Volume and Issue: Vol. 97, No. 5

The Major Questions Doctrine (“MQD”) is the newest textualist interpretive canon, and it has driven consequential Supreme Court decisions concerning issues from vaccine mandates to environmental regulation. Yet, the new MQD is a canon in search of legitimization. Critics allege that the MQD displaces the Court’s conventional textual analysis with judicial policymaking. Textualists have now responded that the MQD is a linguistic canon, consistent with textualism. Justice Barrett recently argued in Biden v. Nebraska that the MQD is grounded in ordinary people’s understanding of language and law, and scholarship contends that the MQD reflects ordinary people’s understanding of textual clarity in “high-stakes” situations. Both linguistic arguments rely centrally on “common-sense” examples from everyday situations.

This Article tests whether these examples really are common sense to ordinary Americans. We present empirical studies of the examples offered by advocates of the MQD, and the results challenge the arguments that the MQD is a linguistic canon. Moreover, the interpretive arguments offered to legitimize the MQD as a linguistic canon threaten both textualism and the Supreme Court’s growing anti-administrative project.

INTRODUCTION

The Supreme Court’s most consequential interpretive canon is a new one: the major questions doctrine (“MQD”). The basic idea is as follows: when an agency undertakes a “major” policy action, the statutory authorization must be clear and specific (rather than unclear or general).¹See infra Section I.A. In several high-profile cases, the Court has used the MQD to strike down agency actions involving vaccine mandates,²Nat’l Fed’n of Indep. Bus. v. Dep’t of Lab., Occupational Safety & Health Admin., 142 S. Ct. 661, 665–66 (2022). environmental regulation,³West Virginia v. EPA, 142 S. Ct. 2587, 2609–16 (2022). and student loan relief.⁴Biden v. Nebraska, 143 S. Ct. 2355, 2375 n.9 (2023). The majority opinion states that the issue is resolved by “statutory text alone,” and its appeal to the Major Questions Doctrine (“MQD”) “simply reflects [the] Court’s familiar practice of providing multiple grounds to support its conclusions.” Id. Given this track record, no wonder critics have argued that the MQD poses an existential threat to the administrative state, since few statutes are likely to provide the requisite clear language, and what constitutes “majorness” is subjective and potentially applicable to a wide range of agency actions.⁵See, e.g., Lisa Heinzerling, The Power Canons, 58 Wm. & Mary L. Rev. 1933, 1938 (2017). But see Kristin E. Hickman, The Roberts Court’s Structural Incrementalism, 136 Harv. L. Rev. F. 75, 76–77 (2022) (arguing that the development of the MQD is more incrementalist than critics have suggested and that it will likely not threaten the administrative state).

Despite its undeniable influence, the MQD is undertheorized, and it remains a canon in search of a justification.⁶See Mila Sohoni, The Major Questions Quartet, 136 Harv. L. Rev. 262, 285–87 (2022) (recounting but disagreeing with these efforts). Scholars and judges have splintered in their understanding of how the doctrine operates on statutory language.⁷See, e.g., Louis J. Capozzi III, The Past and Future of the Major Questions Doctrine, 84 Ohio St. L.J. 191, 219, 222–23 (2023). Compare West Virginia v. EPA, 142 S. Ct. at 2587, with West Virginia v. EPA, 142 S. Ct. at 2616 (Gorsuch, J., concurring), and Biden v. Nebraska, 143 S. Ct. at 2376 (Barrett, J., concurring). For instance, one advocate of the canon describes it as a requirement for a “clear and specific statement from Congress if Congress intends to delegate questions of major political or economic significance to agencies.”⁸Ilan Wurman, Importance and Interpretive Questions, 110 Va. L. Rev. 909, 909 (2024). As we discuss in Section I.B, Wurman’s characterization of the MQD as a clear statement rule notwithstanding, he views the MQD as justifiable as a linguistic canon. Two critics of the MQD have described it similarly as a rule requiring courts “not to discern the plain meaning of a statute using the normal tools of statutory interpretation, but to require explicit and specific congressional authorization for certain [major] agency policies.”⁹Daniel T. Deacon & Leah M. Litman, The New Major Questions Doctrine, 109 Va. L. Rev. 1009, 1009 (2023). In response, Justice Barrett in Biden v. Nebraska has denied that the MQD requires courts “to depart from the best interpretation of the text,” and claims that the canon is not a clear statement rule and does not require explicit congressional authorization of the “precise agency action under review.”¹⁰Biden v. Nebraska, 143 S. Ct. at 2378 (Barrett, J., concurring). These kinds of disagreements, while perhaps technical, influence how the doctrine is defended and employed, and even implicate its future as an interpretive canon.

So far, efforts to legitimize the doctrine have been unpersuasive. The canon is used primarily by self-identified textualists,¹¹See, e.g., West Virginia v. EPA, 142 S. Ct. at 2616 (Gorsuch, J., concurring); Biden v. Nebraska, 143 S. Ct. at 2372–75; id. at 2376 (Barrett, J., concurring). but critics (textualist and non-textualist alike) have alleged that the MQD is inconsistent with textualism, or even is anti-textualist, because it displaces the ordinary meaning of statutory text in the name of normative values.¹²See, e.g., Sohoni, supra note 6, at 282–90; Daniel E. Walters, The Major Questions Doctrine at the Boundaries of Interpretive Law, 109 Iowa L. Rev. 465, 523–37 (2024); Chad Squitieri, Who Determines Majorness?, 44 Harv. J.L. & Pub. Pol’y 463, 480 (2021); Benjamin Eidelson & Matthew C. Stephenson, The Incompatibility of Substantive Canons and Textualism, 137 Harv. L. Rev. 515, 522–33 (2023); Mike Rappaport, Against the Major Questions Doctrine, The Originalism Blog (Aug. 15, 2022, 8:00 AM), https://originalismblog.typepad.com/the-originalism-blog/2022/08/against-the-major-questions-doctrinemike-rappaport.html [https://web.archive.org/web/20240728034527/https://originalismblog.typepad.com/the-originalism-blog/2022/08/against-the-major-questions-doctrinemike-rappaport.html]; Chad Squitieri, Major Problems with Major Questions, L. & Liberty (Sept. 6, 2022), https://lawliberty.org/major-problems-with-major-questions [https://perma.cc/2D3Y-AA4K]. In fact, the MQD’s rise coincides with a surge of skepticism among textualists and commentators about the validity of substantive canons generally.¹³See, e.g., Eidelson & Stephenson, supra note 12, at 517–21. Of course, textualist skepticism about substantive canons is not new. See, e.g., Amy Coney Barrett, Substantive Canons and Faithful Agency, 90 B.U. L. Rev. 109, 110 (2010). The Court’s use of the MQD even prompted Justice Kagan to retract her quip that “we’re all textualists now.”¹⁴Harvard Law School, The 2015 Scalia Lecture Series: A Dialogue with Justice Elena Kagan on the Reading of Statutes, YouTube (Nov. 25, 2015), https://youtu.be/dpEtszFT0Tg [https://perma.cc/L65V-9AET]. She now notes: “It seems I was wrong. The current Court is textualist only when being so suits it.”¹⁵West Virginia v. EPA, 142 S. Ct. at 2587, 2641 (Kagan, J., dissenting). See generally Kevin Tobia, We’re Not All Textualists Now, 78 N.Y.U. Ann. Surv. Am. L. 243 (2023) (providing an overview of the influence and evolution of “all textualist” statements). These critiques allege that the MQD inappropriately licenses textualists to depart from the best reading of statutory text in the name of values or norms. An ideal response for a textualist favoring the MQD would be some account of how the MQD determines the linguistic meaning of a statute.

Increasingly, textualists are making precisely this “linguistic” move. Some textualists now propose that the MQD is a linguistic interpretive canon, consistent with textualism.¹⁶See, e.g., Wurman, supra note 8, at 916–17; Biden v. Nebraska, 143 S. Ct. at 2376 (Barrett, J., concurring). On this account, textualists remain committed to the ordinary reader’s understanding of language, with the MQD simply reflecting how ordinary people, exercising basic “common sense,” generally understand the meaning of statutes delegating authority to agencies.¹⁷See, e.g., Amy Coney Barrett, Congressional Insiders and Outsiders, 84 U. Chi. L. Rev. 2193, 2208–11 (2017) (arguing for statutory interpretation to focus on the understanding of ordinary people rather than Congress). On this “linguistic” picture, normative or substantive values are not relevant to the canon or its application, and they certainly do not lead textualists to depart from the best reading of the text. Instead, the MQD is just like any other linguistic canon—it reflects only a generalization about how ordinary people use and understand language in context.¹⁸On the modern textualist Court’s emphasis on ordinary readers and the relationship between ordinary understanding and linguistic canons, see Kevin Tobia, Brian G. Slocum & Victoria Nourse, Statutory Interpretation from the Outside, 122 Colum. L. Rev. 213, 213 (2022) [hereinafter Tobia et al., From the Outside]. This rebranding of the MQD as a linguistic canon has rapidly moved from the pages of law reviews¹⁹See Wurman, supra note 8, at 909. to the Supreme Court.²⁰Biden v. Nebraska, 143 S. Ct. at 2376 (Barrett, J., concurring). There, Justice Barrett recently denied that the MQD is normatively driven and instead argued that it merely reflects ordinary people’s “common-sense” understanding of instructions, including those given by Congress.²¹Id. at 2384; see also Beau J. Baumann, Let’s Talk About That Barrett Concurrence (on the “Contextual Major Questions Doctrine”), Yale J. on Reg.: Notice & Comment Blog (June 30, 2023), https://www.yalejreg.com/nc/lets-talk-about-that-barrett-concurrence-on-the-contextual-major-questions-doctrine-by-beau-j-baumann [https://perma.cc/8PKB-458K] (discussing Barrett’s arguments).

In this Article, we evaluate the MQD’s “linguistic turn” and subject its premises to empirical study. We study two key issues: (1) Does the MQD follow from ordinary people’s understanding of language and, more specifically, delegating instructions?; and (2) Do ordinary people interpret more cautiously or narrowly in “high-stakes” situations? The empirical results support answering “no” to both questions. Contrary to the MQD proponents’ contentions, the results indicate that ordinary people do not adjust their judgments of textual clarity according to the stakes of interpretation, and they interpret broad delegations broadly, even in situations in which Justice Barrett claims that “common sense” would dictate narrower interpretations of the scope of authorization.²²See infra Part III.

Part I introduces the MQD and the two linguistic arguments that have been offered in defense of the canon. After briefly addressing the defense of the MQD as a substantive canon in Section I.A, we turn in Section I.B to the proposal that ordinary interpretation shifts in “high-stakes” contexts, and that this behavior justifies the MQD as a linguistic canon.²³See Wurman, supra note 8, at 917. The high-stakes argument appeals to an example from analytic philosophy²⁴See, e.g., Keith DeRose, Contextualism and Knowledge Attributions, 54 Phil. & Phenomenological Rsch. 913, 913–18 (1992). and prior legal scholarship²⁵See Ryan D. Doerfler, High-Stakes Interpretation, 116 Mich. L. Rev. 523, 523 (2018). that suggests that high-stakes contexts diminish ordinary knowledge. Thus, as a famous hypothetical illustrates, you might know that the town bank is open on the weekend when planning to deposit a small check with low stakes. In contrast, in a higher-stakes context (for example, if the check is for ten thousand dollars and must be deposited before Monday to avoid an overdraft), you may decide instead that you do not really know that the bank is open. Legal scholarship proposes that this is how ordinary people understand knowledge: ordinary knowledge is stakes sensitive.²⁶See, e.g., Wurman, supra note 8, at 957–59. More importantly for the MQD, an emerging argument builds on this premise to suggest that ordinary understanding of textual clarity is also stakes driven: in high-stakes contexts, a text is less clear.²⁷See id. As such, in those high-stakes (or “major”) cases, courts should require highly specific language to authorize agency action.

Section I.C introduces Justice Barrett’s separate proposal that ordinary language is context sensitive and anti-literal, and therefore a textualist faithful to the ordinary reader should adopt the MQD as a means to determine the best reading of statutory language.²⁸Biden v. Nebraska, 143 S. Ct. 2355, 2376 (2023) (Barrett, J., concurring); see also Barrett, supra note 17, at 2200 (on textualists’ commitment to the ordinary reader, not the ordinary legislator). Justice Barrett’s argument also appeals to an intuitive example: instructing a babysitter to “have fun with the kids” while handing him a credit card might literally permit the babysitter to take them on an overnight trip to an out-of-town amusement park (after all, doing so would be “fun”). But in context, ordinary people employ “common sense” and understand the literal meaning of the instruction to only permit the most reasonable set of applications of the instruction.²⁹Biden v. Nebraska, 143 S. Ct. at 2376 (Barrett, J., concurring). Ordinary people are therefore non-literalists, understanding general delegations to be more limited in meaning than their terms alone might suggest. As such, the argument goes, the MQD is “consistent with how we communicate conversationally,” making it a valid linguistic canon that reflects an interpretive commitment to ordinary people.³⁰Id. at 2379.

Justice Barrett’s argument is important and places her as a leader among the Court’s textualists; she is the only textualist advocate of the MQD who has offered a proposal to square the MQD with textualism. At the same time, the linguistic argument in her brief concurring opinion is not entirely clear. As such, we attempt to charitably reconstruct Justice Barrett’s defense as a workable argument—that is, one that derives the MQD conclusion from the babysitter hypothetical premise.

Part I contributes to the literature by explaining these two new arguments for the linguistic MQD in sufficient detail. Unpacking the arguments clarifies each argument’s theoretical challenges and empirical claims. Both arguments employ hypotheticals about how ordinary people interpret language but, significantly, support these hypotheticals with references to academic philosophy or judicial intuition; neither uses empirical evidence.

Parts II and III investigate these empirical claims, both by engaging with the existing empirical literature on high-stakes knowledge (much of it uncited by proponents of the linguistic MQD) and by conducting original survey experiments of both high-stakes interpretation and how ordinary people interpret instructions. Part II considers the claim that ordinary knowledge is stakes sensitive. This claim has been influential in philosophy,³¹See, e.g., Keith DeRose, Contextualism, Contrastivism, and X-Phi Surveys, 156 Phil. Stud. 81, 81 (2011). legal scholarship,³²Doerfler, supra note 25, at 523. and now the major questions debate.³³Wurman, supra note 8, at 917. Although philosophers claim knowledge is stakes sensitive, many existing studies report that stakes have little or even no effect on ordinary attributions of knowledge.³⁴See generally Jonathan Schaffer & Joshua Knobe, Contrastive Knowledge Surveyed, 46 Noûs 675 (2012) (surveying studies). Other studies report only a small effect. See, e.g., David Rose, Edouard Machery, Stephen Stich, Mario Alai, Adriano Angelucci, Renatas Berniūnas, Emaa E. Buchtel, Amita Chatterjee, Hyundeuk Cheon, In-Rae Cho, Daniel Cohnitz, Florian Cova, Vilius Dranseika, Ángeles Eraña Lagos, Laleg Ghadakpour, Maurice Grinberg, Ivar Hannikainen, Takaaki Hashimoto, Amir Horowitz, Evgeniya Hristova, Yasmina Jraissati, Veselina Kadreva, Kaori Karasawa, Hackjin Kim, Yeonjeong Kim, Minwoo Lee, Carlos Mauro, Masaharu Mizumoto, Sebastiano Moruzzi, Christopher Y. Olivola, Jorge Ornelas, Barbara Osimani, Carlos Romero, Alejandro Rosas Lopez, Massimo Sangoi, Andrea Sereni, Sarah Songhorian, Paulo Sousa, Vera Tripodi, Naoki Usui, Alejandro Vásquez del Mercado, Giorgio Volpe, Hrag Abraham Vosgerichian, Xueyi Zhang & Jing Zhu, Nothing at Stake in Knowledge, 53 Noûs 224, 232–37 (2019) (reporting no effect of stakes on knowledge in fifteen countries, a small effect in three, and a marginal and small effect in the U.S.). For example, in the U.S., over 80% of participants agreed in both the high- and low-stakes cases that there was knowledge; in Japan, a country with the largest difference between high and low stakes, over 70% of participants attributed knowledge in both. Id. And, to our knowledge, no empirical study bears on the question of whether higher stakes reduce textual clarity (a related but different issue). The critical link in one version of the linguistic MQD argument is therefore entirely untested.

Part III presents studies designed to test the empirical claims of the linguistic MQD arguments. Our studies use the exact two cases offered by proponents of the linguistic MQD—the “bank case” and the “babysitter hypothetical”—to conduct original survey experiments. Overwhelmingly, ordinary people in our studies did not interpret these scenarios consistently with the empirical premises of the linguistic MQD arguments.

Part IV develops three sets of implications that follow from our empirical evidence and the textualist efforts to legitimize the MQD as a linguistic canon. These implications concern the empirical evidence for the linguistic MQD (IV.A), challenges that the linguistic MQD poses for textualism (IV.B), and the relationship between empirical evidence of how ordinary people view delegations and administrative law, including intriguing evidence that people are more concerned about underenforcement of instructions compared with overenforcement (IV.C).

In brief, the extant and new empirical findings do not support the linguistic MQD. Specifically, the findings count against the predictions of the two leading linguistic MQD arguments, using the exact cases offered in defense of the linguistic MQD. Of course, we are open to the possibility that study of further examples could weigh against our conclusions. But for interpreters deciding today whether to employ a “linguistic MQD,” there is insufficient empirical support and theoretical clarity to cast the MQD as a valid linguistic canon. Moreover, the results provide stronger support for a new counter-MQD: ordinary people understand general authorizing language as consistent with a broad range of reasonable actions that fall under the text’s meaning. Textualists committed to the “ordinary reader” and “interpretation from the outside” claim to follow those commitments to where they lead—and the current evidence favors an interpretive rule far from the current MQD.³⁵Barrett, supra note 17, at 2208–11 (arguing that courts should interpret from the “outside,” from the perspective of ordinary people, rather than from the “inside,” which would reflect Congress’s perspective).

I.  THE MAJOR QUESTIONS DOCTRINE AND THEORIES OF ITS LEGITIMACY

The MQD has sparked a great deal of scholarly effort to specify exactly what the doctrine is and how it fits into traditional categories of interpretive doctrine. In this Part, we survey these efforts, many of which conclude that the MQD is a substantive, or normative, canon.³⁶See infra Section I.A. These classifications matter because substantive canons are increasingly questioned as being inconsistent with textualism.³⁷See Eidelson & Stephenson, supra note 12, at 517–21; Barrett, supra note 13, at 110. But see Brian G. Slocum & Kevin Tobia, The Linguistic and Substantive Canons, 137 Harv. L. Rev. F. 70, 70–73 (2023) (arguing that an interpretive canon can have both a linguistic and substantive basis). Classifying the MQD as substantive (rather than linguistic) is tantamount to saying it is illegitimate or tenuous, at least on textualist grounds.³⁸But see Walters, supra note 12, at 469–73 (assuming that substantive canons are often acceptable but arguing that the MQD has features that differentiate it from the rest of the canons in troubling ways). Perhaps not surprisingly, some textualist defenders of the MQD have not fully endorsed the idea that the MQD is a substantive canon.³⁹Wurman, supra note 8, at 912. The exception here is Justice Gorsuch, who offered a full-throated endorsement of the MQD as a nondelegation canon in his concurrence in West Virginia v. EPA. See West Virginia v. EPA, 142 S. Ct. 2587, 2617 (2022) (Gorsuch, J., concurring). In fact, as we discuss below, perhaps the most serious attempt to ground the MQD in interpretive law asserts that the doctrine is instead a linguistic, or semantic, canon.⁴⁰See infra Sections I.B & I.C. In theory, at least, this move would legitimize the canon for textualists and everyone else because the doctrine would simply be folded into the relatively uncontroversial search for the ordinary meaning of delegating statutes.⁴¹Wurman, supra note 8, at 916. For a discussion of “ordinary meaning,” see Brian G. Slocum, Ordinary Meaning: A Theory of the Most Fundamental Principle of Legal Interpretation 1–5 (2015).

This pivot to a linguistic defense raises many questions, very few of which have been answered. After describing how the linguistic defense works, we then highlight theoretical limitations, open questions, and the broader implications of defending the MQD as a linguistic canon.

A.  The Canonization of the Major Questions Doctrine

1.  Historical Threads of the Major Questions Doctrine

The MQD is not entirely new; it is in the process of “metamorphosis.”⁴²Walters, supra note 12, at 480–81. It is also, of course, the talk of the town because of fears/hopes that it will be deployed in such a way as to “kneecap” administrative agencies and promote an economic, libertarian conception of American governance. See Matt Ford, The Supreme Court Conservatives’ Favorite New Weapon for Kneecapping the Administrative State, New Republic (Mar. 13, 2023), https://newrepublic.com/article/171093/supreme-court-major-questions-doctrine-administrative-state [https://perma.cc/R3FJ-GVN8]; John Yoo & Robert Delahunty, The Major-Questions Doctrine and the Administrative State, Nat’l Affairs (Fall 2022), https://www.nationalaffairs.com/publications/detail/the-major-questions-doctrine-and-the-administrative-state [https://perma.cc/7NYU-M8FJ]. Arguably, the first appearance of something like the MQD was in the plurality opinion in a 1980 case known as the Benzene Case.⁴³Indus. Union Dep’t v. Am. Petrol. Inst., 448 U.S. 607, 614–15 (1980) [hereinafter Benzene Case]. In that case, the Occupational Safety and Health Administration (“OSHA”) was charged with promulgating standards that “most adequately assure[], to the extent feasible, on the basis of the best available evidence, that no employee will suffer material impairment of health or functional capacity even if such employee has regular exposure to the hazard dealt with by such standard for the period of his working life.”⁴⁴29 U.S.C. § 655(b)(5). Rather than follow OSHA’s argument that the statute, fairly read, seemed to require it to “impose standards that either guarantee workplaces that are free from any risk of material health impairment, however small, or that come as close as possible to doing so without ruining entire industries,” the plurality opinion held that OSHA had only been delegated authority to regulate “significant” risks.⁴⁵Benzene Case, 448 U.S. at 641, 651.

