, a means of identifying cases that are more “central” in their tendency to be cited, but still based on citation counting. Citation studies in the law also include those where the researcher assesses the impact of a patent by the frequency with which it is cited in later patents. See generally Jonathan H. Ashtor, Does Patented Information Promote the Progress of Technology?, 113 Nw. U. L. Rev 943 (2019).

  65. E.g., Merryman, Theory, supra note 54, at 384; Bennardo & Chew, supra note 52, at 107 n.133.

  66. See generally Nina Varsava, Elements of Judicial Style: A Quantitative Guide to Neil Gorsuch’s Opinion Writing, 93 N.Y.U. L. Rev. online feature 75 (2018),; Nina Varsava, Computational Legal Studies, Digital Humanities, and Textual Analysis, in Computational Legal Studies: The Promise and Challenge of Data-Driven Legal Research ___ (Ryan Whalen ed., 2020), available at SSRN:

  67. Long & Christensen, supra note 27, at 181.

  68. Lance N. Long & William F. Christensen, When Justices (Subconsciously) Attack: The Theory of Argumentative Threat and the Supreme Court, 91 Or. L. Rev. 933, 938 (2013).

  69. E.g., Robert Alexy, A Theory of Legal Argumentation 274–79 (Ruth Adler & Neil MacCormick trans., 1989); MacCormick, supra note 1, at 143–61; C.H. Perelman & L. Olbrechts-Tyteca, The New Rhetoric: A Treatise on Argumentation 351–57 (John Wilkinson & Purcell Weaver trans., 1969) (treating argument from precedent as argument by example). See generally Kevin D. Ashley, Precedent and Legal Analogy, in Handbook of Legal Reasoning and Argumentation (Giorgio Bongiovanni et al. eds., 2018).

  70. For example, Feteris analyzed a single U.S. Supreme Court case. Feteris, supra note 40, at 70–77. Plug looked for linguistic cues in a small number of court opinions to characterize their role in marking text as dictum. José Plug, Indicators of Obiter Dicta: A Pragma-Dialectical Analysis of Textual Clues for the Reconstruction of Legal Argumentation, 8 Artificial Intelligence & L. 189, 192–202 (2000). And Kloosterhuis looked at three court opinions for evidence of certain argumentation schemes. Harm Kloosterhuis, Reconstructing Complex Analogy Argumentation in Judicial Decisions: A Pragma-Dialectical Perspective, 19 Argumentation 471, 475–82 (2005).

  71. A genre is “a class of communicative events, the members of which share some set of communicative purposes.” John M. Swales, Genre Analysis: English in Academic and Research Settings 58 (1990); see also Larson, supra note 42. In the law, there are numerous genres, such as the demand letter, complaint, trial motion and brief, appellate brief, etc. See generally Alexa Z. Chew & Katie Rose Guest Pryal, The Complete Legal Writer (2016). These are textual artifacts that have common forms because they address common purposes. See id. at 4 (defining “genre” as “a recurring document type that has certain predictable conventions” and “conventions” as “parts of a genre and the ways that audiences expect a genre to be written”).

  72. Swales, supra note 71, at 137–76.

  73. John M. Swales, Research Genres: Explorations and Applications 228 (2004).

  74. Swales, supra note 73, at 229. For example, in the introduction to a research article, an author typically establishes the territory of the article (Move 1), establishes its niche (Move 2), and then occupies the niche (Move 3). Swales, supra note 71, at 140–43. Moves are then further subdivided into steps. See, e.g., id. at 142–47. This is a functionalist approach akin to that described for citations above. See supra note 49. The approach from this study will be functionalist in similar ways.

  75. Swales, supra note 71, at 172–73.

  76. Swales, supra note 71, at 173.

  77. See generally Breuch & Larson, supra note 47. They analyzed artifacts at the atomistic or sentence level, eschewing the more flexible move, which could be longer or shorter than a sentence. Id. at 186. They found that students used references to previous research much as Swales had suggested, but that a common purpose for that move was also to introduce a new topic or new material, something Swales had not identified. Id. at 193.

  78. Bhatia did so in a book chapter in which he analyzed the common-law “legal case” of Roles v. Nathan [1963] 2 All ER 908. Vijay K. Bhatia, Analysing Genre: Language Use in Professional Settings 129–35 (1993). Though he examined the text carefully, it was still only one court case, not enough to generalize even about opinions by the same author, let alone legal arguments generally.

  79. 276 F. Supp. 3d 34 (S.D.N.Y. 2017).

  80. 510 U.S. 569 (1994).

  81. 276 F. Supp. 3d at 41–42 (citation omitted).

  82. Id. at 43.

  83. See supra note 77 (discussing a study using sentence-by-sentence or atomistic coding).

  84. Breuch and Larson noted the challenges for reliability of defining text segments for analysis. Breuch & Larson, supra note 47, at 188.

  85. See supra notes 42, 71 and accompanying text.

  86. See 17 U.S.C. § 107 (2018).

  87. Monge v. Maya Magazines, Inc., 688 F.3d 1164, 1170 (9th Cir. 2012) (“This affirmative defense presumes that unauthorized copying has occurred, and is instead aimed at whether the defendant’s use was fair.”).

  88. 17 U.S.C. § 107. Courts balance the factors on a case-by-case basis. Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 577 (1994) (“The task is not to be simplified with bright-line rules, for the statute, like the doctrine it recognizes, calls for case-by-case analysis.”).

  89. 28 U.S.C. § 1338. But see T. B. Harms Co. v. Eliscu, 339 F.2d 823, 824 (2d Cir. 1964) (noting that an action where “the purported sole owner of a copyright alleged that persons claiming partial ownership had recorded their claim in the Copyright Office and had warned his licensees against disregarding their interests was not one ‘arising under any Act of Congress relating to . . . copyrights’”).

  90. One concern with copyright opinions is that they are not proportionally distributed across all trial courts at the federal level. In this sample, 137 of the artifacts, or 68.8%, were from district courts in the Second and Ninth Circuits, which tend to have fuller copyright dockets because of the concentration of entertainment and publishing businesses in New York and California. The best I can do with this concern is to note it and suggest that future studies should look at different slices of lawyers’ practices, or at different communities or sub-communities in the genre-theoretic senses. See supra notes 42, 71. See the findings in Part III comparing artifacts from cases in the Southern District of New York and other courts. Part III.A.5. Second, given fair use’s factor-based nature, the relative frequency of the category called example below might be expected to be higher here than in artifacts relating to other bodies of law. The remedy is to conduct further studies in other areas of the law.

  91. See U.S. Courts, Federal judicial caseload statistics 2018, (last visited Nov. 15, 2019) (noting that in 2018, civil filings in regional courts of appeal totaled 49,363; in the Federal Circuit, 1617; in district courts, 358,563); Supreme Court of the United States, Granted & Noted List: October Term 2018 Cases for Argument,, (identifying fewer than eighty cases that received grants of certiorari in that term).

  92. Provenzano & Larson, supra note 5, at 1019. While each appellate filing usually results in one round of lawyers’ briefs and a single opinion, a filing in district court may result in many rounds of briefs and opinions from the court.

  93. Nevertheless, I do recommend in the conclusion to this article that researchers should apply these methods to appellate legal arguments.

  94. For my purposes, ‘dispositive motion’ here includes post-trial briefs.

  95. I intentionally designed this study to avoid the need for human participants: The research materials and data are all available publicly online and thus require no intervention or interaction with the authors or litigants and no access to private information. I therefore did not need to seek review of this project by my institution’s Institutional Review Board (IRB).

  96. For example, in Northland Family Planning Clinic, Inc. v. Center for Bio-Ethical Reform, 868 F. Supp. 2d 962, 966 (C.D. Cal. 2012) (No. SACV 11–731 JVS) (artifact 12.03.00 in this study), the parties raised fair use in their cross-motions for summary judgment. Defendants’ Memorandum of Points & Authorities in Support of Motion for Summary Judgment at 10 (Apr. 16, 2012), ECF No. 40 (artifact 12.03.40 in this study) (raising fair-use defense); Memorandum of Points and Authorities in Support of Plaintiff’s Motion for Partial Summary Judgment at 9 (Apr. 16, 2013), ECF No. 45 (artifact 12.03.45 in this study) (arguing that defendant could not succeed on fair use)

  97. There is a narrow band of exceptions to this statement: Where the copyright owner brought a claim for infringement and then moved for summary judgment, it occasionally omitted any reference to fair use, probably because it is an affirmative defense. If the copyright-claim defendant then raised fair use in its opposition memorandum, the claim plaintiff would address fair use in a reply brief. In this study, we would code the first brief raising fair use and the first brief from the other side responding to it.