As Cass Sunstein notes, although the Court invoked the nondelegation doctrine and constitutional avoidance to arrive at this statutory interpretation, it is impossible to square what the Court did with the “(standard) nondelegation doctrine.”⁴⁶Cass R. Sunstein, There Are Two “Major Questions” Doctrines, 73 Admin. L. Rev. 475, 484 (2021) (calling the MQD a “linear descendant” of the Benzene Case). This relatively recent vintage has been contested by Louis Capozzi, who argues that the Supreme Court deployed the MQD in a series of rate cases in the late 19th Century. Capozzi, supra note 7, at 196–97. However, this analogy has itself been contested. See Capozzi on the Future of the Major Questions Doctrine, Admin Wannabe (Oct. 19, 2022), https://adminwannabe.com/?p=114 [https://perma.cc/FK6S-MGZW]. The interpretation offered by OSHA, in addition to doing little violence to the text of the statute, would “sharply cabin” the agency’s discretion.⁴⁷Sunstein, supra note 46, at 486. Sunstein suggests that the plurality opinion in the Benzene Case instead endorsed the novel idea that “without a clear statement from Congress, the Court will not authorize the agency to exercise that degree of (draconian) authority over the private sector.”⁴⁸Id.

It was hardly clear at the time, however, that the Court was creating something called the “major questions doctrine”; in fact, that would not become clear until very recently. Instead, for several decades, the Court intermittently invoked similar, but often distinct, reasoning from the Benzene Case in regulatory cases involving “extraordinary” circumstances, all while leaving the precise theory behind the reasoning unstated. Paradigmatic of these invocations is FDA v. Brown & Williamson Tobacco.⁴⁹FDA v. Brown & Williamson Tobacco Corp., 529 U.S. 120, 120 (2000). In that case, the Food and Drug Administration (“FDA”) promulgated a rule regulating tobacco products as “drugs” under the Food, Drug, and Cosmetics Act. The Court applied the familiar Chevron two-step analysis and concluded, on the basis of an examination of legislative history, that Congress had unambiguously declined to give the FDA this power.⁵⁰Id. at 160–61. The Court added another reason for its conclusion, though, stating that “[i]n extraordinary cases . . . there may be reason to hesitate before concluding that Congress has intended . . . an implicit delegation.”⁵¹Id. at 159.

As the “implicit delegation” phrase reveals, the Court explicitly couched its consideration of the “majorness” or “extraordinariness” of the power asserted by the FDA as part of the Chevron analysis. Thus, the MQD acted as a “carve-out” or “exception” to the ordinary rule that statutory ambiguities constitute implicit delegations that an agency is given primacy over courts to resolve, so long as it does so reasonably.⁵²Sunstein, supra note 46, at 482. Instead, when “extraordinary” questions are presented by the agency’s claim of delegated authority, the Court itself resolves the ambiguity at Chevron step one.⁵³Id.

The Brown & Williamson opinion’s use of proto-MQD logic departed from the apparent logic of the Benzene Case in an important way. The Benzene Case left little room for an agency interpretation to survive once the doctrine was triggered. The only way to prevail was to point to clear statutory authorization that could not be limited by the Court to avoid the major implications of the agency’s interpretation. Sunstein calls this the “strong version” of the MQD.⁵⁴Id. at 486. By contrast, in Brown & Williamson, Sunstein sees a “weak version” that theoretically allowed an agency’s major action so long as the statutory interpretation could be endorsed by a Court engaged in independent (de novo) review without according the agency any deference.⁵⁵Id. at 484.

As a practical matter, the weak version of the MQD seemed to win out for a while after Brown & Williamson, and on at least one occasion, an agency did win in a major questions case. In King v. Burwell, the Internal Revenue Service (“IRS”) interpreted the Affordable Care Act to make tax credits available even if an individual purchased health insurance on a federal insurance exchange, despite statutory language that limited tax credits to plans purchased through “an Exchange established by the State.”⁵⁶King v. Burwell, 576 U.S. 473, 483 (2015) (citing 77 Fed. Reg. 30378 (2012) and 26 U.S.C. §§ 36B(b)–(c)). Like in Brown & Williamson, the Court noted that there “may be reason to hesitate before concluding that Congress has intended such an implicit delegation.”⁵⁷Id. at 485 (quoting FDA v. Brown & Williamson Tobacco Corp., 529 U.S. 120, 159 (2000)). Unlike in Brown & Williamson, however, the Court concluded that the agency had the power to issue the rule, even on a de novo interpretation of the statute. Although the Court’s interpretation of the statutory language at issue has been criticized,⁵⁸Richard M. Re, The New Holy Trinity, 18 Green Bag 407, 408–09 (2015); Ryan D. Doerfler, The Scrivener’s Error, 110 Nw. U. L. Rev. 811, 811 (2015). the important point is that the “weak version” of the MQD—that is, an “exception,” or “carve-out” from Chevron deference—seemed to rule the day. The only open questions were about where, precisely, to locate the major questions exception: at Chevron step zero,⁵⁹See Thomas W. Merrill & Kristin E. Hickman, Chevron’s Domain, 89 Geo. L.J. 833, 873 (2000); Cass R. Sunstein, Chevron Step Zero, 92 Va. L. Rev. 187, 207–11 (2006) (elucidating a “step zero” that asks whether Chevron deference even potentially applies or whether some other standard of review, such as Skidmore or de novo review, should prevail). Most observers viewed King v. Burwell as deploying the major questions exception at step zero. step one,⁶⁰Most observers viewed Brown & Williamson as deploying the major questions exception at step one. or step two.⁶¹The only case to have apparently located the major questions exception at step two was Utility Air Regulatory Group v. EPA, 573 U.S. 302, 331–33 (2014).

2.  The Modern Major Questions Doctrine and Its Justification

Enter what Mila Sohoni calls the “major questions quartet.”⁶²Sohoni, supra note 6, at 262. If it was unclear exactly which version of the MQD existed before the quartet, the waters have become only murkier afterward. One thing is unmistakably clear though: The Court did not treat the MQD as a mere exception or carve-out from Chevron deference. Instead, it “unhitched the major questions exception from Chevron.”⁶³Id. at 263. In fact, the majority opinion in West Virginia v. EPA,⁶⁴West Virginia v. EPA, 142 S. Ct. 2587, 2587–616 (2022). the leading case in the quartet, did not even mention Chevron in its elaboration or application of the MQD.⁶⁵Part of the reason why Chevron was not mentioned may be because the Court is now generally hostile to the doctrine. See Lisa Schultz Bressman & Kevin M. Stack, Chevron Is a Phoenix, 74 Vand. L. Rev. 465, 466 (2021) (describing the debate about Chevron and arguing that judicial deference to agency interpretations is a foundational aspect of administrative law). As this Article went to press, the Court overruled Chevron. See Loper Bright Enters. v. Raimondo, 144 S. Ct. 2244, 2254 (2024). Instead, the Court offered an almost entirely new gloss on the doctrine:

“[I]n certain extraordinary cases, both separation of powers principles and a practical understanding of legislative intent make us ‘reluctant to read into ambiguous statutory text’ the delegation claimed to be lurking there. To convince us otherwise, something more than a merely plausible textual basis for the agency action is necessary. The agency instead must point to ‘clear congressional authorization’ for the power it claims.”⁶⁶West Virginia v. EPA, 142 S. Ct. at 2609 (citation omitted).

For the vast majority of commentators, these words have been taken to suggest that the current Court, post-quartet, thinks of the MQD as a particularly powerful form of substantive canon: a clear statement rule.⁶⁷Deacon & Litman, supra note 9, at 1012; Sohoni, supra note 6, at 264; Walters, supra note 12, at 480–89. On this reading—which seems similar to the implicit use of the doctrine in the Benzene Case—Congress must have spoken with unmistakable clarity in order for agencies to have the “major” power they are claiming to have been delegated. If there is any ambiguity, and even if the agency has a “plausible” basis for concluding that it has the authority under applicable statutes, the agency cannot exercise that power. Some are not convinced the MQD is a clear statement rule and view it as a weaker substantive canon that resolves ambiguity.⁶⁸See, e.g., Natasha Brunstein & Donald L.R. Goodson, Unheralded and Transformative: The Test for Major Questions After West Virginia, 47 Wm. & Mary Env’t L. & Pol’y Rev. 47 (2022) (noting that the Court in West Virginia v. EPA does not refer to the MQD as a clear statement rule). Accordingly, when the MQD is applicable, any statutory ambiguities should be resolved against the agency’s assertion of power so as to vindicate “separation of powers principles.”⁶⁹West Virginia v. EPA, 142 S. Ct. at 2609. In any event, a common understanding is that the MQD is driven by a normative commitment to a limited role for administrative agencies in the legal system, and perhaps by a “delegation doctrine” that insists that agencies have no power unless it is affirmatively shown that Congress has granted it to them.⁷⁰See generally Jonathan H. Adler, The Delegation Doctrine, Harv. J. Pub. Pol’y: Per Curiam, Summer 2024, at 1.

The MQD is inherently controversial as a substantive canon regardless of whether it is a clear statement rule or a tiebreaker canon. Simply by virtue of being a substantive canon, the “new MQD” is in tension with textualism. As Justice Kagan, a self-avowed textualist, puts it, there is some momentum for “toss[ing] [substantive canons] all out.”⁷¹Transcript of Oral Argument at 60, Ysleta del sur Pueblo v. Texas, 142 S. Ct. 1929 (2022) (No. 20-493). As she noted in her West Virginia dissent, channeling Karl Llewelyn, “special canons like the ‘major questions doctrine’” function as “get-out-of-text-free cards.”⁷²West Virginia v. EPA, 142 S. Ct. at 2641 (Kagan, J., dissenting). Karl Llewellyn famously purported to show that every canon can be countered by an equal and opposite canon, which he argued deprives canons of any probative force in the interpretive process. See Karl N. Llewellyn, Remarks on the Theory of Appellate Decision and the Rules or Canons About How Statutes Are to Be Construed, 3 Vand. L. Rev. 395, 401–06 (1950). Llewellyn’s famous critique, however, overstated the conflict among canons. See William N. Eskridge, Jr., Norms, Empiricism, and Canons in Statutory Interpretation, 66 U. Chi. L. Rev. 671, 679 (1999) (“The large majority of Llewellyn’s competing canonical couplets are presumptions about language and extrinsic sources, followed by qualifications to the presumptions.”). Recently, Benjamin Eidelson and Matthew Stephenson have exhaustively assessed “leading efforts to square modern textualist theory with substantive canons” and ultimately concluded that “substantive canons are generally just as incompatible with textualists’ jurisprudential commitments as they first appear.”⁷³Eidelson & Stephenson, supra note 12, at 520–21; see also Barrett, supra note 13, at 110. This challenge would apply to a range of canons employed by the textualist Supreme Court. The Roberts Court, though textualist, often employs substantive canons. See Nina A. Mendelson, Change, Creation, and Unpredictability in Statutory Interpretation: Interpretive Canon Use in the Roberts Court’s First Decade, 117 Mich. L. Rev. 71, 141 tbl.2 (2018); Anita S. Krishnakumar, Reconsidering Substantive Canons, 84 U. Chi. L. Rev. 825, 825–26 (2017). The MQD, insofar as it is a substantive canon, would not be spared.⁷⁴Eidelson & Stephenson, supra note 12, at 520–21.

Beyond these generalized concerns with substantive canons, some commentators have questioned whether the MQD satisfies basic expectations about the Court’s recognition and use of substantive canons, even assuming that they can sometimes be legitimate aids to interpretation. Simply put, the Court has not been at all clear about the source of the normative foundation of the MQD.⁷⁵Biden v. Nebraska, 143 S. Ct. 2355, 2376 (2023) (Barrett, J., concurring) (noting that “there is an ongoing debate” about the MQD’s “source and status”). For Sohoni, formulating the MQD as a kind of constitutional avoidance rule fails because of the “Court’s failure to say anything about nondelegation”—a failure that “creates genuine conceptual uncertainty about what exactly it was doing in these cases.”⁷⁶Sohoni, supra note 6, at 297. The currently prevailing nondelegation test asks merely whether Congress has provided a “reasonably intelligible policy” to guide an agency’s exercise of discretion.⁷⁷A.L.A. Schechter Poultry Corp. v. United States, 295 U.S. 495, 500 (1935). That test would not have provided anywhere close to a “significant risk” of constitutional invalidity in any of the statutes examined in the major questions quartet.⁷⁸Significant risk is required under the modern form of the constitutional avoidance doctrine. Although Justice Gorsuch in his concurrence in West Virginia v. EPA suggested that the MQD is inspired by the nondelegation doctrine (and probably his preferred version of the nondelegation doctrine, which is not the law currently), the majority pointed more generally to “separation of powers principles.”⁷⁹West Virginia v. EPA, 142 S. Ct. 2587, 2609 (2022). Some have inferred that the Supreme Court might be interested in developing constitutional principles demanding affirmative proof of delegation in certain circumstances—and that the MQD reflects this implicit constitutional project⁸⁰Sohoni, supra note 6, at 312–13; Adler, supra note 70, at 6.––but if so the Court has not been explicit. This uncertainty about the connection between constitutional principles and the MQD also seems to doom the MQD under Justice Barrett’s own test for the legitimacy of substantive canons within textualism, under which there must be a reasonably specific constitutional principle to which a constitutionally inspired substantive canon attaches.⁸¹Barrett, supra note 13, at 178. In other words, if the MQD is a substantive canon, its substance, or normative content, is not clear. Most substantive canons either reflect a broad societal consensus or are tied closely to constitutional law. The MQD at first glance has neither of these attributes.

3.  The Modern Major Questions Doctrine’s Linguistic Turn

Perhaps not surprisingly, given the strong pushback that the MQD has received when it is formulated as a substantive canon, defenders of the MQD are increasingly suggesting that the MQD is not a substantive canon at all. Instead, proponents suggest it is a linguistic canon.

This rebranding is not as far-fetched as it might seem at first. “ ‘[L]inguistic’ validity and ‘substantive’ value are properties of canons.”⁸²Slocum & Tobia, supra note 37, at 73. The standard dichotomy between “linguistic” and “substantive” canons suggests that a canon has at most one property; but, it is conceptually possible for a canon to have both.⁸³Id. There is evidence that some canons that have long been treated as “substantive canons”—such as anti-retroactivity and anti-extraterritoriality—are also consistent with how ordinary people understand rules. For example, when a rule (especially a punitive rule) does not explicitly state whether it applies retroactively, prospectively, or both, people tend to understand it to apply only prospectively.⁸⁴Id. at 82–83. Insofar as textualism is guided by ordinary understanding of language,⁸⁵See, e.g., Barrett, supra note 17, at 2194. textualists have good reason to consider such “substantive” canons as simultaneously linguistic ones. Even some tough critics of substantive canons like Eidelson and Stephenson show some openness to these arguments: “[T]he textualist’s reasonable reader . . . opens the door to recasting some seemingly substantive canons as simply default inferences that a reasonable reader would draw . . . . The presumption against extraterritoriality is a possible example.”⁸⁶Eidelson & Stephenson, supra note 12, at 539.

Could a similar linguistic argument support the MQD? Acknowledging that criticisms of the MQD as a substantive canon “are, to some if not a large extent, warranted,”⁸⁷Wurman, supra note 8, at 912. Professor Ilan Wurman recently rebranded the MQD as a linguistic canon.⁸⁸Wurman, supra note 8, at 916 (“On this conceptualization, the importance of a purported grant of authority would operate as a kind of linguistic canon: ordinarily, lawmakers and private parties tend to speak clearly, and interpreters tend to expect clarity, when those lawmakers or parties authorize others to make important decisions on their behalf.”). Wurman argues that the MQD could be understood as motivated by a theory of linguistic usage about how interpretive uncertainty should be resolved rather than as importation of substantive or normative values into the interpretive enterprise. He appeals to prior work in philosophy and legal philosophy, which argues that “high-stakes” contexts lead to less knowledge or legal clarity.⁸⁹See infra Section I.B.

Even more recently, Justice Barrett has proposed her own, separate linguistic argument for the MQD’s legitimacy. The Supreme Court has made the major questions quartet a quintet with its decision in Biden v. Nebraska. That case concerned President Biden’s 2022 proposal to forgive $10,000 to $20,000 in student loans for low to middle-income borrowers. Biden’s Department of Education traced the authority for their emergency loan relief to the HEROES Act, a 2001 law that grants the U.S. Secretary of Education the ability to “waive or modify” provisions related to federal student loans “in connection with a war or other military operation or national emergency.”⁹⁰20 U.S.C. § 1098bb(a)(1). After Biden announced his administration’s loan forgiveness program as a response to the COVID-19 national emergency, several states challenged the program. That case reached the Supreme Court and divided the Justices 6–3 along conservative-liberal lines. Justice Roberts’s majority opinion proceeded with traditional textual interpretation, concluding that the government’s student loan relief is not within the statutory meaning of “waive or modify” any provision. But the opinion also referenced the major questions doctrine, as an alternative ground for the holding.

Justice Barrett wrote separately to argue that the MQD is not a substantive canon but rather “a tool for discerning—not departing from—the text’s most natural interpretation.”⁹¹Biden v. Nebraska, 143 S. Ct. 2355, 2376 (2023) (Barrett, J., concurring). Candidly, and consistently with her prior writings on substantive canons,⁹²See Barrett, supra note 13, at 110. Justice Barrett conceded that the substantive canon version of the MQD might be “inconsistent with textualism” and therefore “should give a textualist pause.”⁹³Biden v. Nebraska, 143 S. Ct. at 2376 (Barrett, J., concurring). By grounding the MQD in how ordinary readers apply common sense in reading statutory text, Justice Barrett aims to put the MQD on more solid footing, particularly for textualists.

After the opinion, some suggested that Justice Barrett’s argument “mirrors” Wurman’s.⁹⁴See, e.g., Baumann, supra note 21. We disagree: the two arguments both present the MQD as a linguistic canon, but the arguments are distinct. Wurman appeals to high-stakes context and the resolution of interpretive uncertainty, while Barrett appeals to anti-literalism and contextual restriction concerning major actions (with nothing about high stakes). Thus, Wurman’s argument centers on “ambiguity” caused by high stakes, whereas Justice Barrett’s theory is about how ordinary people generally use “common sense” to interpret non-literally (with no mention of “ambiguity”). The next two Sections separately reconstruct Wurman’s (I.B) and Justice Barrett’s (I.C) linguistic arguments in detail and present some theoretical challenges for each.

B.  The Major Questions Doctrine as a High-Stakes Linguistic Canon

One important line of work defending the “linguistic” MQD appeals to the philosophical and legal-philosophical literature on stakes and knowledge.⁹⁵Wurman, supra note 8, at 957–61 (appealing to Doerfler, supra note 25). That theoretical literature proposes that knowledge is sensitive to high stakes: it could be true that one knows a proposition in a low-stakes context (for example, the bank is open) but does not know that proposition, given the same evidence, in a high-stakes context.

The legal literature about stakes and interpretation, including the linguistic MQD defense, takes this claim about knowledge to be important. But the relationship between knowledge and legal interpretation is not entirely clear. Roughly, the argument goes as follows: we are less likely to know a proposition when the practical stakes of its truth are raised, and similarly, we are less likely to assess that a text is clear when the practical stakes of its meaning are raised.⁹⁶E.g., Doerfler, supra note 25, at 523.

The linguistic defense of the MQD is clearly based in part on this philosophical literature about stakes and knowledge. Before interrogating the full argument, however, we must spell it out. Here we attempt to reconstruct the defense.

1.  Reconstruction of the “High Stakes” Linguistic Defense of the Major Questions Doctrine

(1)  [Empirical Premise 1: Stakes-Sensitive Knowledge]: The ordinary reader’s knowledge is sensitive to high stakes.⁹⁷Wurman, supra note 8, at 957 (“[O]rdinary speakers attribute ‘knowledge’—and, in turn, ‘clarity’—more freely or less freely depending upon the practical stakes.” (quoting Doerfler, supra note 25, at 527)).

(2)  [Empirical Premise 2: Stakes-Sensitive Clarity]: The ordinary reader’s understanding of textual clarity is sensitive to high stakes.⁹⁸Id. at 959 (“[O]rdinary readers and speakers are more likely to find the statute ambiguous in that [high-stakes] context than in a relatively lower-stakes context.”); see also id. at 917 (appealing to “how ordinary persons interpret instructions in high-stakes contexts”).

(3)  [Definition: MQD Case]: In a MQD case, the agency’s statutory powers are defined in linguistic terms that are semantically clear but highly general. The agency is exercising “vast powers” of great economic/political significance and pointing to the statutory language as authorization.⁹⁹See id. at 911 (summarizing the MQD as the idea that “[c]ourts should have ‘skepticism’ when statutes appear to delegate to agencies questions of major political and economic significance, which skepticism the government can only overcome ‘under the major questions doctrine’ by ‘point[ing] to “clear congressional authorization” to regulate in that manner’ ” (citation omitted)).

(4)  [Premise]: MQD cases involve a high-stakes context.¹⁰⁰Although Wurman, supra note 8, never explicitly describes the MQD cases as “high stakes,” we assume this premise is uncontroversial as part of a reconstruction of the argument. If they did not involve a high-stakes context, none of the arguments would be relevant.

(5)  [Textualist Premise]: Judges should interpret statutory language from the perspective of the ordinary reader.

(6)  [Minor Conclusion, from 1, 2, 3, 4, 5]: In a MQD case, the text is unclear.

(7)  [Premise]: If a text is unclear with respect to authorizing an agency’s action, it does not authorize that action.

(8)  [Major Conclusion, from 6, 7]: In a MQD case, the agency’s action is not authorized.

Attempting to construct the argument fully and precisely reveals several interesting features and questions. First, consider the two “Empirical Premises” (1 and 2). It is unclear exactly what function the first Empirical Premise (about knowledge) serves. It is included in the argument above because it features repeatedly and centrally in Wurman’s (and Doerfler’s) scholarship on high stakes, but even if that Premise were false, Premise 2 alone could support the argument.

Why, then, does the “high-stakes” literature emphasize knowledge in addition to textual clarity? Perhaps because there is little data bearing on the truth of Premise 2, but there is rich, decades-old philosophical literature that seemingly supports Premise 1.¹⁰¹E.g., Stewart Cohen, Contextualism, Skepticism, and the Structure of Reasons, 13 Phil. Persps. 57, 57 (1999); DeRose, supra note 24, at 913–18. As such, we understand the legal literature to be using Premise 1 as support for Premise 2: philosophers have concluded that knowledge is stakes sensitive, and this conclusion supports also concluding that textual clarity is stakes sensitive.

In Part III, we investigate the stakes-knowledge-clarity relationship empirically, but here we note some initial skepticism about the inference from knowledge to clarity. Law includes technical language,¹⁰²Frederick Schauer, Is Law a Technical Language?, 52 San Diego L. Rev. 501, 501 (2015). and as such, many ordinary people do not have direct knowledge of a law’s meaning. Nevertheless, this does not imply that a particular law is unclear, in the sense of being unclear to a legal expert or inherently indeterminate. Recent empirical work supports this point: ordinary readers understand law to include technical legal meanings, and they defer to legal experts to elaborate those meanings.¹⁰³Kevin Tobia, Brian G. Slocum & Victoria Nourse, Ordinary Meaning and Ordinary People, 171 U. Pa. L. Rev. 365, 365 (2023) [hereinafter Tobia et al., Ordinary People]. The mere fact that laypeople do not know the meaning of a law without further inquiry or assistance strikes us as an implausible basis for judges to treat the law as ambiguous or unclear.

Moreover, the “Minor Conclusion” (6) only follows with a very strong interpretation of the meaning of “sensitive to high stakes” (1) and (2). To conclude that “general” statutory language is unclear because of ordinary sensitivity to a high-stakes context, one must interpret (2) to mean that a high-stakes context eliminates clarity.

Wurman describes the MQD as limited to “resolving statutory ambiguities.”¹⁰⁴Wurman, supra note 8, at 940–41. This is a common way to describe a “tiebreaker” canon. We ultimately find this confusing insofar as Wurman also presents the MQD as a linguistic canon, a rule of thumb that is evidence of linguistic meaning. If “ambiguity” refers to linguistic ambiguity, an applicable “linguistic” canon would render the statute non-ambiguous. For example, in Lockhart the Court faced a linguistic ambiguity.¹⁰⁵Lockhart v. United States, 577 U.S. 347, 361 (2016). Lockhart was convicted under 18 U.S.C. § 2252(a) and faced a mandatory minimum due to an earlier conviction. The penalty increased if the defendant had a prior conviction “under the laws of any State relating to aggravated sexual abuse, sexual abuse, or abusive sexual conduct involving a minor or ward.”¹⁰⁶18 U.S.C. § 2252(b)(1). That final modifier (involving a minor or ward) could modify all three noun phrases (aggravated sexual abuse, sexual abuse, and abusive sexual conduct) or just the last (abusive sexual conduct). The series qualifier canon instructs us to apply the modifier to all three noun phrases. The determination that the series qualifier canon applies qua linguistic canon is a decision that the linguistic meaning of the provision is determinate and has a specific meaning, not that it is ambiguous. If ambiguity persists—for example, if there is a competing linguistic canon that counsels in favor of the opposite interpretation—the Court might resolve ambiguity with some non-linguistic consideration, such as the rule of lenity.

Alternatively, perhaps the argument is that the MQD is “linguistic” in the sense that it represents how ordinary people believe that ambiguity should be resolved, and thus how ordinary people would choose to resolve disputes in MQD cases. But that would be an unusual sense of “linguistic.” Existing linguistic canons help determine the linguistic meaning of a provision; they do not enter the interpretive process after that meaning has been concluded to be indeterminate.

This might all seem pedantic, but it highlights a problem with this linguistic defense of the MQD. We have done our best to explain the argument in a clear form, but we are unsure that there is even a workable argument for the “high stakes” linguistic MQD that arrives at the Major Conclusion (8).

Beyond this general issue (that the logic of the argument itself is unclear), several of the premises are open to debate. For example, perhaps some of the Court’s major questions cases do not involve high stakes or sufficiently high stakes (Premise 4).¹⁰⁷See Deacon & Litman, supra note 9, at 1009–10 (discussing and critiquing the Court’s criteria of majorness); Natasha Brunstein & Richard L. Revesz, Mangling the Major Questions Doctrine, 74 Admin. L. Rev. 217, 219 (2022) (discussing how the Trump Administration distorted the majorness determination by invoking the doctrine “enormously expansively and inconsistently”). Premise 7 is also controversial: just because a text’s meaning is unclear does not necessarily imply that it should be interpreted against an agency delegation (perhaps instead, it should be interpreted with a presumption of judicial nonintervention).¹⁰⁸Wurman acknowledges that this is a contestable claim. See Wurman, supra note 8, at 958 (noting that Doerfler views the question as whether judges should “demand more epistemic confidence” before overturning an expert agency’s interpretation). But Wurman suggests that “the legal system already contingently addresses this question” by presumptively disallowing agency action unless agencies “demonstrate authority for their actions” and thus satisfy their “burden of proof.” Id. at 960. Note the connection here to the theory of exclusive delegation, which is a nascent substantive grounding for the canon, not a linguistic one. See supra note 80 and accompanying text.

Nevertheless, most of our attention in this Article is on the two Empirical Premises, 1 and 2. Whatever the argument is, it is clear that these two premises are central: the “high-stakes” argument repeatedly appeals to these claims.¹⁰⁹See Wurman supra note 8, at 954–55. If these premises—and especially the second premise—are empirically invalid, the entire argument is a nonstarter. Part II of this Article presents evidence bearing on Premise 1, and Part III presents original empirical studies bearing on both Premise 1 and Premise 2. To preview the findings, (1) although academic philosophers have long assumed that higher stakes reduce knowledge, many studies find that stakes have no effect on ordinary people’s knowledge attributions;¹¹⁰See infra Part II. (2) we find a very small effect of stakes on knowledge (far from sufficient to conclude that “the ordinary reader” is stakes-sensitive about knowledge), and no effect of stakes on linguistic clarity.¹¹¹See infra Section III.A.

C.  The Major Questions Doctrine as an Anti-Literal Linguistic Canon

A second argument for the “linguistic” MQD surfaced in summer 2023. Justice Barrett’s concurrence in Biden v. Nebraska proposes that the MQD has a linguistic basis in ordinary people’s anti-literalism and sensitivity to context.

The crux of the argument is an appeal to the predicted reaction of ordinary people to everyday situations, such as Justice Barrett’s “babysitter” hypothetical:

Consider a parent who hires a babysitter to watch her young children over the weekend. As she walks out the door, the parent hands the babysitter her credit card and says: “Make sure the kids have fun.” Emboldened, the babysitter takes the kids on a road trip to an amusement park, where they spend two days on rollercoasters and one night in a hotel. Was the babysitter’s trip consistent with the parent’s instruction? Maybe in a literal sense, because the instruction was open-ended. But was the trip consistent with a reasonable understanding of the parent’s instruction? Highly doubtful. In the normal course, permission to spend money on fun authorizes a babysitter to take children to the local ice cream parlor or movie theater, not on a multiday excursion to an out-of-town amusement park. If a parent were willing to greenlight a trip that big, we would expect much more clarity than a general instruction to “make sure the kids have fun.”¹¹²Biden v. Nebraska, 143 S. Ct. 2355, 2379–80 (2023) (Barrett, J., concurring) (emphasis omitted).

Justice Barrett explains that additional context could make a difference, including (1) “maybe the parent left tickets to the amusement park on the counter,” (2) “[p]erhaps the parent showed the babysitter where the suitcases are, in the event that she took the children somewhere overnight,” (3) “maybe the parent mentioned that she had budgeted $2,000 for weekend entertainment,” (4) the “babysitter had taken the children on such trips before,” or (5) “if the babysitter were a grandparent.”¹¹³Id. at 2380. Notably, not all of these are additions to the text of the statement. We are sympathetic to this view of non-text-based context, but it is arguably a significant departure from traditional text-focused textualism.¹¹⁴See infra Section IV.B.

Moreover, Justice Barrett argues that the babysitter hypothetical illustrates how “we communicate conversationally” and that the MQD merely represents “common sense” in a different context:

In my view, the major questions doctrine grows out of these same commonsense principles of communication. Just as we would expect a parent to give more than a general instruction if she intended to authorize a babysitter-led getaway, we also “expect Congress to speak clearly if it wishes to assign to an agency decisions of vast ‘economic and political significance.’ ” That clarity may come from specific words in the statute, but context can also do the trick. Surrounding circumstances, whether contained within the statutory scheme or external to it, can narrow or broaden the scope of a delegation to an agency.¹¹⁵Biden v. Nebraska, 143 S. Ct. at 2380 (Barrett, J., concurring) (citation omitted).

This justification coheres with Justice Barrett’s “ordinary speaker” approach to interpretation. In Congressional Insiders and Outsiders, Justice Barrett argues that judges should approach language “from the perspective of an ordinary English speaker—a congressional outsider.”¹¹⁶Barrett, supra note 17, at 2194. This generally requires avoiding insider knowledge about Congress: “What matters to the textualist is how the ordinary English speaker—one unacquainted with the peculiarities of the legislative process—would understand the words of a statute.”¹¹⁷Id.

While Justice Barrett’s babysitter example is intriguing, it is not immediately clear how it supports the MQD. A skeptic might read the babysitter-to-MQD argument as committing a “motte” and “bailey” fallacy, conflating one position that is very easy to defend (the motte) with one much harder to defend (the bailey). It is undeniable that context influences interpretation and it would not be surprising that ordinary people are more confident in delegation of power with additional supporting contextual evidence. If the babysitter had previously taken the children on trips ((4) from above) or the agency had a longstanding practice of developing new programs, that context would often make readers equally or more confident that a text delegating authority to that agent encompasses similar action.

But this observation (that context can lend further support to particular actions taken pursuant to a delegation) does not justify the MQD. Justice Barrett’s key claim about ordinary language is much stronger, something like: ordinary people understand general delegations to X to be limited to only the most reasonable ways to X, absent further textual or contextual support for X. Recall Justice Barrett’s argument about the babysitter’s trip: “But was the trip consistent with a reasonable understanding of the parent’s instruction? Highly doubtful.”¹¹⁸Biden v. Nebraska, 143 S. Ct. at 2379–80 (Barrett, J., concurring). The central claim in the strong form of Justice Barrett’s argument is not merely that context matters but that absent supporting context, ordinary delegations are limited to the set of most reasonable applications of the instruction.

To appeal to the “motte” claim in support of the “bailey” claim is to trade an obvious fact about context to support a highly controversial claim about intuitive understanding of delegations. We do not, however, read Justice Barrett to make such a slippery move. There is a more charitable way to read her concurrence (that is, relying on the stronger key claim). This reading relies on an interesting and empirically testable question: When a text delegates an agent the power to X with general language, do people intuitively understand the delegation to be limited to only the set of the most reasonable/natural ways to X, or do they understand the delegation more broadly (even if not entirely literally)? For example, when a parent instructs a babysitter to “use this credit card to make sure the kids have fun this weekend,” does that authorize only the most reasonable actions (for example, ordering pizza, ordering a movie), or does it also authorize some actions that would be understood as less reasonable (for example, taking the kids to an amusement park)? Similarly, when Congress delegates to an agency, is the agency limited to only the set of most reasonable understandings (absent supporting context), or do people understand delegations to communicate a broader (if not quite literal) authorization?

Justice Barrett’s “linguistic defense” of the MQD leaves some questions open—the quotations above capture the bulk of the defense. Our formal reconstruction of the arguments follows.

1.  Reconstruction of the “Anti-Literalism” Defense of the Major Questions Doctrine

(1)  [Definition: Ordinary Majorness]: For a given rule, an action is “major” if the ordinary reader understands it, absent additional context, as not among the set of most reasonable ways to follow the rule.¹¹⁹A “major” action may be consistent with the rule’s literal meaning. The appeal to “reasonableness” generates an interesting feature of this definition: an action could be “major” in the sense of exceeding the reasonable set of actions or subceeding it. For example, imagine the babysitter responds by choosing to simply play board games with the kids, without using the credit card. It is possible that this is not among the most reasonable ways to follow the rule.

(2)  [Definition: MQD Case]: In a MQD case, the agency’s statutory powers are defined in linguistic terms that are semantically clear but highly general. The agency is exercising “vast powers” of great economic/political significance and pointing to the statutory language as authorization.

(3)  [Empirical Premise: MQD Cases Involve Ordinary Majorness] The ordinary reader takes MQD cases to involve a “major” action (for example, in the MQD cases, the ordinary reader takes the contested action, absent additional context, as not among the most reasonable ways to follow the rule).

(4)  [Textualist Premise]: Judges should interpret statutory language from the perspective of the ordinary reader.¹²⁰Some textualists might adopt a weaker premise: “In interpreting statutes, judges should employ some principles that guide the ordinary reader, some principles that guide an idealized or informed reader (for example, ‘reasonable reader’), and some principles that guide the expert reader (for example, ‘ordinary lawyer’).” Insofar as Justice Barrett’s linguistic MQD argument adopts something like this weaker premise, the argument only goes through if the weaker premise is supplemented with a further premise: “In MQD cases, textualists should employ the principles that guide an ordinary reader’s understanding of delegations of authority to agents.” Justice Barrett’s MQD argument relies heavily on her ordinary babysitter example, suggesting that—at least for the purpose of major questions cases—judges’ approach to language should include the ordinary reader’s understanding of delegations (including how the literal meaning of a delegation is restricted by context). For simplicity, our main argument uses the simpler but stronger premise, but it could also use the weaker (but more complicated) pair of premises.

This weaker premise also reveals hard questions for textualists, which are beyond the scope of this Article: When, exactly, should a textualist adopt one or other of these perspectives and principles? We are skeptical about textualists that freely shift among these perspectives, with no guiding principles. Justice Barrett herself has not clearly answered this question, sometimes treating the ordinary reader as the lodestar for interpretation and other times pointing to legally trained readers. See Barrett, supra note 17, at 2202. A defense of the MQD on the grounds that it reflects lawyerly training is arguably more substantive than linguistic, and more circular than logical, but we do not purport to address this defense of the MQD in this Article. See also Tobia et al., Ordinary People, supra note 103, at 432–34 (arguing that standards like “appropriately informed interpreter” are more normative than descriptive).

(5)  [Empirical Premise]: Absent additional context, the ordinary reader understands rules that delegate power to an agent to have significant contextual limitations against all “major” actions; such a rule’s communicative content is limited to authorizing only the set of most reasonable actions.

(6)  [Conclusion]: In MQD cases, absent additional context, judges interpreting delegations should interpret delegations to exclude all major actions.

II.  PHILOSOPHICAL AND EMPIRICAL BACKGROUND

The previous Part introduced the two linguistic MQD arguments, one concerning high stakes and one concerning anti-literalism. This Part provides background from philosophy and empirical studies related to these arguments.

Some of the questions at the heart of the “high-stakes” MQD defense have been long debated by epistemologists (philosophers who specialize in the study of knowledge). More recently, the same questions have been studied empirically by psychologists and experimental philosophers.¹²¹See generally A Companion to Experimental Philosophy (Justin Sytsma & Wesley Buckwalter eds., 2016) (for an overview of experimental philosophy). Much of this work challenges a premise in the high-stakes MQD argument: although philosophers have claimed high stakes impact knowledge, high stakes have (at most) a small effect on ordinary judgments of knowledge. Section II.A reviews this research.

Section II.B provides background related to Justice Barrett’s claims about context and anti-literalism. Context matters in interpretation, and recent research has found that ordinary people understand law in line with anti-literalism, as Justice Barrett notes. However, there is no extant research that supports the stronger empirical premise in the anti-literalism argument.

A.  Stakes and Knowledge

1.  Philosophical Epistemology of Stakes and Knowledge

For decades, philosophers have evaluated stakes’ impact on knowledge with hypothetical “thought experiments.”¹²²See, e.g., Cohen, supra note 101, at 58–60. Consider a pair of cases as an example.¹²³Rose et al., supra note 34, at 237–39 (for a discussion of this version of the case); see also DeRose, supra note 24, at 913–16. The only differences between cases are highlighted in italics.

(1)  Low-Stakes Bank Deposit:

Bob and Jane are considering whether to stop at the bank to deposit a check on a Friday. Nothing turns on whether they deposit the check in the next week. The line is long, and they consider coming back on Saturday. Bob says that he remembers that the bank was open last Saturday, and Jane replies that banks sometimes change their hours. Bob says, “I know the bank will be open tomorrow.”

In this case, many philosophers claim that Bob knows that the bank will be open tomorrow.¹²⁴Keith DeRose, The Ordinary Language Basis for Contextualism, and the New Invariantism, 55 Phil. Q. 172, 176 (explaining that “almost any speaker in my situation would claim to know the bank is open on Saturdays” in this low stakes case). Now consider a slight variation on this case.

(2)  High-Stakes Bank Deposit:

Bob and Jane are considering whether to stop at the bank to deposit a check on a Friday. It is critical that the check is deposited on one of the next two days. On Sunday, there will be a large debit to Bob’s account, which does not currently have enough funds, and the check is Bob’s only means to cover that expense. The line is long, and they consider coming back on Saturday. Bob says that he remembers that the bank was open last Saturday, and Jane replies that banks sometimes change their hours. Bob says, “I know the bank will be open tomorrow.”

In this case, philosophers say that Bob’s statement is false.¹²⁵Id. at 177 (“Almost everyone will accept [‘I don’t know if the bank is open’] as a reasonable admission, and it will seem true to almost everyone.”). He does not know the bank will be open tomorrow.

The epistemology literature has taken philosophers’ shared reactions to these cases as intuitive data. And philosophers have offered different theories to make sense of that data. These are rich and complicated philosophical debates, which this Article does not have the space to rehearse or explore deeply.¹²⁶For example, “contextualism” holds that “to know” is context sensitive, such that the truth conditions for knowledge attributions vary across contexts. Cohen, supra note 101, at 57; DeRose, supra note 24, at 914; see also Keith DeRose, Solving the Skeptical Problem, 104 Phil. Rev. 1, 4–5 (1995). “Interest-relative invariantism” (“IRI”) rejects the claim that knowledge is context sensitive; instead, IRI holds that practical factors impact whether knowledge obtains. Jason Stanley, Knowledge and Practical Interests 85–89 (2005). Our principal interest is in how this work has informed recent debates in legal philosophy.

Legal-philosophical scholarship has drawn on this work in epistemology in support of the claim that high-stakes legal interpretation differs from lower-stakes interpretation. Ryan Doerfler suggests that high-stakes contexts influence textual clarity,¹²⁷Doerfler, supra note 25, at 523; see also William Baude & Ryan D. Doerfler, The (Not So) Plain Meaning Rule, 84 U. Chi. L. Rev. 539, 546–47 (2017). and Wurman piggybacks on this premise to argue that stakes sensitivity supports the MQD.¹²⁸Wurman, supra note 8, at 957–61. Importantly, these legal applications appeal to “ordinary speakers”¹²⁹Doerfler, supra note 25, at 523, 542. and “ordinary epistemic justification,” especially reactions to the bank cases described above.¹³⁰Id. at 575. A starting premise is that, for ordinary speakers of ordinary language, stakes impact knowledge; this is typically illustrated by the low- and high-stakes bank example.

2.  Do Stakes Impact Knowledge? Empirical Perspectives

Despite the pedigree of the stakes-knowledge literature, there is one big problem: many empirical studies report that stakes have no effect on ordinary attributions of knowledge. As Joshua Knobe & Jonathan Schaffer explain, “[l]ooking at this recent evidence, it is easy to come away with the feeling that the whole contextualism debate was founded on a myth. The various sides offered conflicting explanations for a certain pattern of [stakes-sensitive] intuitions, but the empirical evidence suggests that this pattern of intuitions does not exist.”¹³¹Schaffer & Knobe, supra note 34, at 675–76.