  98. My RAs and I read the entire briefs. See Larson, supra note 12 (describing the process for selecting the coding span).

  99. See Larson, supra note 12. Some of the artifacts in the data repository are subject to U.S. Copyright law, which vests in authors of texts like the briefs here the exclusive right to make copies of them. 17 U.S.C. § 106(1). The opinions of the federal judges in this study are not, however, subject to copyright. See 17 U.S.C. § 105. Providing copies of working documents relating to opinions in the data repository thus creates no copyright liability. As for the lawyers’ briefs, even wholesale copying and distribution of lawyers’ briefs by commercial services Westlaw and Lexis has been held to be fair use under the Copyright Act. White v. West Pub. Corp., 29 F. Supp. 3d 396, 400 (S.D.N.Y. 2014). I contend that providing these copies in association with this study is a fair use.

  100. One to ten lawyers, eleven to fifty lawyers, fifty-one to 200 lawyers, and more than 200 lawyers. I considered the possibility that some readers might assume that practices they consider sub-optimal might be coming from sub-optimal practitioners. I gathered firm-size information to be able to assess whether large firms (with firm size being in some people’s minds directly related to quality of lawyers) engaged in different practices than smaller ones.

  101. Regarding copyright fair use, I use the terms ‘secondary user’ and ‘rights holder’ in this way: A rights holder is a party in copyright litigation that owns a copyright that another party to the litigation may have infringed. A rights holder is often the plaintiff in a copyright action, though that is not necessary, as for example when the rights holder is the defendant in a declaratory judgment action. See, e.g., Richards v. Merriam-Webster, Inc., 55 F. Supp. 3d 205, 210 (D. Mass. 2014) (holding there was no fair use where the plaintiff was a pro se secondary user seeking a declaration that his proposed use of portions of defendant/rights holder’s dictionary would be fair use). I used this term instead of ‘plaintiff,’ ‘defendant,’ ‘moving party,’ or ‘non-moving party’ when describing a case for coding. A secondary user copies (or is alleged to have copied) at least a portion of the work of the rights holder and asserts that any such copying is protected under the fair-use doctrine. Am. Geophysical Union v. Texaco Inc., 60 F.3d 913, 917 (2d Cir. 1994) (“The traditional fair use analysis . . . [adjusts] the competing interests of authors—the author of the original copyrighted work and the author of the secondary work that ‘copies’ a portion of the original work . . . .”) (emphasis mine). See also id. at 920 (referring to “secondary user”). The secondary user is often the defendant in a copyright case, but it may be the plaintiff if it has brought a declaratory judgment action against the rights holder.

  102. It was possible that each had a partial victory.

  103. Practically speaking, judges’ opinions were the only artifacts in the neither category. Briefs could obviously be in support or opposition to a motion, but sometimes a brief functioned in both ways.

  104. See generally William C. Mann & Sandra A. Thompson, Rhetorical Structure Theory: Toward a Functional Theory of Text Organization, 8 Text–Interdisciplinary Journal for the Study of Discourse 243 (1988); Maite Taboada & William C. Mann, Applications of Rhetorical Structure Theory, 8 Discourse Studies 567 (2006); Maite Taboada & William C. Mann, Rhetorical Structure Theory: Looking Back and Moving Ahead, 8 Discourse Studies 423 (2006); Marie-Francine Moens & Rik De Busser, First Steps in Building a Model for the Retrieval of Court Decisions, 57 International Journal of Human-Computer Studies 429 (2002); C. Reed & A. Daskalopolu, Modelling Contractual Arguments, 4th International Conference on Argumentation, pp. 686–92, Amsterdam, The Netherlands (1998).

  105. I have not adopted the full RST framework, as I am interested only in the uses of cases as support. RST is much more thorough and operates a much smaller scale than I need to (or want to, given the effort involved).