Much of this evidence comes from “experimental philosophy.” Rather than relying on the intuitions of philosophers (some of whom might have a lot at stake in intuitions about contextualism), experimental philosophers examine the understandings of ordinary people. Moreover, they often conduct experiments, which present different participants with different versions of the same scenarios, varying in only one respect (for example, higher stakes). This allows experimenters to draw inferences about whether certain factors (for example, stakes) affect people’s judgments in these cases. Some readers may be familiar with experimental philosophy’s testing of the well-known “trolley dilemma.”¹³²See, e.g., Joshua D. Greene, R. Brian Sommerville, Leigh E. Nystrom, John M. Darley & Jonathan D. Cohen, An fMRI Investigation of Emotional Engagement in Moral Judgment, 293 Sci. 2105, 2105 (2001). Many have also poured substantial effort into testing the influence of stakes on knowledge, especially in the “bank cases.”

Do stakes affect lay attributions of knowledge? Many studies report no.¹³³Adam Feltz & Chris Zarpentine, Do You Know More When It Matters Less?, 23 Phil. Psych. 683, 697 (2010); Wesley Buckwalter, Knowledge Isn’t Closed on Saturday: A Study in Ordinary Language, 1 Rev. Phil. & Psych. 395, 403 (2010); Wesley Buckwalter & Jonathan Schaffer, Knowledge, Stakes, and Mistakes, 49 Noûs 201, 228 (2015); Rose et al., supra note 34, at 245; Kathryn B. Francis, Philip Beaman & Nat Hansen, Stakes, Scales, and Skepticism, 6 Ergo 427, 450–52 (2019); Joshua May, Walter Sinnott-Armstrong, Jay G. Hull & Aaron Zimmerman, Practical Interests, Relevant Alternatives, and Knowledge Attributions: An Empirical Study, 1 Rev. Phil. & Psych 265, 272–73 (2010). As one important example, consider the study conducted by David Rose and other contributing authors. They gave participants versions of the bank case described at the start of this Section. They collected data from over 3,500 participants across 16 countries. The vast majority of countries show no significant effect, and for the few that show an effect, the size is very small (about a 10% difference in low- versus high-stakes cases). The researchers conclude that, overall, there is “virtually no evidence that stakes affect knowledge attribution.”¹³⁴Rose et al., supra note 34, at 233.

Other papers report a complicated pattern for other epistemic notions besides knowledge. For example, Mark Phelan finds no effect of stakes on judgments about how (epistemically) confident someone should be in a between-subjects study, but he finds an effect in a within-subjects study (when the same participant considered matched cases).¹³⁵Mark Phelan, Evidence that Stakes Don’t Matter for Evidence, 27 Phil. Psych. 488, 505 (2014); see also May et al., supra note 133, at 272 (reporting an effect of stakes on confidence but not knowledge).

Other studies report stakes effects for more complicated (and perhaps controversial) measures of knowledge. As an example, consider Alexander Dinges and Julia Zakkou’s study.¹³⁶Alexander Dinges & Julia Zakkou, Much at Stake in Knowledge, 36 Mind & Language 729, 746 (2021). For another recent example, see generally Brian Porter, Kelli Barr, Abdellatif Bencherifa, Wesley Buckwalter, Yasuo Deguchi, Emanuele Fabiano, Takaaki Hashimoto, Julia Halamova, Joshua Homan, Kaori Karasawa, Martin Kanovsky, Hackjin Kim, Jordan Kiper, Minha Lee, Xiaofei Liu, Veli Mitova, Rukmini Bhaya, Ljiljana Pantovic, Pablo Quintanilla, Josien Reijer, Pedro Romero, Purmina Singh, Salma Tber, Daniel Wilkenfeld, Stephen Stich, Clark Barrett & Edouard Machery, A Puzzle About Knowledge Ascriptions, Noûs: Early View, July 4, 2024, at 1, available at https://onlinelibrary.wiley.com/doi/10.1111/nous.12515?af=R (finding no effect for questions like “[name] knows/only thinks he knows that [. . .],” but an effect for questions like “how many times do you think [name] has to check the logs before he knows [. . .]”). The weight of current evidence suggests that there is a small or no effect of stakes on knowledge attribution, but there is an effect of stakes on these other measures, such as questions about whether you “stand by” your claim or whether you should “check” your evidence more times. This study instructed participants to consider a scenario in one of three versions. All scenarios began with the following:

Picture yourself in the following scenario:

You and Hannah have been writing a joint paper for an English class. You have agreed to proofread the paper. You’ve carefully proofread the paper 3 times and used a dictionary if necessary. You spotted and corrected a few typos, but you didn’t find any typos in the last round anymore.

You meet up with Hannah to finally submit the paper. Hannah asks whether you think there are no typos in the paper anymore. You respond:

“I know there are no typos anymore.”

At this point, . . .

Then, the scenarios proceeded in either a “neutral,” “stakes,” or “evidence” version. The “stakes” manipulation sought to change the practical significance of the knowledge claim, while the “evidence” manipulation sought to change the evidence base on which the knowledge claim rests.

Neutral: . . . Hannah reveals to you for the first time that she’s always been a big fan of the Backstreet Boys. You’ve never liked the Backstreet Boys, but since you like Hannah, you promise to listen to a few songs she particularly recommends. You doubt that it will change your mind but agree that it doesn’t hurt to give it a try. As you’re about to submit the paper, Hannah asks whether you stand by your previous claim that you know there are no typos in the paper. You respond:

Stakes: . . . Hannah reveals to you for the first time that it is extremely important for her to get an A in the English class. Her scholarship depends on it, and she’ll have to leave college if she loses the scholarship. If there is a typo left in the paper, she’s very unlikely to get an A, so it is extremely important to her that there are no typos in the paper. As you’re about to submit the paper, Hannah asks whether you stand by your previous claim that you know there are no typos in the paper. You respond:

Evidence: . . . Hannah reveals to you for the first time that she’s secretly read your previous term papers and always spotted lots of typos in them even when you said you had carefully proofread them. She apologizes for not telling you earlier. You are slightly disappointed but forgive her. Hannah is a good friend, and you appreciate that she was honest with you in the end. As you’re about to submit the paper, Hannah asks whether you stand by your previous claim that you know there are no typos in the paper. You respond:

All scenarios ended with: “I do” or “I don’t,” asking participants to pick the response they would be more likely to give.

Using this “stand by” question, the researchers found a difference. In the “Neutral” version, 94% of participants stood by their knowledge claim (“I do”); in the “Stakes” version, 76% of participants stood by; and in the “Evidence” version, 42% stood by. The researchers found similar results in a bank case. The Neutral-Stakes difference suggests that stakes can impact knowledge attributions. The Stakes-Evidence difference indicates that other factors (for example, an attributor’s evidence base) also matter and can have a larger effect than stakes. This difference (76% versus 42%) is one of the larger differences reported in the literature.¹³⁷Dinges & Zakkou, supra note 136, at 735.

It is not clear if agreement with “standing by” a claim is equivalent to agreement with knowledge of a claim. To “stand by” a claim calls to mind the action associated with the claim (that is, going to the bank today or not). From a cost-benefit perspective, stakes are relevant to action. The rising expected cost of failing to act in light of a possible bank closure or paper typo is relevant to a rational actor’s decision-making. Arguably, some of the observed small impacts of stakes on lay attributions of knowledge could be reflecting lay participants’ actionability judgments: in the high-stakes context, Bob’s knowledge has not changed, but whether he should go to the bank has changed.

Overall, the evidence is mixed concerning whether stakes impact ordinary knowledge attributions. Historically, many philosophers had stakes-sensitive knowledge intuitions, predicted that others would, and developed complex theories about those effects.¹³⁸See, e.g., DeRose, supra note 24, at 913–18. Yet, a large number of empirical studies of thousands of ordinary participants, across many languages and cultures, have found no impact of stakes, or only a very small effect, on knowledge.¹³⁹See Schaffer & Knobe, supra note 34, at 703. Very recently, one new study has reignited the debate, finding some support for the impact of stakes on epistemological judgments.¹⁴⁰See Dinges & Zakkou, supra note 136, at 729. Another forthcoming paper also adopts a nuanced position that normative facts influence knowledge. See N. Ángel Pinillos, Bank Cases, Stakes and Normative Facts, in 5 Oxford Studies in Experimental Philosophy 375 (Joshua Knobe & Shaun Nichols eds., 2024). Yet, another recent study reports that stakes do not affect judgments about knowledge¹⁴¹Su Wu, Are Folks Purists or Pragmatic Encroachers? New Discoveries of Relation Between Knowledge and Action from Experimental Philosophy, Episteme 1, 11 (2023) (studying Chinese participants). but do affect judgment about action.¹⁴²Id. at 12. In total, there is evidence pointing in both directions. Resolving the debate will require further empirical research as well as systematic theorizing of the seemingly conflicting empirical results.

Consequently, it remains far from settled that high stakes reduce knowledge for “the ordinary person.” Most studies have found that stakes do not impact knowledge in this way. And even for the studies that do report an effect, it is small. If 95% of participants evaluate that there is knowledge in a low-stakes case, and 80% evaluate that there is knowledge in a comparable high-stakes case, does this imply that the “ordinary person” has stakes-sensitive knowledge intuitions? Advocates of ordinary stakes sensitivity need to spell out why stakes-sensitivity manifesting in 10–15% of ordinary participants implies that the ordinary reader has stakes-sensitive knowledge.

The claim that high stakes impact knowledge figures prominently in the argument for a high-stakes linguistic MQD.¹⁴³See supra Section II.A. Extant legal literature has drawn heavily on this claim in supporting that “high-stakes” interpretation differs from lower-stakes interpretation. In doing so, it has drawn primarily from hypotheticals in academic philosophy (the “bank cases”) and intuitions about those hypotheticals offered by academic philosophers. Insofar as the legal literature concerns stakes’ impact on ordinary people’s knowledge attributions,¹⁴⁴See, e.g., Wurman, supra note 8, at 956–61. those legal debates would benefit from greater engagement with the large body of recent empirical work summarized in the previous Section.

3.  From Philosophy to Legal Philosophy

The previous two Subsections have introduced the debate about stakes and knowledge in epistemology. But it is important to recall that the connection of this debate to legal philosophy requires another step. For example, Doerfler proposes a connection between “clarity” or “plain meaning” of a statute and knowledge about the statute’s meaning: “[T]o say that the meaning of a statute is ‘clear’ or ‘plain’ is, in effect, to say that one knows what that statute means.”¹⁴⁵Doerfler, supra note 25, at 527 (emphasis omitted); see also Baude & Doerfler, supra note 127, at 545. The logic appears to be that clarity attributions are a subset of knowledge claims, such that a property demonstrated to affect knowledge claims should transitively affect clarity claims.

Ultimately, this relationship between knowledge and clarity is outside the scope of our Article (the relevant question for the linguistic MQD is stakes’ impact on clarity). However, there are some philosophical questions to raise about the proposed relationship between clarity and knowledge. One, which we described earlier, concerns technical meaning. A layperson might not know what a statute means because it is technical, yet the statute may not be “unclear” to that person in the relevant sense of clarity (that is, ambiguous). As another difference, consider factivity. Philosophers often propose that knowledge is factive: I know p only if p. But it is not obvious that clarity is factive. The meaning of a statute might appear clear (that is, not ambiguous) to an agent while the agent is wrong about the statute’s meaning, and thus the agent lacks knowledge of the statute’s meaning. Such a case would be a counterexample to the claim that an agent knows what a statute means if and only if the meaning of the statute is clear.

Most importantly, the empirical evidence about ordinary attributions of knowledge reviewed here—to the extent that it even does support stakes sensitivity—does not necessarily extend to ordinary determinations of whether statutory text is clear. The studies to date mostly used the bank case, but the bank case presents no rule to which clarity judgments might attach. It might be possible that the clarity of rules is reduced for ordinary people in higher-stakes contexts. Indeed, it is theoretically possible that clarity judgments about textual rules are more sensitive to stakes than knowledge more generally. But it is just as possible that there is a breakage: that is, that clarity claims are not simply a subset of knowledge claims but a special and different kind of knowledge claim. However, as far as we are aware, these are entirely untested empirical hypotheses. Without any empirical evidence specific to clarity claims, it would not be possible to bootstrap ordinary stakes-sensitive clarity from ordinary stakes-sensitive knowledge (moreover, as we have argued, ordinary stakes-sensitive knowledge is also empirically dubious). Part III therefore tests this clarity claim.

B.  Context and Anti-Literalism

Justice Barrett’s concurring opinion in Biden v. Nebraska offers a different argument for the MQD as a linguistic canon. For Justice Barrett, the MQD simply reflects “common sense” inferences about how broader context restricts language’s (literal) meaning.¹⁴⁶Biden v. Nebraska, 143 S. Ct. 2355, 2379 (2023) (Barrett, J., concurring) (“Context also includes common sense, which is another thing that ‘goes without saying.’ Case reporters and casebooks brim with illustrations of why literalism—the antithesis of context-driven interpretation—falls short.”). Justice Barrett illustrates this with the babysitter example, claiming that ordinary people understand a delegation to a babysitter to have implicit limits (although a babysitter’s attempt to transgress those normal limits might be allowed by a supplemental clear authorization). This, Justice Barrett suggests, is precisely how an ordinary reader would read a statute delegating authority to an agency, and therefore a canon requiring a clear statement from Congress is justified.¹⁴⁷See supra Section I.C.

1.  Anti-Literalism and Context in Ordinary Language

Anti-literalism is an important feature of ordinary language. Consider François Recanati’s discussion of the “You are not going to die” example from Kent Bach:

[Imagine] a child crying because of a minor cut and her mother uttering . . . [“you are not going to die”] in response. What is meant is: “You’re not going to die from that cut.” But literally the utterance expresses the propositions that the kid will not die tout court—as if he or she were immortal. The extra element contextually provided (the implicit reference to the cut) does not correspond to anything in the sentence itself; nor is it an unarticulated constituent whose contextual provision is necessary to make the utterance fully propositional.¹⁴⁸François Recanati, Literal Meaning 8–9 (2004).

This example helpfully illustrates that we often understand propositions anti-literally, in light of context, and that the relevant context need not come from the statement itself. The very same words “you’re not going to die,” convey a different meaning when uttered after a child gets a cut than they would in some other context where the literal meaning would be the correct meaning.

The powerful influence of context is not limited to anti-literalism. Extratextual context can also disambiguate. As an example, consider the statement “Do not take drugs and alcohol.” Does this mean “Do not take either one?” Or does it mean “Do not take the two together?” The answer varies across contexts.

If this rule were presented in the context of a substance abuse counseling session, our extratextual knowledge about that session leads us to understand this text [to prohibit each individually]: Don’t take drugs; don’t take alcohol. However, if this rule were presented in the context of a patient’s annual physical, in which the doctor prescribed cholesterol-reducing medications, our extra-textual knowledge about that session encourages [understanding the rule to prohibit the combination].¹⁴⁹Kevin Tobia, Jesse Egbert & Thomas R. Lee, Triangulating Ordinary Meaning, 112 Geo. L.J. 23, 51 (2023).

2.  Anti-Literalism in Ordinary Understanding of Legal Rules

Justice Barrett’s argument is attractive in its appeal to context and anti-literalism. And Justice Barrett is not the only modern textualist to appeal heavily to anti-literalism; Justices Gorsuch and especially Kavanaugh have also called attention to the perils of overliteral interpretation.¹⁵⁰The Justices use “literal” in various ways, but Justice Gorsuch and Kavanaugh have recently called attention to avoiding inappropriate literalism. See, e.g., Bostock v. Clayton County, 140 S. Ct. 1731, 1750 (2020) (Gorsuch, J.); id. at 1825 (Kavanaugh, J., dissenting) (“[C]ourts must follow ordinary meaning, not literal meaning.”).

For modern textualism, this is a welcome development. Analysis of the (linguistic) meaning of legal rules should attend to context and exceed pure literalism. As one example, consider the linguistic canons. Many linguistic canons reflect intuitive contextual restrictions from literal meaning. “No cars, trucks, or other vehicles may enter the park” might literally prohibit bicycles from the park, as most ordinary people take a bicycle to be a vehicle.¹⁵¹Kevin P. Tobia, Testing Ordinary Meaning, 134 Harv. L. Rev. 726, 757 (2020) (reporting that most laypeople, law students, and judges agree that a bicycle is a “vehicle”). However, the principle of ejusdem generis instructs interpreters to construe the broad, catchall term “vehicle” in light of the listed items (“cars,” “trucks”).¹⁵²See McBoyle v. United States, 283 U.S. 25, 26–27 (1931). Even if laypeople are not familiar with the name “ejusdem generis,” they intuitively apply this kind of reasoning when analyzing both legal and ordinary rules.¹⁵³Tobia et al., From the Outside, supra note 18, at 259–60.

People also apply other types of contextual restrictions from literal meaning. This includes some contextual rules that are not currently recognized by courts as linguistic canons. For example, people understand that universal quantifiers like “any” often do not mean literally any.¹⁵⁴Id. (reporting studies demonstrating that laypeople intuitively apply a ejusdem generis principle); see also Tobia, supra note 151, at Appendix (reporting that most laypeople do not take “no vehicles in the park” to prohibit a bicycle from the park, even though most laypeople agree that a bicycle is a “vehicle”). If this tendency were at least as systematic in ordinary understanding as those underlying conventional linguistic canons (for example, the tendency to restrict catchall terms as ejusdem generis reflects), a textualist committed to the ordinary reader should employ those new canons (for example, the “quantifier domain restriction canon”).

Recent legal scholarship has also asked whether thinking about context and anti-literalism might reveal that some “substantive” canons are also linguistic canons.¹⁵⁵Slocum & Tobia, supra note 37, at 73. Some clear statement rules—such as anti-retroactivity and anti-extraterritoriality—could be seen as linguistic canons, based on our understanding of context. Taken literally, many statutes would seem to apply at all times, in all places.¹⁵⁶E.g., 18 U.S.C. § 2119 (“Whoever, with the intent to cause death or serious bodily harm takes a motor vehicle that has been transported, shipped, or received in interstate or foreign commerce from the person or presence of another by force and violence or by intimidation, or attempts to do so, shall . . . be fined under this title or imprisoned not more than 15 years, or both.”). But people understand statutes to communicate temporal and geographical restrictions: while there is some division among laypeople, overall, people tend to understand rules to apply only prospectively, and only territorially.¹⁵⁷Slocum & Tobia, supra note 37, at 81–96.

Textualists may rhetorically privilege the “ordinary reader” and express support for anti-literalism, but they have not yet adopted many of these suggestions. For instance, no textualist has adopted an anti-literal “quantifier domain restriction canon” or theorized anti-retroactivity as a linguistic canon (although it is a long-standing clear statement rule). These context-sensitive rules are relatively robust and systematic and are supported by empirical evidence. We have reservations about a textualism that ignores such systematic patterns of anti-literalism while also freely adopting “ad hoc” anti-literal arguments related only to particular cases. On this score, Justice Barrett’s concurrence in Biden v. Nebraska is commendable in hypothesizing about a broader contextual principle that generally guides ordinary understandings of delegations (that is, a principle applying across cases, not an ad hoc appeal to context and anti-literalism related only to the authorization of emergency student loan relief). Whether Barrett’s contextual principle is systematic and empirically supported is a separate question.

Anti-literalism and contextual restriction are powerful ideas that accurately reflect language usage, but if textualists have no theory about when one can appeal to them, there is a danger that textualists can freely frame different readings as “literal” and “anti-literal,” choose liberally among them, or simply ignore non-literal meanings when doing so is convenient.¹⁵⁸See id. at 106–08; see also William N. Eskridge, Jr., Brian G. Slocum & Kevin Tobia, Textualism’s Defining Moment, 123 Colum. L. Rev. 1611, 1612–27 (2023) (documenting twelve theoretical choices facing modern textualists and arguing that textualists’ failure to explain their answers to these choices facilitates cherry-picking and undermines rule of law values like predictability). The claim that “in context,” a text does not “literally” mean what it says is also a powerful way for motivated interpreters to escape a text’s clear meaning.

Context matters. But if textualists have no theory about what counts as context and when they must appeal to it, ad hoc appeals to context are like “looking out over a crowd and picking out your friends.”¹⁵⁹See Confirmation Hearing on the Nomination of John G. Roberts, Jr. to Be Chief Justice of the United States Before the S. Comm. on the Judiciary, 109th Cong. 200–01 (2005) (on looking to foreign law in U.S. constitutional interpretation). Except here, the “friends” are not even limited to preexisting sources; they also include entirely novel hypothetical examples generated by the judge.

3.  Contextual Restriction of Delegations?

As Section II.B argued, the “anti-literalism” argument of the linguistic MQD needs a stronger premise than simply “people sometimes understand language non-literally.” The mere fact that “you are not going to die” has a nonliteral meaning does not justify the MQD.

The premise necessary to the argument involves a new claim about ordinary understanding of delegations. Justice Barrett proposes that there is some MQD-like principle that is part of ordinary people’s common sense, concerning the limited authorization from a general delegating instruction. It is for this reason that she relies on the babysitter hypothetical, an anti-literalism intuition-pump about an ordinary instruction that delegates power to an agent. General delegation language, Justice Barrett posits, has an anti-literal limitation. Unless there is further specific authorization, that general language is understood to be limited to only the most reasonable actions.

This is an interesting and empirically testable proposition: ordinary people understand general delegations to be limited to only the most reasonable actions falling under the language of the delegation. As far as we know, there is no empirical study that has examined this question. We present a new study to do so in Section III.B.

III.  NEW EMPIRICAL EVIDENCE

This Part tests key empirical claims at the core of the linguistic arguments for the MQD. In both tests, we seek to reduce our researcher degrees of freedom (that is, eliminate cherry-picking scenarios) by relying on the exact cases that advocates of the linguistic defense offer: the high-stakes “bank case” and the “babysitter hypothetical.”

Section III.A presents a study that tests whether ordinary people’s judgments about knowledge are lowered in high-stakes contexts (using the bank case). It also examines, for the first time, whether people’s understanding of a rule is impacted: Are rules perceived as less clear in high-stakes contexts?

Section III.B presents a study to examine the babysitter case: a parent instructs the babysitter to use a credit card to “make sure the kids have fun.” Do ordinary people understand this instruction to license taking the children on a road trip to an amusement park, or do they understand it to be limited to only more reasonable actions?

Section III.C responds to the primary two objections to the studies that have appeared in print since we first publicized this Article’s empirical findings.

A.  Do High Stakes Reduce Knowledge and/or Clarity? The Bank Case

1.  General Overview

The first study examined whether (high) stakes reduce ordinary attributions of (1) knowledge and (2) clarity of rules. We randomly assigned participants to either a low-stakes¹⁶⁰Rose et al., supra note 34, at 231. Low:

Bob and his wife are driving home on a Friday afternoon. They both received some money earlier in the day, and so they plan to stop at the bank on the way home to deposit it. But as they drive past the bank, they notice that the lines inside are very long, as they often are on Friday afternoons. Although they generally like to deposit any money they receive at the bank as soon as possible, it is not especially important in this case that it be deposited right away, and so Bob suggests that they drive straight home and deposit their money on Saturday morning. His wife says, “Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.” Bob replies, “No, I know the bank will be open. I was just there two weeks ago on Saturday. It was open until noon.” As a matter of fact, the bank will be open on Saturday morning.

Id. or high-stakes¹⁶¹Id. High:

Bob and his wife are driving home on a Friday afternoon. They both received some money earlier in the day and so they plan to stop at the bank on the way home to deposit it. But as they drive past the bank, they notice that the lines inside are very long, as they often are on Friday afternoons. They have recently written a very large and very important check. If the money is not deposited into their bank account before Monday morning, the important check they wrote will not be accepted by the bank, leaving them in a very bad situation. Bob suggests that they drive straight home and deposit their money on Saturday morning. His wife says, “Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.” Bob replies, “No, I know it’ll be open. I was just there two weeks ago on Saturday. It was open until noon.” As a matter of fact, the bank will be open on Saturday morning.

Id. condition of the bank case. In each condition, participants read a version of the famous bank case, in which Bob and his wife discuss whether a bank is open on Saturday. Participants answered two types of knowledge questions, drawn from the previous literature.¹⁶²See id. at 229–32. The basic knowledge question asks:

In your personal opinion, when Bob says “I know the bank will be open” is his statement true?

Yes, Bob’s statement is true.

No, Bob’s statement is not true.

Defenders of context sensitivity have argued that this question more accurately tracks debate about contextualism than questions that simply ask participants to rate “knowledge.”¹⁶³See DeRose, supra note 31, at 82. The “strict” knowledge question asks:

In your personal opinion, which of the following sentences better describes Bob’s situation?

Bob knows that the bank will be open on Saturday.

Bob thinks he knows that the bank will be open on Saturday, but he doesn’t actually know it will be open.

Next, we randomly assigned participants to one type of rule: Clear, Ambiguous 1, Ambiguous 2, Unclear. The study presented a vignette explaining that Bob’s wife now used her phone to find the bank’s policy on its website. We randomly presented participants with one of four types of rules:

• [Clear] The bank is open on Saturdays.
• [Ambiguous 1] The bank is closed on Sundays.
• [Ambiguous 2] The bank is closed only on Sundays and federal holidays.
• [Unclear] The bank is open during regular business hours.

Participants rated whether the rule is clear or unclear concerning whether the bank is open on Saturday:

Now imagine that Bob’s wife uses her phone to search for the bank’s policy. She finds a website for the local bank branch. The website’s text states: “[RULE]” In your personal opinion, is this rule’s meaning clear or unclear concerning whether the bank is open on Saturday?

Clear: The bank is open on Saturday.

Clear: The bank is closed on Saturday.

Unclear.

In sum, we experimentally varied two factors: Stakes (low, high) and Rule Type (Clear, Ambiguous 1, Ambiguous 2, Unclear). This study examines whether Stakes affect lay judgment of knowledge (basic and strict). The study also examines whether Stakes affect lay judgment of a rule’s clarity across hypothesized clear, ambiguous, and unclear rules.

2.  Methodological Details

All study materials, hypotheses, exclusion criteria, and primary analyses were preregistered at Open Science.¹⁶⁴Kevin Tobia, Stakes and Legal Interpretation, Center for Open Science (July 12, 2023, 09:21 AM), https://osf.io/adw2n [https://perma.cc/9MVV-BR2J]. The study data is also available at the same site. A total of 501 participants were recruited from Prolific.co and compensated $1.00 ($12.00/hr) for a 5-minute task. To be eligible, participants must have completed at least 10 tasks on Prolific, with a 100% approval rating, and they must currently reside in the United States.

Within the study, there were several check questions. First was a simple attention check question, which asked participants to select the answer “purple” in a long list of colors. There was also a manipulation check, clearly labeled as an “attention check”: “Attention check question: According to the story, which of the following statements is correct?” The options were “it is very important that Bob and his wife deposit their money” [correct answer in high-stakes condition] and “it is not very important that Bob and his wife deposit their money” [correct answer in low-stakes condition]. Later in the study, there was a third multiple choice attention check: “Alex is taller than Sam, and Sam is taller than John. Who is the shortest?” [correct answer = “John”; incorrect answers = “Alex,” “Sam,” “They are all the same height”]. Finally, all participants were asked to complete a CAPTCHA. Participants who answered any one of these questions incorrectly were excluded from the analyses. Thirty-two (out of 501; i.e., 6%) participants were excluded from these criteria.

3.  Results

A total of 469 participants were included in the data analysis (mean age = 39.58; 50% men, 48% women, 1% non-binary).

A binomial logistic regression revealed an effect of Stakes on knowledge. Participants attributed knowledge less in high-stakes cases (prob. = 0.86, 95% CI = [0.81, 0.90]) than in low-stakes cases (prob = 0.95, 95% CI = [0.91, 0.97]), odds ratio = 0.35, 95% CI = [0.18, 0.70], z = -2.99, p = 0.003.¹⁶⁵See infra Figure 1.

Figure 1.

Figure 1. Percentage attributing knowledge (top panel) and strict knowledge (bottom panel), in low- and high-stakes bank cases. In the high-stakes case, knowledge attributions were slightly (about 10%) lower. Overall, the majority of participants attributed knowledge in low- and high-stakes cases.

A binomial logistic regression revealed an effect of Stakes on strict knowledge. Participants attributed strict knowledge less in high-stakes cases (prob. = 0.66, 95% CI = [0.60, 0.71]) than in low-stakes cases (prob = 0.78, 95% CI = [0.72, 0.83]), odds ratio = 0.55, 95% CI = [0.37, 0.83], z = -2.85, p = 0.004.¹⁶⁶See supra Figure 1.

A multinomial logistic regression examined the effect of Stakes (low, high) and Rule Type (Clear, Ambiguous 1, Ambiguous 2, Unclear) on judgment of the bank rule’s clarity (clearly open, clearly closed, unclear). First, consider the effect of Stakes. Comparing clearly open and clearly closed responses, there was no effect of Stakes, z = 0.06, p = 0.956. Comparing clearly closed and unclear responses, there was no effect of Stakes, z = 0.38, p = 0.705. Next, consider the effect of Rule Type. Comparing clearly open and clearly closed responses, there was a significant effect of the clear versus unclear rule, z = -3.07, p = .002. There was no significant effect among the other rule types, |zs| < 0.21, ps > 0.8. Comparing clearly closed and unclear responses, there were no significant rule type effects, |zs| < 0.2, ps > 0.85. Finally, there were no significant Stakes * Rule Type interactions, |zs| < 0.41, ps > 0.68.¹⁶⁷See infra Figure 2.

Figure 2.

Figure 2. Percentage attributing a clear meaning (open or closed) or unclarity for four different rules in low- and high-stakes cases. There were large and significant differences among the rules’ perceived meaning: the “Obviously Clear” and “Ambiguous 2” rules were generally understood to mean clearly open; the “Ambiguous 1” rule was understood to be unclear or mean clearly open; and the “Obviously Unclear” rule was unclear. However, there was no impact of high stakes on clarity judgments for any type of rule, whether the rule was clear (for example, Obviously Clear), ambiguous (for example, Ambiguous 1), or unclear (for example, Obviously Unclear).

4.  Discussion

The results regarding stakes and knowledge are consistent with the prior literature. Some previous studies have found a small effect of stakes on knowledge in the United States.¹⁶⁸E.g., Rose et al., supra note 34, at 235 (finding a small pattern in the U.S., but not in most other countries). Here, we find a similar small effect: In the low-stakes bank case, 95% attribute knowledge, but in the high-stakes bank case, this number drops to 86%. The “strict knowledge” measure reflects a similarly sized difference (78% versus 66%).

i.  Is Knowledge “Sensitive” to Stakes?

The empirical results clarify the importance of refining this philosophical question: What is it for ordinary knowledge to be “sensitive to stakes”? One (weak) interpretation is that in some circumstances, for some people, stakes affect knowledge attributions. A stronger interpretation is that for most or all people, there are some cases in which knowledge is lost in high-stakes contexts. The strongest interpretation is that in many or most circumstances, high stakes defeat knowledge (for many or most people).

Once we have greater philosophical clarity about what it means to say knowledge is sensitive to stakes, we can analyze those theses in light of the empirical results. The results here straightforwardly provide support for the weak interpretation: the high-stakes manipulation affects (some participants’) attributions of knowledge. But the results do not support the “stronger” or “strongest” interpretations. The vast majority of participants attributed knowledge in low- and high-stakes cases. And even for the “strict knowledge” question, most participants still judged that there was (strict) knowledge in the high-stakes scenario. In other words, for the vast majority of participants, stakes did not impact knowledge.

ii.  Do High Stakes Reduce Clarity?

The results provide a more straightforward answer to this question. The high- versus low-stakes manipulation had no impact on whether people understood rules to be clear or unclear. Importantly, we used four types of rules, which varied in their basic level of clarity. With respect to whether the bank is open Saturday, “the bank is open on Saturday” is obviously clear; “the bank is closed on Sunday” is ambiguous; “the bank is closed only on Sundays and federal holidays” is ambiguous;¹⁶⁹Note, we hypothesized that this rule has some ambiguity, given that the scenario does not specify whether the following Saturday is a federal holiday. Participants generally overlooked this possibility or assumed that the next day was not a holiday. Thus, the “Ambiguous 2” stimulus could be treated as another example of “obviously clear” text. The “Ambiguous 1” rule was much more often understood as unclear. and “the bank is open during regular business hours” is unclear. For all of these rules, high stakes did not increase the base level of unclarity.¹⁷⁰See supra Figure 2.

B.  Ordinary Understanding of Delegations: The Babysitter Case

The second study examines how ordinary Americans understand delegations in an ordinary context. This Study takes inspiration from Justice Barrett’s recent concurrence in Biden v. Nebraska, which offered a new linguistic defense of the MQD.

1.  General Overview

The second study examined Premise 5 from Justice Barrett’s argument, the second empirical premise: When assessing whether an agent has followed or disobeyed a rule granting authority to perform some actions, do ordinary people restrict the rule’s literal meaning to only the set of most reasonable actions (absent additional context)?¹⁷¹See supra Section I.C. Study 2 examines this question by presenting participants with an ordinary rule granting authority, followed by one of five possible actions. These five actions varied in their anticipated reasonableness, and we examined whether participants evaluated each as following or violating the rule.

As in Study 1, we sought to minimize our researcher degrees of freedom by relying on existing and important test cases that have been offered by advocates of the linguistic MQD. For Study 2, we chose Justice Barrett’s “babysitter” hypothetical, as well as Justice Barrett’s proposed “major” action: a babysitter taking children to an amusement park in response to the instruction “Use this credit card to make sure the kids have fun this weekend.”

We randomly varied the conventional gender of the parent’s name (Patrick or Patricia) and babysitter’s name (Blake or Bridget). This did not affect rule violation judgment. Below is the text of the scenarios with the names Patricia and Blake:

Imagine that Patricia is a parent, who hires Blake as a babysitter to watch Patricia’s young children for two days and one night over the weekend, from Saturday morning to Sunday night. Patricia walks out the door, hands Blake a credit card, and says: “Use this credit card to make sure the kids have fun this weekend.”

Next, the scenario continued in one of five ways:

[MISUSE] Blake only uses the credit card to rent a movie that only he watches; Blake does not use the card to buy anything for the children.

[MINOR] Blake does not use the credit card at all. Blake plays card games with the kids.

[REASONABLE] Blake uses the credit card to buy the children pizza and ice cream and to rent a movie to watch together.

[MAJOR] Blake uses the credit card to buy the children admission to an amusement park and a hotel; Blake takes the children to the park, where they spend two days on rollercoasters and one night in a hotel.

[EXTREME] Blake uses the credit card to hire a professional animal entertainer, who brings a live alligator to the house to entertain the children.

All scenarios concluded with:

The kids have fun over the weekend.

We anticipated that the five scenarios would be seen as varying in their “reasonableness” as a response to the rule “Use this credit card to make sure the kids have fun this weekend,” with the REASONABLE scenario as maximal and the others as less reasonable. As we describe below, this prediction was borne out.

In all of the questions, we randomly varied whether the scenario described the parent’s directive as an “instruction” or “rule.” This also had no effect on rule violation judgment. Below we present the questions using the term “instruction.” After reading the scenario, participants first answered a comprehension question:

Attention check question: According to the story, which of the following statements is correct?

[CORRECT] Patricia’s instruction was “Use this credit card to make sure the kids have fun this weekend.”

Patricia’s instruction was “Do not use this credit card to make sure the kids have fun this weekend.”

Patricia’s instruction was “Use this credit card for anything this weekend.”

Patricia’s instruction was “Do not use this credit card for anything this weekend.”

Next, participants answered the rule violation question:

[Rule Violation] In your personal opinion, which better describes this situation?

Blake followed the instruction.

Blake violated the instruction.

We also measured participants’ judgment of the rule’s literal meaning and purpose.¹⁷²[Literal Meaning] “Think about what the instruction ‘Use this credit card to make sure the kids have fun this weekend’ means literally. In your personal opinion, did Blake’s actions comply with or violate the literal meaning of the instruction? Blake complied with the rule’s literal meaning; Blake violated the rule’s literal meaning” and [Purpose] “Think about the underlying purposes of Patricia’s instruction. In your personal opinion, did Blake’s actions support or oppose the instruction’s underlying purposes? Blake’s actions supported the instruction’s underlying purpose; Blake’s actions opposed the instruction’s underlying purposes.” Finally, we measured participants’ evaluation of whether the babysitter’s action was a reasonable response to the instruction:

[Reasonableness] Think about how Blake responded to Patricia’s instruction. In your personal opinion, is this an unreasonable or reasonable way to respond to that instruction?

(completely unreasonable) 1  2  3  4  5  6  7 (completely reasonable)

2.  Methodological Details

As for Study 1, all Study 2 materials, hypotheses, exclusion criteria, and primary analyses were preregistered at Open Science.¹⁷³See Tobia, supra note 164. The study data is also available at the same site. A total of 500 participants were recruited from Prolific.co and compensated $1.00 ($12.00/hr) for a 5-minute task. To be eligible, participants must have completed at least 10 tasks on Prolific, with a 100% approval rating, they must currently reside in the United States, and they must not have taken Study 1. Within the study, there were the same two check questions used as exclusion criteria in Study 1 (attention check and transitivity) and the new comprehension check described in the previous Section. Twenty-four (out of 499; i.e., 4.8%) participants were excluded with this criteria.

3.  Results

A total of 475 participants were included in the data analysis (mean age = 37.74; 48% men, 50% women, 2% non-binary).

First, we examined whether the five acts differed in their perceived reasonableness with respect to the rule. A linear regression revealed significant effects of the Action (misuse, minor, reasonable, major, extreme). Compared to ratings for the “reasonable” act (buying pizza and a movie for the kids), ratings for the misuse act (buying a movie for only the babysitter) were significantly lower, β = -1.67, 95% CI = [-1.89, -1.46], p < .001; ratings for the minor act (playing cards rather than purchasing anything) were significantly lower, β = -0.48, 95% CI = [-0.69, -0.27], p < .001; ratings for the major act (purchasing the amusement park trip) were significantly lower, β = -1.03, 95% CI = [-1.24, -0.82], p < .001; and ratings for the extreme act (purchasing the alligator entertainer) were significantly lower, β = -1.77, 95% CI = [-1.98, -1.56], p < .001.¹⁷⁴See infra Figure 3.

Figure 3.

Figure 3: Reasonableness Ratings. Ordinary judgments of an action’s reasonableness in the babysitter hypothetical. Higher scores indicate greater reasonableness (1–7 scale).

Next, we examined which of the five acts participants understood as instances of following or disobeying the instruction. A binomial logistic regression revealed effects of Act type on rule violation. For the misuse case, rule following prob. = 0.15, 95% CI = [0.09, 0.24]; for the minor case, rule following prob. = 0.51, 95% CI = [0.41, 0.61];¹⁷⁵This differed significantly from the misuse case, odds ratio = 5.88, 95% CI = [2.94, 11.79], z = 5.00, p < 0.001. for the reasonable case, rule following prob. = 1.00, 95% CI = [0.00, 1.00];¹⁷⁶All participants in the reasonableness condition answered, “rule followed.” for the major case, rule following prob. = 0.92, 95% CI = [0.84, 0.96];¹⁷⁷This differed significantly from the misuse case, odds ratio = 62.07, 95% CI = [24.73, 155.79], z = 8.79, p < 0.001. and for the extreme case, rule following prob. = 0.90, 95% CI = [0.82, 0.94].¹⁷⁸This differed significantly from the misuse case, odds ratio = 49.66, 95% CI = [20.88, 118.11], z = 8.83, p < 0.001.

4.  Discussion

This Study aimed to test the empirical claims underlying the “babysitter hypothetical,” an example that has been used to support claims in a linguistic defense of the MQD.

i.  Do People Understand Different Actions to Vary in Their Reasonableness as a Response to the Rule “Use This Credit Card to Make Sure the Kids Have Fun This Weekend”?

Yes. People evaluated some actions as highly reasonable, such as buying pizza and a movie for the kids. Other actions appeared less reasonable, like taking the kids to an amusement park or simply playing cards (and not buying anything). Others were even less reasonable, such as hiring an alligator entertainer or using the card to only purchase something for the babysitter. These results are unsurprising, but this variation is essential to test the key claim that the babysitter hypothetical has been offered to demonstrate.

ii.  Do People Understand Authorizing Rules to Be Limited to Only the Set of Most Reasonable Actions?

No. Although people evaluate Justice Barrett’s “major” action (taking the kids to an amusement park) as less reasonable than at least one alternative, they nevertheless understand it as consistent with the rule. Moreover, people evaluated the even more extreme example of bringing a live alligator to the house as consistent with the rule.

To be sure, people did rule out some actions as impermissible. In particular, the respondents overwhelmingly said that misuse of the credit card for the babysitter’s benefit rather than that of the children violated the rule. They also divided roughly evenly over the babysitter’s decision to forgo using the credit card at all. We will have more to say about these interesting patterns in Part IV,¹⁷⁹See infra Part IV. but for now, the most important thing to note is that two of the less reasonable actions that tested the boundaries of the instruction were nevertheless deemed to be within the parent’s rule.

iii.  Why Do People’s Judgments About an Act’s Reasonableness and Rule Violation Differ?

Our survey also included questions about the rule’s literal meaning and the rule’s purposes. First consider reasonableness judgments by considering the results for purpose and literal meaning. Figure 4 presents the results for the purpose question. On inspection, this pattern of purpose attributions across actions is similar to the pattern of reasonableness ratings (Figure 3): actions seen as more reasonable were also the ones seen as most supportive of the rule’s purposes. The ratings for purpose and reasonableness, r = 0.63, 95% CI = [0.57, .0.68], p < .001, were more highly correlated than the ratings for purpose and literal meaning, r = 0.39, 95% CI = [0.31, .0.47], p < .001.

Next consider judgments about rule violation. Both literal meaning and purpose were correlated with rule violation judgment, but rule violation was more strongly correlated with literal meaning, r = 0.67, 95% CI = [0.62, .0.72], p < .001, than purpose, r = 0.49, 95% CI = [0.42, .0.56], p < .001.

Figure 4.

Figure 4: Purpose Ratings. Ordinary judgments of whether an action supports (rather than opposes) the rule’s purposes in the babysitter hypothetical.

These analyses are exploratory and further work is required to more fully understand the differences in participants’ judgments about whether an action is reasonable and whether it violates a rule, but the Study here clearly shows a difference in these judgments.¹⁸⁰See supra Table 1. The question of whether the rule was violated and the question of whether the action was a reasonable response to the rule are understood differently by ordinary people: These questions are not synonymous. The comparisons to the purpose measure suggest a stronger relationship between reasonableness and purpose than rule violation and purpose.