  106. See, e.g., Richard K. Neumann, Jr. & Sheila Simon, Legal Writing 125–30 (2008); Deborah A. Schmedemann & Christina L. Kunz, Synthesis: Legal Reading, Reasoning, and Writing 106–08 (3d ed. 2007); Laurel Currie Oates & Anne Enquist, The Legal Writing Handbook 134–37 (4th ed. 2006); see generally Richard K. Neumann, Jr., Combining Proofs of Separate Conclusions of Law, in Legal Reasoning & Legal Writing 131 (6th ed. 2009).

  107. See Larson, supra note 12 (more completely describing the process for text segmentation). This was the text segment that contained the top-level nucleus or claim—usually that a party’s motion for dismissal, summary judgment, new trial, etc.—on the issue of whether there was fair use. In short, they identified the segment(s) of the brief or court opinion that addressed fair use.

  108. Every citation to any court case in the coding span counted, including short forms and “id.

  109. Note that raw citations did not include what I call “secondary citations”; these are citations within another citation that explain that the cited source was citing or quoting another source or that provide subsequent or prior history for the cited source. See Larson, supra note 12. This choice does not reflect any particular theoretical stance. Rather, for the sake of reliability in coding, it is a choice that must be made, and having been made, it should be expressed.

  110. At each level, coders highlighted the statement of the nucleus/claim, and in those cases where the nucleus was not clearly stated, the coders inserted bracketed language that did that work and highlighted the bracketed language.

  111. See, e.g., Christine Coughlin, Joan Malmud Rocklin & Sandy Patrick, A Lawyer Writes: A Practical Guide to Legal Analysis 93–95 (3d ed. 2018).

  112. Consider the reconstruction of the argument in the text in Part III of the Coding Guide. Larson, supra note 12.

  113. For an extended example, refer to Part III of Larson, supra note 12.

  114. See supra Part I.A.

  115. See, e.g., Breuch & Larson, supra note 47, at 191.

  116. See text accompanying note 23, supra.

  117. Reasoning by legal analogy is a key component of legal reasoning generally. For a fuller discussion, see generally Larson, supra note 1, and sources cited there, including particularly those at 664–65 nn.2–3.

  118. See text accompanying note 25, supra.

  119. See infra Part III.B.

  120. All the data are available in the data repository, an online supplement. Larson, supra note 12.

  121. Larson, supra note 12.

  122. I was one of the two coders on each of 169 (84.9%) of the artifacts. Because I recognize that different readers can perceive a case use as functioning in different ways, we took a greedy approach to codes: If one coder saw something the other did not, they resolved the difference in a conference, accepting the code if both coders thought it was at least plausible. A stingy approach, where a unit receives a code only if both coders saw it independently, is also justifiable but represents a conceptually different emphasis. We might interpret greedy coding as identifying plausible uses of opinions and stingy coding as identifying predominant uses of opinions. Because of this strategy, I was not particularly concerned about assessing inter-rater reliability. Nevertheless, I summarize IRR checks here: Mean pairwise, per-case-file observed agreement among coders was 86.9%. The Cohen’s Kappa statistic measures the amount of non-chance agreement two coders achieve. Jean Carletta, Assessing Agreement on Classification Tasks: The Kappa Statistic, 22 Computational Linguistics 249, 252 (1996). In this study, mean pair-wise, per-case-file Cohen’s Kappa was 0.52. Interpretations of the Kappa statistic are “clearly arbitrary” but provide “benchmarks.” J. Richard Landis & Gary G. Koch, The Measurement of Observer Agreement for Categorical Data, 33 Biometrics 159, 165 (1977). Landis and Koch provide the following interpretations: 0.21–0.40 “fair,” 0.41–0.60 “moderate,” 0.61–0.80 “substantial,” and 0.81–1.00 “almost perfect.” Id. The calculation of agreement did not include thirty-eight artifacts that we coded using an approach that did not preserve individual coders’ codes (case files 12.01, 12.02, 13.01, 13.02, 15.01, 16.01, 17.01 and 17.03).

  123. See the examples for this and other codes in Part III of Larson, supra note 12.

  124. Consider 13.01.00 CU09, where the author implied but did not directly state that the secondary user’s use of more of the original work would make the use less likely transformative.