Textualists concerned with the ordinary meaning of rules would presumably favor the rule violation question over the reasonableness question. Textualists who place significant weight on whether an action was “reasonable” with respect to a rule may be incorporating purposive reasoning, which is not as clearly relevant to ordinary people’s straightforward understanding about whether an act violates a rule.

The results reported here about laypeople’s rule violation judgments are consistent with prior work. Previous studies have found that both text (operationalized as literal meaning) and purpose influence rule violation judgment, but the former has a stronger influence.¹⁸¹Ivar R. Hannikainen, Kevin P. Tobia, Guilherme da F. C. F. de Almeida, Noel Struchiner, Markus Kneer, Piotr Bystranowski, Vilius Dranseika, Niek Strohmaier, Samantha Bensinger, Kristina Dolinina, Bartosz Janik, Eglė Lauraitytė, Michael Laakasuo, Alice Liefgreen, Ivars Neiders, Maciej Próchnicki, Alejandro Rosas, Jukka Sundvall & Tomasz Żuradzki, Coordination and Expertise Foster Legal Textualism, 119 Proc. Nat’l Acad. Scis., no. 44, 2022, at 1, 6; Kevin Tobia, Experimental Jurisprudence, 89 U. Chi. L. Rev. 735, 783–91 (2022) (summarizing research on lay judgment about legal interpretation). See generally Guilherme da Franca Couto Fernandes de Almeida, Noel Struchiner & Ivar Hannikainen, Rules, in Cambridge Handbook of Experimental Jurisprudence (Kevin Tobia ed., forthcoming 2024) (reviewing recent empirical studies about the effect of text and purpose on laypeople’s rule violation judgments). In sum, ordinary people lean towards textualism, but not the “common sense” limitations claim at the heart of the linguistic MQD.

C.  Objections

This Section considers the two primary objections that have been raised in print about the results since we first circulated a draft of this Article.

1.  Objection 1: Subjects Must Be Sensitive to Stakes

One objection concerns stakes sensitivity. Wurman writes, “In conversation, Ryan Doerfler has pointed out that it does not appear that the participants [in this Article’s Study 1] were asked whether the rule was clear to Bob, as opposed to themselves, and Bob is the one sensitive to stakes in the example.”¹⁸²Wurman, supra note 8, at 961 n.271.

It is not clear why this observation constitutes an objection. One version of this objection is that only the judgments of those directly impacted by the stakes are relevant to legal theory, and because our study’s participants are not themselves impacted by the bank’s closure, their judgments about knowledge and clarity are not useful. This objection proves too much. The legal literature theorizing the effects of stakes-on-knowledge and stakes-on-clarity draws heavily on philosophical thought experiments (especially the bank case about Bob). None of these examples involve high stakes for the thought experimenter. The stakes are always for the subject described in the scenario, like Bob. The assumption is that those considering the scenarios can evaluate the significance of stakes (for some other person). If this objection undercuts our experiments, it also undercuts the merit of the original philosophical thought experiments offered to support Wurman’s argument.

A different way to elaborate this observation into an objection is to propose that (1) there is a more subjective relationship between stakes and clarity and (2) that (subjective sense) of clarity is relevant to legal interpretation. For a particular judge, that judge’s determination of clarity depends on the practical stakes to that particular judge. We do not have the space to fully engage with the merits of this theory, but some of its consequences are unusual. Because the subjective practical stakes of a decision may vary between judges, on this subjective view of stakes and clarity, such differences in subjective stakes would appropriately correspond to differing evaluations of clarity. A judge experiencing high practical stakes could deem a text unclear, while a judge experiencing lower practical stakes could deem the same text clear. However, many would think that whether a legal text is clear or unclear (in the sense relevant to legal interpretation) should not vary among judges in this way.¹⁸³See Richard M. Re, A Law Unto Oneself: Personal Positivism and Our Fragmented Judiciary, 110 Va. L. Rev. (forthcoming 2024) (manuscript at 5), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4687303 [https://perma.cc/E6DW-U4P2] (acknowledging that the dominant approach in jurisprudence is to “identify a ‘general’ theory of law that assertedly applies to most or all legal systems” but also arguing for the possibility and desirability of some degree of “personal” law implemented by judges). On this highly subjective view, to predict whether a law is correctly identified as “clear” (in the eyes of a particular judge), one must know what practical stakes the (particular) judge faces.

Although we find this an unusual view about what clarity means in current legal interpretation, this objection’s underlying claim is an empirically testable one. As such, we investigate this empirical question as a robustness check: Do stakes affect people’s judgment about whether the rule is clear to Bob?

2.  Objection 2: Only Parents’ Views About the Babysitter Hypothetical Count

A recurring objection to our study about the babysitter hypothetical concerns the population surveyed. Both Josh Blackman and Ilan Wurman have suggested that the appropriate audience for Justice Barrett’s hypothetical is parents.¹⁸⁴See Josh Blackman, Major Questions or Lax Parents?, Reason (July 27, 2023, 10:28 PM), https://reason.com/volokh/2023/07/27/major-questions-or-lax-parents [https://perma.cc/J74R-XGX7] (“Justice Barrett . . . may be referring to people who are familiar with the process of hiring babysitters. . . . Diversity of views is very important. One facet of diversity is having children.”); Wurman, supra note 8, at 961 n.271 (“It would be worth testing how many participants would agree that the instruction was followed if they were the parents.”). Because Justice Barrett’s hypothetical involves a parent, the objection goes, we should look to (only) the views of parents in understanding the meaning of the parent’s instruction.

Our original study did not collect data about participant’s parental status because we see it as irrelevant to the legal theory debate about the ordinary meaning of the parent’s instruction to the babysitter—more on that below. However even if we had that data in the first study, it is extremely unlikely that filtering by parental status would result in a different bottom line result given that the overall results lean so strongly in one direction.¹⁸⁵Consider a back-of-the-envelope calculation. Only 8% of participants responded that the babysitter who took the children to an amusement park violated the rule. Assume, favorably to the objectors, that these 8% of responders all had children. The majority of all parents’ responses would favor the babysitter hypothetical intuition if at least 92% (85/92) of the other participants did not have children. This would imply that, at most, 16% of all participants had children. Given that we recruited a sample of Americans, it is likely that much more than 16% had children. Only 8% of participants shared the intuition that the babysitter violated the parent’s instruction, and of the other 92% of participants, it is unlikely that that the vast majority (say 90%) did not have children. In the next Section, we present a new study that collects this additional demographic data. The results do not vary depending on parental or babysitter-hiring status.

Our more fundamental responses to this objection are theoretical and appeal to longstanding principles of interpretation. First, consider the question of audience: Should the babysitter hypothetical be limited to parents? That is, should textualist interpretation’s “ordinary reader” be limited to only a small subset of ordinary readers?

First, the objection assumes the wrong interpretive perspective. In Justice Barrett’s hypothetical, the parent is the lawgiver and the babysitter is the audience. But the correct textualist focus, according to Justice Barrett, is on interpretation from the “outside[],” not from the “inside[].”¹⁸⁶See Barrett, supra note 17, at 2194. Textualists typically view interpretation from the perspective of a “hypothetical reasonable person,” not from the perspective of the lawgiver. Fidelity to the text of the statute, as understood by an ordinary reader, is the best way to remain a faithful agent of Congress (or, as Justice Barrett would have it, as a faithful agent of the people). Thus, even if one specific focus in the babysitter hypothetical were deemed more appropriate, for modern textualists that focus would more likely be that of a babysitter (the instruction’s reader), not a parent (the instruction’s author).¹⁸⁷See Tara Leigh Grove, Testing Textualism’s “Ordinary Meaning”, 90 Geo. Wash. L. Rev. 1053, 1057 (2022).

Second, textualists do not subdivide the general class of ordinary people that determines ordinary meaning. Instead, the “Supreme Court tends to employ a one-size-fits-all approach to interpretation.”¹⁸⁸David Louk, The Audiences of Statutes, 105 Cornell L. Rev. 137, 193 (2019). We recognize the importance of interpretive communities and the observation that statutes have audiences.¹⁸⁹See generally id. (arguing that the varied audiences of statutes may have differing expectations about statutory meaning). However, the concept of audience is most often used in textualist theory and otherwise to support the use of technical meanings (instead of ordinary meanings) for certain specialized statutes. Otherwise, the same statute might mean different things to the different groups subject to it, a position that Justice Scalia (writing for the Court) condemned.¹⁹⁰Clark v. Martinez, 543 U.S. 371, 380 (2005) (rejecting the argument that the same statutory provision can have a different meaning depending on the group subject to it). The proposal to find ordinary meaning only in the views of the people most directly implicated by the law is thus a radical departure from modern textualism.

This suggestion (the legal interpretive equivalent of “ask only parents”) also strikes us as unworkable. If an interpreter aimed to limit “ordinary meaning” to the meaning a statute has to the people most directly impacted by it, how do we identify the people in this community? Even in Justice Barrett’s more straightforward babysitter hypothetical (and again setting aside that the babysitter is the audience, not the parent), we could ask: Are the relevant readers all parents, parents who go away for weekends, parents who can also afford babysitters, parents who would be willing to hand a credit card to a babysitter, or parents who would be willing to hand a credit card to a babysitter with limited instructions?

Even if the relevant subcommunity could be identified, it is not clear a judge would be well positioned to identify this narrow subcommunity’s understanding. If the textualist interpretive inquiry shifted from one about ordinary meaning to one about “ordinary meaning for only the audience most directly impacted by this statute,” might judicial intuition be especially unreliable if judges were not part of this latter subcommunity?

This suggestion becomes more bizarre as we shift from the babysitter hypothetical to real legal examples. In Biden v. Nebraska, who is the relevant interpretive community of people: the Department of Education, people with student loans, or some other group? If we take this objection and analogy seriously, it seems we should ask who is the “parent” in Biden v. Nebraska? Presumably, it is Congress. Do Blackman and Wurman suggest that Congress’s views are most relevant in interpretation? If so, this objection offered by Justice Barrett’s defenders, emphasizing a narrow subgroup of people who give or implement this instruction, is inconsistent with Justice Barrett’s broader approach to interpretation, which emphasizes judges as faithful agents of the “people,” not Congress.¹⁹¹See supra notes 186–87 and accompanying text.

In sum, the objection to “ask parents” about the babysitter hypothetical is not persuasive. Theoretically, the legal-interpretive analogue to “ask parents” is unmotivated, unworkable, and inconsistent with modern textualism. Nevertheless, we address this objection in the next part of this Section, in a replication study that asks for the participants’ parental status. Empirically, the results are no different for participants who are parents or who have hired a babysitter.

3.  An Additional Empirical Study

We are not persuaded by the theory underlying these two objections, but we are grateful to those who have raised them. And, setting the theoretical issues aside, it is possible to test these objections empirically. To do so, we conducted one final study.

Five hundred participants were recruited from Prolific to complete Study 1 and Study 2, with a few minor modifications aimed at addressing the objections described previously. A total of 445 participants passed the same attention checks described in Study 1 and Study 2 above and were included in the analysis (mean age = 37.9, 47% men, 51% women, 2% nonbinary). The final demographics section also asked about whether the participants had children (38% yes, 59% no, 2% prefer not to respond), had hired a babysitter (21% yes, 77% no, 2% prefer not to respond), and had worked as a babysitter (48% yes, 51% no, 1% prefer not to respond).

i.  Testing Clarity to Bob (the Agent Sensitive to Stakes)

Participants first read the Study 1 materials concerning Bob and the bank. They were again randomly assigned to high or low stakes and one of the four rule types (Obviously Clear, Ambiguous 1, Ambiguous 2, Obviously Unclear). Participants answered the same questions about knowledge and strict knowledge, as well as a new question that Wurman recommends about clarity to Bob:

[Clarity to Bob] Now imagine that Bob’s wife uses her phone to search for the bank’s policy. She finds a website for the local bank branch. The website’s text states [rule text varying by scenario].

Consider Bob’s perspective on this scenario.

Is this rule’s meaning clear or unclear to Bob concerning whether the bank is open?

Clear: The bank is open on Saturday.

Clear: The bank is not open on Saturday.

Unclear

A multinomial logistic regression examined the effect of Stakes (low, high) and Rule Type (Clear, Ambiguous 1, Ambiguous 2, Unclear) on judgment of the bank rule’s clarity to Bob (clearly open, clearly closed, unclear). First, consider the effect of Stakes. Comparing clearly open and clearly closed responses, there was no effect of Stakes, z = 0.95, p = 0.341. Comparing clearly closed and unclear responses, there was no effect of Stakes, z = 1.26, p = 0.209. There was no significant effect of Rule Type and no significant Stakes * Rule Type interactions.

The results for these questions about whether the rule is clear to Bob also show no effect of stakes. For the Obviously Clear rule, stakes did not affect judgments of clarity to Bob (2% of participants selected unclear in high stakes; 2% in low stakes); for the Ambiguous 1 rule, stakes did not affect judgments of clarity to Bob (29% of participants selected unclear in high stakes; 38% in low stakes); for the Ambiguous 2 rule, stakes did not affect judgments of clarity to Bob (4% of participants selected unclear in high stakes; 12% in low stakes); and for the Obviously Unclear rule, stakes did not affect judgments of clarity to Bob (61% of participants selected unclear in high stakes; 59% in low stakes).

The results for knowledge and strict knowledge were similar to the results found in Study 1. A binomial logistic regression revealed an effect of Stakes on knowledge. Participants attributed knowledge less in high-stakes cases (prob. = 0.85, 95% CI = [0.80, 0.89]) than in low-stakes cases (prob = 0.94, 95% CI = [0.89, 0.96]), odds ratio = 0.38, 95% CI = [0.19, 0.73], z = -2.89, p = 0.004. A binomial logistic regression revealed an effect of Stakes on strict knowledge. Participants attributed strict knowledge less in high-stakes cases (prob. = 0.64, 95% CI = [0.58, 0.70]) than in low-stakes cases (prob = 0.81, 95% CI = [0.75, 0.86]), odds ratio = 0.42, 95% CI = [0.27, 0.65], z = -3.91, p < 0.001.

In sum, one objection to our original Study 1 is that it fails to ask the right question about clarity: it should ask whether the text is clear to Bob, not clear in general. This follow-up study tested that question about clarity to Bob, finding identical results: participants’ judgments about clarity to Bob were not sensitive to high stakes.

ii.  Parents Only

The second objection is that we should consider only the views of parents. Consider the results of the same study, replicated, broken out by whether participants are parents, and have hired a babysitter.¹⁹²See infra Table 2.

Comparing across all participants, parent participants, and those who have hired babysitters, the results are essentially identical. Participants generally disagreed that the babysitter violated the rule/instruction by taking the children to an amusement park overnight, and this did not depend on whether those participants were themselves parents or had hired a babysitter.

4.  Additional Objections

We have responded to the two major objections leveled against the studies since we made a draft of this Article public. However, there are two other objections that strike us as worth pursuing, but which we do not have the space to fully explore here.

The first is that in our babysitter experiment, we should have asked a different question. As a reminder, we asked “which better describes this situation?”—that the babysitter “followed” the instruction/rule or “disobeyed” the instruction/rule? This strikes us as a straightforward way to capture textualists’ concern: What does the rule mean to the ordinary reader? Wurman suggests that we should have asked other questions, like whether participants agree that the instruction “include[s] authorization” to undertake this action, or whether participants think “ordinary, reasonable interpreters of this parent’s instruction would have interpreted it to include this scenario.”¹⁹³Wurman, supra note 8, at 961 n. 271. (“The question’s framing effectively required the participants to answer whether the babysitter literally violated the instruction. And the answer is of course not. But if the question had been asked another way—‘does the best reading of the parent’s instruction include authorization to undertake this action?’ or ‘do you think the parent’s instruction was intended to include this scenario?’ or ‘do you think ordinary, reasonable interpreters of the parent’s instruction would have interpreted it to include this scenario?’—the answer almost certainly would have been different.”). Wurman does not motivate these suggestions with much theory, and it is not obvious why these phrasings would identify participants’ understanding of the meaning of the rule. For example, recall that there are many theories of interpretation: textualism, purposivism, and consequentialism. It is not obvious that most people think the “reasonable interpreter” is a textualist. Perhaps people think that the “ordinary reasonable interpreter” is not a pragmatist. If so, asking people about their views of “the reasonable interpreter” would reliably generate non-textualist judgments.

Nevertheless, in our third study, we also asked these two additional questions: (1) “In your personal opinion, which better describes this situation?”—(a) The parent’s instruction/rule “authorized” the babysitter “to undertake this action”; or (b) The parent’s instruction/rule “did not authorize” the babysitter “to undertake this action”; and (2) “In your personal opinion, which better describes this situation?”—(a) “An ordinary person interpreting” the parent’s instruction/rule “would understand it to allow what” the babysitter did; or (b) “An ordinary person interpreting” the parent’s instruction/rule “would not understand it to allow what” the babysitter did. The results did not differ in the dramatic way that Wurman predicts. For the first authorization question, 85% of participants agreed that the “major” action was authorized. For the second “reasonable interpreter” question, a majority (57%) agreed that this reasonable interpreter would give the textualist response to the major action case: the reader would understand the instruction/rule to allow what the babysitter did.

A final objection states that the parent-babysitter analogy is a poor analogy for the Congress-agency relationship. This objection is sometimes offered as a critique of the MQD, not a defense, and as a reason why we should not indulge a faulty analogy. Less frequently, it is raised as a defense to the MQD—suggesting that the context of a real-world delegation would surely include consideration of constitutional structure.¹⁹⁴See generally Chad Squitieri, Placing Legal Context in Context (Oct. 23, 2023) (unpublished manuscript), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4610078 [https://perma.cc/G7CL-UZBV]. We find the basic point that context matters persuasive, but the muscular vision of ordinary understanding of legal context that Squitieri offers raises serious problems. It cannot be the case that the MQD is supported because an “ordinary” reader who studied the question of congressional delegation closely enough might become skeptical of the delegation of major power to agencies. First, questions of delegation are highly contested on many grounds—even trained lawyers and judges disagree vehemently about the legality and propriety of delegation. See Julian Davis Mortenson & Nicholas Bagley, Delegation at the Founding: A Response to the Critics, 122 Colum. L. Rev. 2323, 2323 (2022) (providing an overview of contemporary debates about the original public meaning of the Constitution’s vesting of legislative power in Congress). Second, on Squitieri’s approach, there must be some limit on the amount of legal context that can be assumed to be known by the “ordinary” reader. Otherwise, there would be nothing constraining judges in the elucidation of matters of ordinary meaning through their own trained, but highly subjective, minds. Ultimately, if the concept of the ordinary reader is to do any work within a textualist theory that constrains judges, it must provide some limits on the amount of legal context that can be assumed by the judge. We are exploring the idea that the babysitter hypothetical misses important relevant context to the evaluation of delegations in a future piece, but for now we set it aside. This Article takes the babysitter hypothetical from Biden v. Nebraska on its own terms. For readers that are skeptical that such an analogy provides insight into the meaning of statutory language, our study offers a second line of critique: the analogy does not offer insight into statutory meaning, and even if it did, its assumption about ordinary readers is faulty.

IV.  IMPLICATIONS

The recent pivot to a linguistic defense of the MQD is a watershed moment for two fields of law that often intersect: statutory interpretation and administrative law. Through the narrowest lens, the reframing of the MQD as “linguistic” attempts to insulate the nascent MQD from scrutiny as hypocritical anti-textualism, allowing conservative judges to use the doctrine to curb the power of the administrative state without turning in their textualist cards.¹⁹⁵See supra Section I.A. But the move also resonates much more deeply. If accepted, the connection being drawn between ordinary people and the MQD would move textualism further towards an “outsider” orientation, with implications well beyond the narrow purview of the MQD.¹⁹⁶See Barrett, supra note 17, at 2199; Tobia et al., Ordinary People, supra note 103, at 383. Likewise, if accepted, the linguistic defense of the MQD would tend to reinforce trends toward an explicitly “libertarian administrative law,”¹⁹⁷Cass R. Sunstein & Adrian Vermeule, Libertarian Administrative Law, 82 U. Chi. L. Rev. 393, 393 (2015). backing it with the force of supposedly ordinary people’s commonsense understanding of how government should work.

The theoretical critiques and original empirical evidence presented thus far in this Article support skepticism about the arguments to adopt the MQD as linguistic. In this Part, we explain why, and we also reflect on what our evidence says more generally about the fields of statutory interpretation and administrative law.

We start in Section IV.A by discussing how our investigation and findings challenge the conclusion that the MQD is a valid linguistic canon. In light of existing empirical work, our new empirical studies, and our new theoretical analysis and objections, we conclude that the two “linguistic defenses” of the MQD do not have adequate empirical support or theoretical clarity to succeed. Of course, defenders of the MQD might propose new arguments or different evidence, but for now, it is difficult to see on what basis one could employ the MQD as a valid linguistic canon.

Section IV.B explains that Justice Barrett and Wurman’s attempts to establish the MQD as a linguistic canon raise serious challenges to textualism. Justice Barrett’s arguments about “common sense” and “context” are so general that they threaten to undermine textualism’s commitment to enforcing the rule of law by privileging semantic content, even when unexpected applications are at issue. In turn, Wurman’s defense of the MQD necessarily involves a broad conception of “ambiguity.” This broad framing of ambiguity has been criticized by Justices Scalia and Kavanaugh and, like Justice Barrett’s arguments, would result in courts using “ambiguity” as a pretext to avoid the semantic meanings of statutes.

Finally, Section IV.C addresses broader implications for administrative law and regulation. We have reservations about any strategy to ground judicial interpretation in “ordinary people’s” understanding of ordinary examples, especially for a topic as technical as administrative law. Nevertheless, for the sake of argument, we consider where such an “ordinary” approach should take textualist interpreters. Empirical evidence about ordinary understanding of law and language suggests a dramatically different approach than what Justice Barrett suggests for the MQD. Ordinary people understand broad delegations to include a wide range of reasonable actions consistent with the delegation. Moreover, our findings reveal something we did not expect: ordinary people are fairly skeptical that underimplementation of delegated authority is consistent with facially broad delegations. These facts do not support the MQD, but they might support other linguistic canons—many of which have more in common with Chevron than the MQD—and they may counsel some rethinking of administrative law’s indifference to agency inaction.

A.  The Major Questions Doctrine Is Not a Valid Linguistic Canon

The most immediate question motivating our studies is whether there is a valid basis for considering the MQD as a linguistic canon of statutory interpretation. As discussed above, canons are traditionally distinguished according to whether they are justified by normative or legal principles (in which case they are substantive) or whether they help determine the linguistic meaning of statutory language (in which case they are linguistic).¹⁹⁸The conventional understanding of canons takes these options to be mutually exclusive: the MQD is either a linguistic canon, a substantive canon, or neither—but it cannot be both. See, e.g., Biden v. Nebraska, 143 S. Ct. 2355, 2376 (2023) (Barrett, J., concurring) (proposing that the MQD is a linguistic canon and noting skepticism about (all) substantive canons). Although a canon can be both substantive and linguistic,¹⁹⁹Slocum & Tobia, supra note 37, at 70 (arguing that a canon could have both a valid linguistic and substantive basis). the MQD’s defenders have emphasized the MQD’s supposed linguistic properties because of growing concerns among textualists about both substantive canons generally and the MQD in particular. The existing empirical evidence reviewed in Part II and original empirical studies in Part III suggest this is a false start: the linguistic properties identified by the MQD’s defenders do not find support in the intuitions (or “common sense”) of ordinary people. Consequently, at least in the absence of further empirical studies, the MQD cannot, and should not, be defended as a valid linguistic canon capturing how ordinary readers understand delegating statutes.

1.  The Evidence Does Not Support a “High-Stakes” Linguistic Major Questions Doctrine

i.  High Stakes and Knowledge

Start with the theory that the MQD is justified on the grounds that, for ordinary people, the stakes of an interpretive dispute impact the text’s clarity.²⁰⁰Wurman, supra note 8, at 957. This argument begins by appealing to analytic philosophy and legal theory that posits a relationship between stakes and knowledge claims.²⁰¹See, e.g., DeRose, supra note 24, at 914–15; Doerfler, supra note 25, at 523. The central example is the bank case: when little depends on the bank being open on Saturday, we know that it is open; but, when the stakes of the Saturday deposit are higher, we do not know that it is open.

However, a large empirical literature reports this claim to be false,²⁰²See supra Part II. and the entire philosophical literature to be “founded on a myth” about people’s reactions to these cases.²⁰³Schaffer & Knobe, supra note 34, at 675. Many studies find that high stakes have no effect at all on knowledge. Moreover, most of these studies use the exact case (the bank case) to which defenders of the linguistic MQD appeal.

The comparatively fewer studies that find an effect on knowledge report a small effect. In those studies, high stakes reduce knowledge for around 10% of participants, but not for the vast majority.²⁰⁴See, e.g., Rose et al., supra note 34, at 233. This Article’s new large empirical study (N = 500) finds a similarly small effect on knowledge, only a 9% difference between the low- and high-stakes cases.²⁰⁵See supra Section III.A.

Textualists are not always clear about how to construct their “ordinary reader,” but it is difficult to see how even this small difference (95% of people in low stakes agree there is knowledge, and 86% of people in high stakes agree there is knowledge) is sufficient to conclude that “the ordinary reader” has less knowledge in high-stakes contexts. For the vast majority of ordinary participants, high stakes have no impact on knowledge; the foundational premise in the “high-stakes” MQD seems to reflect an unordinary epistemology.

ii.  High Stakes and Clarity

The “high-stakes” argument for the linguistic MQD uses this (false) premise about knowledge as a theoretical foundation to support a technically distinct, and to date untested, claim that ordinary people follow the same epistemological pattern when making judgements about the clarity of statutory language. Assuming that people do this, the argument concludes that a high-stakes situation can render otherwise clear statutory language unclear.

The recent “high-stakes” legal interpretation literature seems to assume that statutory interpretation essentially involves a kind of knowledge claim, such that high stakes’ impact on knowledge necessarily carries over into the interpretive context.²⁰⁶See Wurman, supra note 8, at 957; Doerfler, supra note 25, at 523. Conceptually, we disagree with this literature’s equation of knowledge about a text’s meaning and textual clarity: language can be clear (in the relevant sense) even if laypeople do not have knowledge of its meaning. Consider books that report statements like: “The Art Nouveau movement preceded the Art Deco movement,” or “The Sarbanes-Oxley Act established the Public Company Accounting Oversight Board.” Even if a layperson does not have full knowledge about what these statements mean (that is, cannot accurately assess the statements’ truth or falsity, or explain what they mean to someone in reasonable detail), this does not imply that the statements are in any way unclear (in the sense of appearing ambiguous or indeterminate) to that layperson. Our data (the only we are aware of on this point) is not consistent with this transitive logic.²⁰⁷See supra Section III.B. We found that high stakes have a small effect on knowledge, but no effect at all on textual clarity. This finding supports the conclusion that ordinary judgments of knowledge do not rise and fall consistently with ordinary judgments of textual clarity.

More importantly, we find that high stakes have no effect on clarity for texts of varied levels of baseline ambiguity. High stakes did not reduce ordinary people’s sense of clarity for a fairly clear text or even for texts that were initially more ambiguous.²⁰⁸See supra Section III.B. This finding challenges the more critical premise in the “high-stakes” MQD defense (concerning clarity, not knowledge).

Together, these two problems count against the “high-stakes” linguistic defense of the MQD. High stakes have (at best) a small impact on knowledge and no impact on clarity. We have also noted various other theoretical issues with the “high-stakes” linguistic argument. For example, even if high stakes had the hypothesized effects, it is not clear why reduced knowledge or textual clarity puts more weight on judges’ readings of the statutes or implies anti-agency interpretation rather than putting more weight on agency interpretations of the statutes.²⁰⁹See Wurman, supra note 8, at 954–55 and accompanying text.

2.  The Evidence Does Not Support an “Anti-Literalist” Linguistic Major Questions Doctrine

i.  The Data Do Not Support the Stronger Claim Necessary to the “Anti-Literal” Linguistic Major Questions Doctrine

The previously discussed considerations about anti-literalism²¹⁰See supra Section II.B. are insufficient to support a strong conclusion about the MQD. Just because people sometimes interpret non-literally and display context sensitivity does not imply that courts should interpret general delegating language to authorize only a small subset of agency actions that fall under the text’s meaning. One could easily agree that (1) delegations should not always be interpreted literally, while also holding that (2) anti-literalism does not lead to the MQD.

In Section III.B, we reconstructed Justice Barrett’s argument in sufficient detail to deliver the MQD conclusion. We understood her key empirical claim to be the following: absent additional context, ordinary people understand rules that grant authority to an agent to have significant contextual limitations against all “major” actions; such a rule’s communicative content is limited to authorizing only the set of most reasonable actions. Here, an action is “major” if readers understand it, absent additional context, as not among the set of most reasonable ways to follow the rule. While this is a much stronger premise than mere anti-literalism, an even stronger premise is necessary to conclude that in MQD cases, absent additional context, judges should interpret delegations to exclude all major actions.

Our empirical study tested this claim about ordinary understanding of grants of authority.²¹¹See supra Section III.B. Here, we again sought to minimize researcher degrees of freedom and chose cases that have been offered by advocates of the linguistic MQD. In Study 2, we examined Justice Barrett’s “babysitter case.” We found that most ordinary people do not take the babysitter’s actions, that is, taking children on a multi-day trip to an amusement park, to be unauthorized by the parent’s instruction to use the parent’s credit card to ensure that the kids have fun over the weekend. To the contrary, 92% of respondents took the babysitter’s actions to be consistent with the rule/instruction. When we looked at a more extreme hypothetical—bringing a zookeeper to the house to entertain the kids with a live alligator—respondents judged the babysitter’s actions less reasonable but virtually just as authorized by the parent’s instruction to “make sure the kids have fun.”

However, our respondents did not simply think anything followed the rule. Fully 85% of them thought that the babysitter’s decision to use the credit card for something other than the children’s entertainment violated the instruction, and 49% believed that it was a violation of the instruction to entertain the children too little.

Importantly, these different actions varied in their perceived reasonableness. Participants agreed that it is more reasonable to respond to the parent’s instruction by buying the kids pizza, and less reasonable to take the kids to an amusement park or hire an animal entertainer. Nevertheless, participants judged that these latter actions—while not part of the most reasonable set of responses—are fully consistent with the rule.

Ultimately, these findings suggest that even if Justice Barrett is right that context matters for interpreting grants of authority to administrative agencies, that fact alone does not justify the strong MQD. To point to “common sense” and “context” may be entirely reasonable for a judge—we will have more to say about this in the next Section—but referring to them does not rule out “major” or less reasonable agency actions, at least in the minds of ordinary readers.

3.  Limits of the Evidence, and the Bottom Line

Our two studies test the central examples that have been offered by proponents of the MQD as a linguistic canon. Both of those arguments appeal centrally to claims about how ordinary readers understand language; neither of those claims is supported by the studies conducted here. Of course, this Article’s focus is on the linguistic arguments, not the many other defenses of the MQD.²¹²See, e.g., Randolph J. May & Andrew Magloughlin, NFIB v. OSHA: A Unified Separation of Powers Doctrine and Chevron’s No Show, 74 S.C. L. Rev. 265, 289–91 (2022) (discussing the MQD as a separation of powers principle); Nathan Richardson, Keeping Big Cases from Making Bad Law: The Resurgent “Major Questions” Doctrine, 49 Conn. L. Rev. 355, 359 (2016) (discussing the MQD as a safety valve for Chevron deference). And concerning the linguistic case, we are open to future arguments and empirical studies: some future revision of a linguistic defense of the MQD could possibly succeed. In this Section, we briefly highlight some of the limits of our studies and the doors they leave open for proponents of the MQD. We also summarize our “bottom line” about the MQD.

i.  Substantive Arguments for the Major Questions Doctrine

First, and perhaps most obviously, our studies do not foreclose a substantive basis for the MQD. That is, rather than grounding the doctrine in how text is understood, proponents of the MQD might point to constitutional or normative values that should lead judges to depart from the best reading of statutory language when agencies take major actions. The fact that none of the other Supreme Court justices joined Justice Barrett’s concurrence might suggest that at least five justices are comfortable with the idea that the MQD is solely substantive rather than partly or entirely linguistic.

So far, the Court has not clearly articulated the substantive basis of this canon: for Justice Gorsuch, the source of normative substance appears to be the nondelegation doctrine; for Chief Justice Roberts, the source is general separation of powers principles. But this lack of clarity about from where the justices are drawing the MQD’s substantive content does not mean that the MQD might eventually come, through an incremental process, to coalesce around some common narrative that would suffice to justify the MQD as a substantive canon alongside the many other substantive canons that our legal system recognizes. Given the growing textualist skepticism of substantive canons, as well as the contestable premises of the nondelegation doctrine and separation of powers, we doubt that such a defense would be uncontroversial,²¹³See Walters, supra note 12, at 521 (discussing the limits of the argument in favor of the MQD as simply another substantive canon). but this is a topic that falls outside the scope of this Article.

ii.  Linguistic but Non-Ordinary Arguments for the Major Questions Doctrine

Second, our studies focus on linguistic defenses that tie themselves explicitly to appeals to the construct of the “ordinary reader.” While we think this focus is defensible, given the larger textualist commitment to the ordinary reader as the anchor for interpretation,²¹⁴See Barrett, supra note 17, at 2194. it is also possible to defend a linguistic MQD on the grounds that it represents some kind of generalization about how Congress likely intends delegating statutes to be interpreted. The move here is to ground the MQD in what Beau Baumann calls the “descriptive case”: that is, an empirical assertion about the ordinary context of delegating statutes and the way Congress operates when it passes delegating statutes.²¹⁵Beau J. Baumann, The Major Questions Doctrine Fiction 11–12 (Mar. 14, 2023) (unpublished manuscript) (on file with authors).

Indeed, the Court in West Virginia v. EPA said as much when it cited a “practical understanding of legislative intent” as a basis for the MQD;²¹⁶West Virginia v. EPA, 142 S. Ct. 2587, 2609 (2022). both Wurman and Justice Barrett nod to this possibility as well.²¹⁷As Wurman, supra note 8, at 955–56 puts it:

Deliberate ambiguity benefits both parties when it comes to issues that are not sufficiently important as a general matter to scuttle an entire piece of legislation. But whether to tackle climate change through CO2 regulation, or to regulate cigarettes, or to allow a public health agency to prohibit evictions, are probably not the kinds of things legislators leave to strategic ambiguity; they are the kinds of things that one side wins and the other loses. On Wurman’s account, it makes sense as a linguistic matter to bake this contextual evidence of how Congress treats important questions into our reading of delegating statutes—that is, to interpret ambiguous statutes as not intended to delegate important matters. Justice Barrett’s concurrence in Biden v. Nebraska makes a similar move. After noting that all interpreters seek to “situate[] text in context,” Justice Barrett posits that “[b]ackground legal conventions . . . are part of the statute’s context.”²¹⁸Biden v. Nebraska, 143 S. Ct. 2355, 2378 (2023) (Barrett, J., concurring). In a principal-agent relationship, “ ‘the context in which the principal and agent interact,’ including their ‘prior dealings,’ industry ‘customs and usages,’ and the ‘nature of the principal’s business or the principal’s personal situation’ ” help form the background legal conventions that govern delegation.²¹⁹Id. at 2379. From there, Justice Barrett argues that we know from the context of how Congress usually delegates to agencies that Congress is “more likely to have focused upon, and answered, major questions, while leaving interstitial matters [for agencies] to answer themselves in the course of a statute’s daily administration.”²²⁰Id. at 2380.

These kinds of arguments based on the “descriptive case” run into persistent empirical problems—namely, there is ample evidence that Congress often does intend to delegate major questions to agencies through vague language, and only weak and contested evidence that Congress does not so intend.²²¹See, e.g., Blake Emerson, “Policy” in the Administrative Procedure Act: Implications for Delegation, Deference, and Democracy, 97 Chi.-Kent L. Rev. 113, 113 (2022); Alison Gocke, Chevron’s Next Chapter: A Fig Leaf for the Nondelegation Doctrine, 55 U.C. Davis L. Rev. 955, 970–71 (2021); Heinzerling, supra note 5, at 1933–34. Both Wurman and Barrett make much of a study of congressional staffers conducted by Abbe Gluck and Lisa Schultz Bressman that found that over 60% of staffers thought that drafters typically intend for Congress, not agencies, to decide important policy questions. See Wurman, supra note 8, at 951, 954–56 (citing Abbe R. Gluck & Lisa Schultz Bressman, Statutory Interpretation from the Inside—An Empirical Study of Congressional Drafting, Delegation, and the Canons: Part I, 65 Stan. L. Rev. 901, 1003–06 (2013)); Biden v. Nebraska, 143 S. Ct. at 2380 (same). However, the Gluck and Bressman study is at best weak support for the proposition that Congress intends to reserve major questions for itself. See Walters, supra note 12, at 533–34; Ronald M. Levin, The Major Questions Doctrine: Unfounded, Unbounded, and Confounded, 112 Calif. L. Rev. (forthcoming 2024) (manuscript at 145–47), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4304404 [https://perma.cc/W3X3-5GXM]. These kinds of arguments are also in significant tension with textualism, which generally eschews evidence of legislative intent except insofar as it is “objectified” in statutory language. However, given the evidence presented in this Article, these arguments may still be more promising for proponents of the MQD than a linguistic defense premised on ordinary meaning.

On the whole, then, it does not seem like the doors that are left open by our study are ones that would be attractive to the textualist justices who have given us the MQD. But we cannot deny another possibility: that textualism itself may evolve (or dissolve?) in ways that accommodate the MQD on these other grounds. We turn to that topic in the next Section, but before doing that, we would reiterate that the ordinary-meaning defense of the MQD is, by all appearances, a total dead end. Textualists would be hard-pressed to continue to defend the MQD on this theory of the case and this record of decision.

iii.  The Bottom Line

This Section has briefly noted some limitations of the Article. We make no claims about other (non-linguistic) defenses of the MQD. And we are, of course, open to the possibility that some future argument or evidence could rehabilitate the linguistic defense of the MQD.

However, it is important to emphasize that we endorse a firm conclusion about the current state of affairs for the linguistic MQD and textualists’ use of the canon. The two extant linguistic defenses of the MQD depend on empirical claims about specific hypotheticals (for example, the bank case) that are not supported by empirical studies of ordinary Americans. Proponents of the linguistic MQD may offer new, more workable arguments, with different thought experiments, or different empirical support. But until then, there is no basis to employ it as a linguistic canon, and there is now significant evidence counting against core claims of the two publicly stated linguistic arguments.

Second, even for textualist judges with no interest in the linguistic defense, the empirical data about ordinary readers counts against the MQD’s consistency with ordinary language. Given ordinary readers’ understanding of language, there is more evidence in favor of treating the MQD as an anti-linguistic canon than a linguistic canon.²²²For example, it appears false that people intuitively understand delegations to be limited to the most reasonable set of actions consistent with the language’s literal meaning. With further empirical study, one could imagine refining a canon that captures ordinary judgment about delegation. Most plausible candidates are at odds with the MQD. We discuss this idea further in Section IV.C, infra. And as Justice Kagan remarked, judges who appeal to such non-principles over linguistic interpretation are not really textualists.²²³West Virginia v. EPA, 142 S. Ct. 2587, 2641 (2022) (Kagan, J., dissenting).

B.  Broader Implications for Modern Textualism

Justice Barrett and Wurman’s arguments have implications for textualism beyond the narrow (but hugely important) issue of whether the MQD is a linguistic canon. Textualism’s claim to distinctiveness centers on a commitment to interpretation according to a text’s linguistic meaning, thereby promoting rule of law values.²²⁴See William N. Eskridge, Jr., Brian G. Slocum & Kevin Tobia, Textualism’s Defining Moment, 123 Colum. L. Rev. 1611, 1613 (2023) (explaining how textualism claims to promote the rule of law). Textualism thus abjures judicial discretion to depart from that linguistic meaning.²²⁵Tara Leigh Grove, Comment, Which Textualism?, 134 Harv. L. Rev. 265, 269 (2020) (advocating for formalistic textualism). As Justice Scalia emphasized, judges should not exercise an unbounded “personal discretion to do justice.”²²⁶See Antonin Scalia, The Rule of Law as a Law of Rules, 56 U. Chi. L. Rev. 1175, 1176 (1989). Instead, judges should be restrained even when some results may have been unanticipated by the legislature.²²⁷See John F. Manning, The Absurdity Doctrine, 116 Harv. L. Rev. 2387, 2392–95 (2003).

Justice Barrett’s expansive view of “context,” “common sense,” and non-literal interpretation expands, but also challenges, these foundations of textualism. Justice Barrett admirably argues for a sophisticated version of textualism that rejects literalism and recognizes implied terms.²²⁸See supra Section I.C; see also Kevin Tobia, Brian G. Slocum & Victoria Nourse, Progressive Textualism, 110 Geo. L.J. 1437, 1475 (2022) (arguing that textualism should more willingly acknowledge that linguistic meaning can often include implied terms). Even so, existing interpretive canons that recognize implied terms are narrow, and thus do not undermine textualism’s commitment to linguistic meaning.²²⁹See supra Section I.C. In contrast, Justice Barrett’s “common sense” interpretive canon is unbounded, granting judges considerable discretion to claim that a wide range of actions fall outside of the text’s meaning (or “reasonable meaning”).

Wurman’s arguments also have implications that threaten to expand, if not unravel, textualism. Recall that Wurman, unlike Justice Barrett, frames the MQD as a tiebreaker canon that resolves statutory ambiguity.²³⁰See supra Section I.B. Wurman is correct that the Court has referenced “ambiguity” in MQD cases. This framing of the MQD, however, requires a broad view of ambiguity that would make its determination even more discretionary, and likely more pretextual.

1.  Justice Barrett’s Theory of Non-Literal Interpretation

Justice Barrett’s general appeals to context and non-literal interpretation are consistent with modern textualist scholarship and thinking. Justice Kavanaugh has also repeatedly emphasized the distinction between literal and ordinary meaning and has insisted that courts should avoid overly literalist meanings.²³¹See Bostock v. Clayton County, 140 S. Ct. 1731, 1825 (2020) (Kavanaugh, J., dissenting) (“[C]ourts must follow ordinary meaning, not literal meaning. And courts must adhere to the ordinary meaning of phrases, not just the meaning of the words in a phrase.”). Similarly, John Manning argues that “the literal or dictionary definitions of words will often fail to account for settled nuances or background conventions that qualify the literal meaning of language and, in particular, of legal language.”²³²Manning, supra note 227, at 2393.