  125. See Douglas Walton, Chris Reed & Fabrizio Macagno, Argumentation Schemes 332 (2008) (describing the “Argument from Positive Consequences”).

  126. Part III.B. provides further elaboration of what I found.

  127. See supra notes 86–91 and accompanying text.

  128. See supra Part I.A.

  129. See infra Part III.B. For definitions of ‘rational’ and ‘tactical’ appeals, see text accompanying notes 23 and 25, supra.

  130. See supra notes 42, 71 and accompanying text.

  131. A small proportion of the opinions (9 or 16% of the total) provided outcomes that were mixed on fair use, so I did not categorize the associated briefs based on prevailing party.

  132. A small proportion of the briefs (10 or 7% of the total) qualified as both supporting and opposing motions, as in the Plaintiff’s Opposition to Defendants’ Motion for Summary Judgment and Cross Motion for Partial Summary Judgment (Nov. 5. 2012), ECF. No. 41 (artifact 12.02.41 in this study), in National Football Scouting, Inc. v. Rang, 912 F. Supp. 2d 985 (W.D. Wash. 2012), where the party opposing a motion also makes a motion in its memo of opposition. There was also a small proportion of briefs (4 or 3% of the total) where neither party was moving party, as in BWP Media USA, Inc. v. Gossip Cop Media, Inc., 196 F. Supp. 3d 395 (S.D.N.Y. 2016) (artifact 16.02.00 in this study), and Barcroft Media, Ltd. v. Coed Media Group, LLC, 297 F. Supp. 3d 339, (S.D.N.Y. 2017) (artifact 17.04 in this study), where the artifacts were the parties’ post-trial briefs or letters to the judge.

  133. The Appendix addresses some technical issues regarding statistical and practical significance. For a further discussion of practical significance, see text accompanying infra, notes 193 and 195.

  134. As Table 2 shows, there was great variation in the length of artifacts and in the numbers of citations and case uses in each case file. Table 3 shows the same statistical characteristics (except for number of artifacts) across the artifacts themselves. It also provides a ratio of case uses to total raw citations in the artifact, where the denominator is the total number of times the case was cited (including ‘id.’ used to cite a case fully cited earlier).

  135. This was the case with artifact 13.02.00.

  136. See Table 4 in the Appendix.

  137. This is a form of data display that may require some explanation. In a box plot, the box represents the interquartile range of the values, with the vertical line across it indicating the median value. The horizontal lines extending from the boxes—sometimes called ‘whiskers’— represent 1.5 times the inter-quartile range (IQR). If there are any outlier values beyond the whiskers, they are represented with small circles. The box-and-whiskers plots have the virtue of representing both the dispersion and skewness of the data. So in Figure 1, for example, we can see that the values for rule case uses are clustered in the IQR between 3.36 and 6.90 uses per 1000 words and those for example case uses between 1.45 and 3.73, while the outliers span a larger range, up to 15.33 for rule and 8.92 for example. The values for these two statistics are thus fairly tightly dispersed around the medians, but notice that both values exhibits some skewness: the rule values are slightly left-skewed, because the portion of the box to the left of the median is slightly larger than that to the right. The example values are slightly right-skewed.

    The box-and-whiskers plot exhibits one characteristic that may seem counterintuitive to readers familiar with bar charts and histograms. In bar charts and histograms, the size of a bar indicates how many observations fall within the category indicated by the bar: the more observations, the bigger the bar. With the box plot, the size of the box indicates the dispersion of the values. So in Figure 1, if we look at example case uses, we can see that the part of the box to the right of the median is larger than the part of the box to the left. This does not mean that there are fewer observations in the area represented by the left part. In fact, there are exactly as many observations represented by the right part of the box as represented by the left—one quarter of the observations in the sample. The comparative thinness of the left portion of the box represents a concentration of one quarter of the overall values in a smaller range; the comparative thickness of the right part of the box represents a lower concentration, one quarter of the overall values in a larger range.

  138. They were, however, statistically significant for generalization and policy.

  139. Recall that I use ‘typical’ to refer to values falling within the interquartile range; I refer to values above or below that as ‘unusually’ large or small.

  140. The differences were statistically significant for all types of case use except generalization.

  141. Comparing artifacts from the Second Circuit (n = 99) with those from other circuit-court territories (n = 100) produced similar results. Comparing artifacts from the Ninth Circuit (n = 38) with those from other circuit-court territories (n = 161) did not produce statistically or practically significant differences.

  142. Differences between those authors on policy, example, and quotation are all statistically significant.

  143. Note first that I concluded that around 4% of the case uses coded other probably should not have been. Often, this was because one of the two original coders did not see a rational appeal or quotation, marking the unit as other, but later agreed in conference with the other coder that one of the rational appeals was present. In the future, using these methods, it might be wise to ask the original coders to remove the other code in this instance. It’s also true that coders may have missed uses of cases other than the four rational appeals that I identified above because their attention was focused specifically on those four. Future research will have to address that possibility.

  144. Footnote omitted.

  145. A further example appears in artifact 13.08.1032. At the very end of the secondary user’s fair-use argument, the point where it might have chosen to balance the four factors, it concluded: “Taken together, the public-minded and scholarly purpose of the [secondary uses] . . . compels a finding that the public purpose of copyright law ‘would be better served by allowing the use than by preventing it.’ Blanch, 467 F.3d at 251.” The quotation is from Blanch v. Koons, a Second Circuit case binding on the district court in this artifact. But we have no statement of any rule about balancing the factors from Blanch nor any facts from Blanch to show how that case balanced the factors. At most, this could perhaps be read as a policy statement (and indeed, that is how we coded it) from a binding court stitched together with an assertion about the instant case.

  146. Citations to the record in the case omitted.

  147. Footnotes omitted.

  148. Of course, that was not a fair-use case, and at this point in the fair-use analysis, the advocate should be asserting how creative the original work is, not whether it’s copyright-protected at all.

  149. Case files 12.01, 12.05, 15.07, and 15.09.

  150. E.g. 12.07.79 CU02 (asserting relevance of transformativeness in context of a court decision before Campbell); 18.01.00 CU07 (citing a case that stresses the importance of analysis of the first fair-use factor).

  151. 510 U.S. 569, 584 (1994).

  152. E.g., 12.05.00 CU03, 12.05.64 CU15, 15.06.166 CU11.

  153. For the distinction between these types of appeals, see text accompanying notes 23–25, supra.

  154. See supra Part I.A.

  155. See supra Part III.B.

  156. See supra note 27 and sources cited there.

  157. See supra Part I.A.

  158. Brian N. Larson, Bridging Rhetoric and Pragmatics with Relevance Theory, in Relevance and Irrelevance: Theories, Factors and Challenges 69, 79 (Jan Straßheim & Hisashi Nasu eds., 2018) (defining ‘explicatures’ as “inferences about disambiguation of word senses and reference resolution” or “identification of explicit content”) (citations and quotations omitted); id. at 73 (defining ‘implicatures’ as “conclusions that the Hearer draws about what the Speaker implied, suggested, or meant that are different from what the Speaker said, which is closely related to the conventional meaning of the words uttered”) (citations and quotations omitted).

  159. Though there may still be very many.

  160. Weinreb, supra note 1, at 8–9. See supra Part I.A.

  161. Schauer, supra note 1, at 582.

  162. For use of the term ‘covering rule,’ see Weinreb, supra note 1, at 61 n.31.

  163. I noted previously:

    Schauer offers an example to help define what he means by categories of assimilation. He asks us to “[i]magine a faculty meeting considering a request from a student for an excused absence from an examination in order to attend the funeral of his sister.” Assuming the faculty grants the request without comment, for what is it a precedent? Can students expect to be excused “to attend the funerals of grandparents, aunts, uncles, cousins, nieces, nephews, close friends, and pets”? The answer, he tells us, depends on how broadly the category of assimilation is drawn. The decedent in the first case could be characterized as “a sibling, a relative, a blood relative, and one with whom the student has a ‘meaningful relationship.’”

    Larson, supra note 1, at 680–81 (quoting Schauer, supra note 1, at 577–79) (notes omitted).