Textualism, though, purports to privilege semantic meaning, thereby giving a relatively limited role to non-literal meanings informed by context and pragmatics. Thus, while Manning endorses some non-literal interpretation, his “background conventions” are narrow ones relevant to the “relevant linguistic community” subject to the law, such as common law criminal defenses.²³³Id. at 2466–67. Besides these limited examples, according to Manning, judges “have a duty to enforce clearly worded statutes as written, even if there is reason to believe that the text may not perfectly capture the background aims or purposes that inspired their enactment.”²³⁴See John F. Manning, Second-Generation Textualism, 98 Calif. L. Rev. 1287, 1290 (2010). Doing so ensures “Congress’s ability to use semantic meaning to express and record its agreed-upon outcomes.”²³⁵Id.

A coherent textualism would thus recognize a narrow role for implied terms. Crucially, an implied term must be one that would be obvious to the discourse participants, rather than one imposed by the interpreter for other reasons. An implied term must therefore reflect a presupposition about meaning that is warranted in the circumstances.²³⁶See Emiel Krahmer, Presupposition and Anaphora 3 (1998); Alan Cruse, A Glossary of Semantics and Pragmatics 139 (2006) (explaining that presuppositions are a ubiquitous aspect of language).

Statutes are often drafted at a high level of generality, and Justice Barrett is correct that readers of those rules understand that sometimes the rules expressed are not meant to be taken literally in all respects. Crucially though, the relevant existing interpretive canons are implicated in narrow circumstances and provide relatively specific rules for limiting literal meaning.²³⁷See Tobia et al., From the Outside, supra note 18, at 281–87 (providing examples of textual canons that narrow literal meaning); Slocum & Tobia, supra note 37, at 75 (providing examples of substantive canons that are also linguistic and which serve to narrow literal meaning). Furthermore, empirical evidence supports these narrow rules as linguistic and thus consistent with how ordinary people interpret legal texts.²³⁸See Slocum & Tobia, supra note 37, at 75.

Justice Barrett’s view of implied terms as governed by “common sense” and “context” is similar to Richard Fallon’s approach. Fallon argues that “[o]rdinary principles of conversational interpretation call for us to ascribe a reasonable meaning to prescriptions and other utterances unless something about the context indicates otherwise.”²³⁹Richard H. Fallon, Jr., The Meaning of Legal “Meaning” and Its Implications for Theories of Legal Interpretation, 82 U. Chi. L. Rev. 1235, 1260–61 (2015). Fallon reasons that “[i]n ordinary conversation, we do not waste time and breath offering elaborations and qualifications of our utterances that ought to be obvious to any reasonable person.”²⁴⁰Id. at 1261. Instead, a “reasonable person” understands that “[t]he moral reasonableness of a particular ascribed meaning possesses a distinctive importance.”²⁴¹Id. at 1261–62. Both Fallon and Justice Barrett draw on principles of conversational communication and context, and while Fallon references “reasonable meaning” and Justice Barrett “common sense,” the two are essentially the same idea. In fact, Justice Barrett uses the word “reasonable” in relation to interpretation eleven times in her Biden v. Nebraska opinion (e.g., “reasonable understanding,” “reasonable view,” “reasonable interpreter”).²⁴²Biden v. Nebraska, 143 S. Ct. 2355, 2376–84 (2023) (Barrett, J., concurring). Furthermore, her appeal to “common sense” and “reasonable” interpretations has, like Fallon’s view, room for moral and normative beliefs to motivate non-literal interpretations.

The similarities between the interpretive approaches of Justice Barrett and Fallon should be surprising and troubling to textualists. Fallon’s interpretive principle is in furtherance of his decidedly anti-textualist view of interpretation.²⁴³See generally Richard H. Fallon, Jr., The Statutory Interpretation Muddle, 114 Nw. U. L. Rev. 269 (2019) (arguing against the idea that statutes have determinate linguistic meanings). Justice Barrett’s principle of “common sense,” guided by “context,” is supposedly in furtherance of textualism, but it raises questions that do not have easy textualist answers. Can the principle always defeat the literal meaning of a statute? How can “common sense” even be defined? Even if “common sense” could be defined, do judges share the same “common sense” as ordinary people, or do judges speak with what Eskridge and Nourse refer to as an “upper-class accent?”²⁴⁴See William N. Eskridge, Jr. & Victoria F. Nourse, Textual Gerrymandering: The Eclipse of Republican Government in an Era of Statutory Populism, 96 N.Y.U. L. Rev. 1718, 1811 (2021). The ability for judges to appeal, with little restraint, to “common sense” and “context,” calls to mind Scalia’s fears about non-textualist judging: “personal discretion to do justice” as the judges saw fit.²⁴⁵Scalia, supra note 226, at 1176.

2.  The Anti-Textualist Broad View of Ambiguity

An additional threat to textualism is posed by a broad view of “ambiguity.” Recall that Wurman argues that the MQD is a linguistic canon that resolves statutory ambiguity.²⁴⁶See supra Section I.B. In support of this claim, Wurman quotes from MQD decisions where the Court argues that the relevant statutes are “ambiguous.”²⁴⁷See Wurman, supra note 8, at 915. This defense of the MQD is unsurprising. Textualism is much more permissive about available arguments and interpretive sources when a provision has been deemed “ambiguous.”

There are two key drawbacks in viewing the MQD as serving a tiebreaking role in resolving ambiguity. First, doing so understates the MQD’s role in the Court’s precedents. The MQD has not merely resolved “ties” between meanings; it has caused the Court to choose meanings it would not otherwise have selected. Second, Wurman’s view requires a definition of ambiguity that should be especially troubling to textualists, and the significance of the issue extends beyond the MQD.

Wurman’s argument raises an essential question: On what basis can a provision be deemed “ambiguous”? Wurman suggests that a provision can be “ambiguous” even when a court can determine the provision’s “best reading.”²⁴⁸Id. Thus, crucially, the question of ambiguity does not require that a provision be indeterminate. In other words, the semantic meaning of the provision’s terms could be clear (even if broad) but still “ambiguous,” based on non-textual considerations like the novelty and importance of an agency’s actions.

Use of the “ambiguity” label often obscures rather than clarifies linguistic issues. Specifically, it glosses over the distinctive linguistic features of the prototypical statute involved in MQD cases, which is a statute with broad but semantically clear terms. These features—broad but semantically clear—should represent for textualists a prima facie case against the MQD. After all, textualists assert that courts should focus on the semantic meaning of statutes.

Outside of MQD cases, some textualists have recognized the potential dangers associated with a judicial focus on “ambiguity.” Most significantly, Justice Kavanaugh has criticized “ambiguity” as an interpretive doctrine because its identification is standardless and subjective.²⁴⁹See Brett M. Kavanaugh, Fixing Statutory Interpretation, 129 Harv. L. Rev. 2118, 2121 (2016) (reviewing Robert A. Katzmann, Judging Statutes (2014)). Its discretionary identification and legitimizing power, however, make “ambiguity” an especially attractive interpretive tool for judges. “Ambiguity” is extremely useful because it gives a court cover to interpret a statute narrowly or broadly on the basis of normative concerns. For instance, an explicit announcement of ambiguity allowed the Court in King v. Burwell to “avoid the type of calamitous result that Congress plainly meant to avoid” and gave it justification for “interpret[ing] the Act in a way that” improves health insurance markets and does not destroy them.²⁵⁰King v. Burwell, 576 U.S. 473, 498 (2015).

“Ambiguity’s” legitimizing power explains why the Court in MQD (and other) cases is motivated to label a provision as “ambiguous” without much consideration about whether it is applying a coherent definition of ambiguity. It may be activist to interpret a clear statute narrowly because doing so would be in tension with the provision’s linguistic meaning. In contrast, resolving statutory “ambiguity” is necessary to decide the interpretive dispute, and choosing the narrower interpretation does not conflict with the provision’s linguistic meaning. Thus, if a provision is problematically broad, labeling it as “ambiguous” does not require the Court to explicitly reject its literal meaning.

If a provision can be “ambiguous” even when a court can nevertheless determine its “best reading,” “ambiguity” would mean something like “any uncertainty about the meaning of a provision.” But this sort of definition would make ambiguity ubiquitous and is inconsistent with how it is used in Chevron and other tiebreaker canons like the rule of lenity.²⁵¹See Eskridge, Slocum & Tobia, supra note 224, at 1656 (discussing how the Court’s textualists determine ambiguity). If instead “ambiguity” means that a provision must actually be indeterminate, there is no “best reading” of a provision, but merely possible competing meanings.

The question of ambiguity thus hinges on whether “ambiguity” is synonymous with “indeterminacy.” Even if the terms are synonymous, framing the MQD in terms of “ambiguity” should be unappealing to textualists. The MQD would still be a matter of judgment that depends on how one weighs semantic and pragmatic evidence. In other words, a combination of meaning and context makes a provision clear or, conversely, ambiguous. Univocal semantics and univocal pragmatics may uncontroversially result in a clear provision, and multivocal semantics and multivocal pragmatics in an ambiguous provision, but other combinations are contestable and subject to normative resolution via highly discretionary judgments.

The choice is thus between a narrow definition of “ambiguity” that would require the semantic meaning of the statutory text be indeterminate in some way, and a broad definition that would allow even semantically clear language to be deemed “ambiguous” based on non-language concerns like statutory purpose. Justice Scalia argued that the broad view of ambiguity is “judge-empowering” and mocked the idea that “[w]hatever has improbably broad, deeply serious, and apparently unnecessary consequences . . . is ambiguous!”²⁵²Bond v. United States, 572 U.S. 844, 870 (2014) (Scalia, J., concurring in the judgement). A broad definition of ambiguity would allow the label to be used at any time by emphasizing any number of pragmatic considerations, such as the problematically broad semantic meaning of terms or the “novelty” of an agency’s interpretation. If instead, as Justice Scalia argues, pragmatic evidence can only clarify semantically indeterminate text, ambiguity would therefore require indeterminate semantic meaning and be a narrower, less discretionary doctrine.²⁵³See id.

Textualists in MQD cases should be honest about their use of “ambiguity.” If they use the term broadly, they should explain why Justice Scalia’s critique of the broad definition is mistaken. If they instead agree with Justice Scalia, the MQD cases involving clear (but broad) semantic meaning should thus be viewed by textualists as similar to situations not involving ambiguity. In such cases, if the Court wishes to narrow the literal meaning of the language, it should state so explicitly, giving reasons for why such narrowing is consistent with the judicial function.

C.  Broader Implications for Administrative Law

This Article has taken textualists’ defenses of the MQD at face value. But some harbor a more realist or critical take on the MQD. Perhaps the MQD is animated by neither constitutional values nor language, but rather by the aim of limiting the administrative state’s power. And perhaps leaving questions about the MQD’s legitimacy unresolved allows strategic ambiguity, which is better for this purpose.²⁵⁴See Sohoni, supra note 6, at 266; see also Patrick J. Sobkowski, Of Major Questions and Nondelegation, Yale J. on Reg.: Notice & Comment (July 3, 2023), https://www.yalejreg.com/nc/of-major-questions-and-nondelegation-by-patrick-j-sobkowski [https://perma.cc/23GL-D2G6] (noting that the MQD is currently marked by “strategic ambiguity” that “allows the Justices to strike down or uphold policies without being criticized by other actors for judicial activism and aggrandizement”). Some go even further to argue that the justices are engaged in a form of constitutional hardball, seeking to aggrandize themselves vis-à-vis the other branches of government.²⁵⁵See, e.g., Josh Chafetz, The New Judicial Power Grab, 67 St. Louis U. L.J. 635, 635 (2023); Allen C. Sumrall & Beau J. Baumann, Clarifying Judicial Aggrandizement, 172 U. Pa. L. Rev. Online 24, 24 (2023); Mark A. Lemley, The Imperial Supreme Court, 136 Harv. L. Rev. F. 97, 97 (2022). It is certainly difficult to overlook the hostility that many of the justices express toward modern administrative government and the legislative acts that authorized it.²⁵⁶See generally Beau J. Baumann, Americana Administrative Law, 111 Geo. L.J. 465 (2023) (discussing examples including National Federation of Independent Businesses v. Department of Labor, Occupational Safety and Health Administration, 142 S. Ct. 661, 669 (Gorsuch, J., concurring); Gundy v. United States, 139 S. Ct. 2116, 2134–35 (2019) (Gorsuch, J., dissenting); and City of Arlington v. FCC, 569 U.S. 290, 327 (2013) (Roberts, C.J., dissenting)).

Yet, turning our attention away from the Supreme Court and toward the broader legal community, our findings about how ordinary people understand delegations of authority have significant implications for administrative law well beyond the MQD. While we acknowledge that there are good reasons to be skeptical about outsourcing questions of administrative law to laypeople, insofar as textualist principles animate the statutory interpretation questions at the heart of administrative law, it is worth asking where ordinary people’s intuitions lead.²⁵⁷Indeed, an emerging literature does just this, often using survey experiments to investigate questions important to administrative law and the administrative state. See generally Brian D. Feinstein, Legitimizing Agencies, 91 U. Chi. L. Rev. 919 (2024); Edward Stiglitz, The Reasoning State (2022). Below, we highlight a couple takeaways from this exercise. An irony of textualist’s turn to “ordinary people” to support the MQD may be that it actually supports a significantly cabined judicial role in controlling delegation of authority to the administrative state. Far from endorsing a kind of “libertarian administrative law” that treats delegations of authority to administrative agencies with suspicion and seeks almost perfunctorily to narrow them,²⁵⁸See Sunstein & Vermeule, supra note 197, at 410. ordinary people appear to take general ordinary delegations to license a range of reasonable actions.

To be sure, we considered ordinary judgments of an ordinary, private delegation (that is, the babysitter), but critics of the administrative state have made that ordinary context relevant by insisting that general principles of private agency and/or ordinary delegations law should inform public law delegation.²⁵⁹Biden v. Nebraska, 143 S. Ct. 2355, 2379–80 (2023) (Barrett, J., concurring); Phillip Hamburger, Is Administrative Law Unlawful? 386 (2014); Gary Lawson & Guy Seidman, “A Great Power of Attorney”: Understanding the Fiduciary Constitution 104 (2017). We are also skeptical that there is an easy way to study the “ordinary person’s” view of specific cases. As prior research has shown, interpreters’ values affect their interpretation.²⁶⁰Ward Farnsworth, Dustin F. Guzior & Anup Malani, Ambiguity About Ambiguity: An Empirical Inquiry into Legal Interpretation, 2 J. Legal Analysis 257, 259 (2010). Asking ordinary people whether the EPA has authority to issue broad climate change regulations under the Clean Air Act is likely to tell us more about people’s values and politics than their understanding of language. Thus, the implications we spell out depend on the validity of this ordinary analogy—the one made by the linguistic MQD’s defenders (recall the “high stakes” appeal to the ordinary bank case and the “common sense” appeal to the ordinary babysitter case).

To start, our study of the babysitter hypothetical revealed a surprising result about what ordinary people would think of the amusement park hypothetical. Taking the children to the amusement park might not be the most reasonable response to the instruction to “use this credit card to make sure the kids have fun this weekend,” but it certainly does not violate it (after all, an amusement park is “fun”). The study also revealed that the vast majority of ordinary people believe that the parent’s instruction extends to the even more unusual action of bringing a live alligator to the house. This surprising finding suggests that people do not limit delegations to only the most reasonable actions or the ones most consistent with the rule’s purpose.

Ordinary readers approached the limits of broad delegations through a textual and purposive lens. Compared with the amusement park, alligator, and movie scenarios, respondents were far more likely to say that the babysitter violated the instruction when the babysitter failed to achieve the purpose of the instruction (as in the case of not using the credit card and potentially shortchanging the children’s fun) and when the babysitter actively undermined it (by using the credit card for the babysitter’s own enjoyment). This finding is difficult to understand unless ordinary readers understand delegations in large part as remedial—that is, as seeking to empower the agent to solve a problem or achieve some goal—rather than exclusively delimiting—that is, as setting out the scope of the agent’s power.²⁶¹This explanation is largely consistent with Brian Feinstein’s discovery that ordinary people are prompted to increase their trust in government when they believe it is being undertaken by an agent with expertise to fulfill social functions. See Feinstein, supra note 257, at 919. In both Feinstein’s studies and ours, delegations are understood by ordinary people to be about problem solving.

The modern textualist commitment to ordinary people’s understanding as a basis for interpretation²⁶²See, e.g., Barrett, supra note 17, at 2194. and linguistic canons²⁶³See, e.g., Wurman, supra note 8, at 909. opens the door to uncovering a linguistic basis for other canons, including new canons.²⁶⁴See Tobia et al., From the Outside, supra note 18, at 288–90. As a hypothetical, imagine if a textualist were to carefully consider evidence about ordinary people’s understanding of delegating language (e.g., in the babysitter case) and attempt to “canonize” those intuitions into administrative law doctrine. The result would probably be a fundamental recalibration of the field—but not in the way the MQD imagines. Were one to follow the evidence, it seems to instead support canonizing a sort of “counter-MQD” that presumes that general delegations should be interpreted broadly (or at least not as restrictively as Justice Barrett’s argument claims), significantly curtailing judicial power to limit Congress’s attempts to empower administrative agencies.

In addition, and relatedly, our findings are in some tension with administrative law’s traditional approach to questions of underimplementation of statutory delegations. A variety of administrative law doctrines insulate agency discretion to decline to enforce the law: for instance, Heckler v. Chaney provides that agency nonenforcement decisions are almost never reviewable by courts,²⁶⁵Heckler v. Chaney, 470 U.S. 821, 821 (1985). and Norton v. Southern Utah Wilderness Alliance makes it impossible for challengers to force agency action unless they can point to a discrete duty (rather than a more general failure to pursue broad policy goals of a statute).²⁶⁶Norton v. S. Utah Wilderness All., 542 U.S. 55, 55 (2004). These doctrines insulate agency underuse of delegated regulatory authority from judicial scrutiny. Yet our findings suggest that ordinary readers may be more troubled by delegated authority’s underuse than uses that fit with the language but exceed an observer’s sense of reasonableness.²⁶⁷Both using the credit card for only the babysitter’s needs (“misuse”) and bringing an alligator to the house for entertainment (“extreme”) were judged as “unreasonable,” while failing to use the card and entertaining the children with card games (“minor”) was judged as “reasonable.” But rule violation judgments did not rise and fall with these evaluations of reasonableness. The extreme action was more consistent with the rule than the minor action, and both were more consistent than the misuse action. On the flip side, when agencies do take action pursuant to their delegations, judges often artificially narrow those delegations.²⁶⁸Cass R. Sunstein, Nondelegation Canons, 67 U. Chi. L. Rev. 315, 315–16 (2000). Canons that might theoretically push in the opposite direction—toward liberally construing “remedial” statutes, for instance—have fallen into disrepute.²⁶⁹Dir., Off. of Workers’ Comp. Programs, Dep’t of Lab. v. Newport News Shipbuilding & Dry Dock Co., 514 U.S. 122, 135 (1995) (noting that the remedial canon is the “last redoubt of losing causes”). This basic asymmetry in the treatment of delegations to agencies—deep skepticism of exercises of delegated authority coupled with indifference toward failures to exercise delegated authority at all²⁷⁰See Daniel E. Walters, Symmetry’s Mandate: Constraining the Politicization of American Administrative Law, 119 Mich. L. Rev. 455, 455–56 (2020).—may be exactly backwards if ordinary people’s intuitions are to be the guide.

Again, we do not endorse any particular changes to administrative law here. There are many good reasons, such as the institutional constraints under which agencies operate, to disfavor outsourcing administrative law into ordinary people’s linguistic or legal intuitions (whatever those may be).²⁷¹Eric Biber, The Importance of Resource Allocation in Administrative Law, 60 Admin. L. Rev. 1, 1–2 (2008). There are also many countervailing concerns, such as fair notice and due process, that may justify curtailing expansive ordinary readings of delegating statutes.²⁷²See, e.g., Jennifer Lee Koh, Crimmigration and the Void for Vagueness Doctrine, 2016 Wis. L. Rev. 1127, 1153–59 (2016). But we also believe that for those inclined to remake administrative law through the eyes of the ordinary reader, it is worth grappling with facts rather than judicial hypotheticals about those ordinary readers. People are far more comfortable with broader interpretation of general-language delegations than many textualists have assumed, and they appear to be disproportionately uncomfortable with violations through underuse of delegated authority.

CONCLUSION

The MQD is the most influential interpretive development at the modern Supreme Court.²⁷³See supra notes 2–4 and accompanying text. Yet it lacks a compelling theoretical basis and a satisfactory explanation of its consistency with textualism, the interpretive theory held by the MQD’s advocates. The new “linguistic MQD” purports to solve both problems: because the MQD reflects ordinary understanding of language, it is a valid linguistic canon and thus consistent with textualism.

This Article has taken this linguistic defense on its own terms and studied the two central ordinary examples offered by its advocates. We find that ordinary people do not understand language as textualists have assumed. High stakes do not undermine knowledge or impact textual clarity, and people do not understand general delegations to be limited to only the most reasonable set of actions. These results challenge the essential empirical claims at the heart of the arguments for the linguistic MQD. While scholarly debate should continue, judges must take stock of the evidence and decide whether to employ the canon—and whether to do so in the name of linguistics and ordinary people. In our view, there is insufficient empirical support and theoretical clarity to cast the MQD as a valid linguistic canon. Arguably, the linguistic defense is the only viable theory for textualists to consistently employ the MQD. Thus, unless they offer a successful alternative, the results here support the broader conclusion that consistent textualists should not employ the MQD.