  164. See supra Part I.B.

  165. Initial pairwise observed agreement (86.62%) and Cohen’s Kappa (0.20) were lower on this category than any other. For discussion of inter-rater reliability on this study, see supra, note 122.

  166. See supra Part I.B.

  167. I’m grateful to an anonymous reviewer for raising this issue.

  168. He noted: “[t]he sorts of evidence relevant in [legal] cases of different kinds will naturally be very variable. To establish negligence in a civil case, willful intent in a case of murder, the presumption of legitimate birth: each of these will require appeal to evidence of different kinds.” Toulmin, supra note 28, at 16.

  169. Swales, supra note 71, at 24–27.

  170. Id.

  171. Personal correspondence with Professor Kevin Bennardo. Copy on file with the author.

  172. See Larson, supra note 12.

  173. I’m preparing the findings of that study now.

  174. See Bennardo & Chew, supra, note 52.

  175. I’m preparing the findings of that study now, as well.

  176. See Kirsten K. Davis, The Reports of My Death Are Greatly Exaggerated: Reading and Writing Objective Legal Memoranda in a Mobile Computing Age, 92 Or. L. Rev. 471, 482 (2013) (critiquing critiques of the “traditional legal memorandum”).

  177. Of course, further studies might also uncover flaws in this study or its methods.

  178. See generally Larson, supra note 12.

  179. It may be surprising that there was a single instance where we coded an opinion but none of its briefs. Case file 14.06 amazingly had only one artifact, the court’s opinion in Calibrated Success, Inc. v. Charters, 72 F. Supp. 3d 763 (E.D. Mich. 2014). All three relevant briefs (ECF Nos. 11, 15, and 16) mention fair use, but none has a single citation to a court opinion in support of that discussion!

  180. The outliers here deserve a brief mention: For case file 14.06, which had only one artifact, see supra note 185. Case file 14.03 had only two artifacts, the court’s opinion in Caner v. Autry, 16 F. Supp. 3d 689 (W.D. Va. 2014), and the defendant’s motion, Brief in Support of Motion to Dismiss or in the Alternative, for Summary Judgment (Nov 26, 2013). The plaintiff did not oppose the defendant’s motion with a brief. On the other end of the spectrum is White v. West Publishing Corp., 29 F. Supp. 3d 396 (S.D.N.Y. 2014), an opinion precipitated by three motions for summary judgment (one by the plaintiff and one by each of two defendants), with a brief on each side of each motion.

  181. Failure to add to 100% is due to rounding. We did not record data for whether an opinion favored the moving or non-moving party because the opinions frequently adjudicated cross motions for summary judgment or dismissal, making both parties effectively moving and non-moving.

  182. Part III.A.5.

  183. I say ‘roughly’ because I performed all statistical tests using R Studio, and the quartile functions in R Studio count the median value (if it exists in the observations and is not calculated as the mean of the two values in the middle of the sample) as part of the second and third quartiles. It gets counted twice, and consequently, ‘roughly’ is appropriate here.

  184. See supra note 138.

  185. I performed all statistical tests using R Studio.

  186. András Vargha & Harold D. Delaney, A Critique and Improvement of the “CL” Common Language Effect Size Statistics of McGraw and Wong, 25 J. Educ. & Behav. Stat. 101, 102 (2000).

  187. This statistic bears a linear relationship to Cliff’s delta, with which some readers may be more familiar. Id. at 104.

  188. The researcher-selected threshold is always arbitrary, but there is not a conventionally accepted value here as there is with alpha. Vargha and Delany propose that the effect size is “small” where A ≥ 0.56, “medium” where A ≥ 0.64, and “large” where A ≥ 0.71. Id. at 106. In my view, any medium or large effect size would be practically significant, but I also believe it’s wise to relax that standard slightly for purposes of this study. The best practice going forward would be to study how big the effect size needs to be before a reader is likely to perceive any difference in an argument based solely on the size of this effect.

  189. Note that in this case, one could look just as easily at the median for opinions (6.23), which is just below the upper end of the IQR for briefs (6.37). If the distribution of either statistic had more skew, this reciprocal relationship might not hold.

  190. This is a heuristic method, of course, and there is no statistical reality that compels this correspondence in any given instance. Vargha and Delaney’s A is not isomorphic with the IQR for the values.