Auer deference holds that reviewing courts should defer to agen­cies when the latter interpret their own preexisting regulations. This doc­trine relieves pressure on agencies to undergo costly notice-and-com­ment rulemaking each time interpretation of existing regulations is neces­sary. But according to some leading scholars and jurists, the doc­trine actually encourages agencies to promulgate vague rules in the first instance, augmenting agency power and violating core separation of pow­ers norms in the process. The claim that Auer perversely encourages agencies to “self-delegate”—that is, to create vague rules that can later be informally interpreted by agencies with latitude due to judicial defer­ence—has helped to persuade the Supreme Court to take up this term the question of whether to overturn the doctrine. Yet, surprisingly, the self-delega­tion thesis has never been tested.

This Article scrutinizes the thesis empirically, using an original and extensive dataset of the texts of federal rules from 1982 to 2016. My lin­guistic analysis reveals that agencies did not measurably increase the vagueness of their writing in response to Auer. If anything, rule writ­ing arguably became more specific over time, at least by one measure, despite Auer’s increasing prominence.

These findings run against common wisdom, but they should not be at all surprising. The self-delegation thesis depends on a model of agency behavior that is inconsistent with what is known about the insti­tutional pressures and cognitive horizons that cause agencies to pur­sue clarity in rule writing. By revealing the failures of theoretical predictions about Auer, this Article more generally draws attention to the need to test behavioral theories of administrative law against empiri­cal reality before unsettling settled law.


    1. Early Origins—The Road from Seminole Rock to Auer
    2. Recent Developments—Auer’s Second Revolution
    1. The Critique of Auer—Incentives for Self-Delegation
    2. The Defense of Auer
    3. The Need for Evidence in Assessing the Self-Delegation Critique
    1. Data
      1. Sample
      2. Measures of Vagueness
    2. Aggregate Trends
    3. Agency-Specific Trends
    4. Searching for a Link Between Incentives and Effect in Auer
    5. Testing the Effect of the Second Revolution
    6. Other Robustness Checks
      1. Alternate Treatment
      2. Sample Bias
    7. Summary of Empirical Findings
    1. The Limits of Comprehensive Rationality in Administrative Agencies
    2. Bounded Rationality’s Impact on the Timing of Interpretive Clarity in Administrative Action
      1. Goal Ambiguity
      2. Tactical Uncertainty
      3. Professional Norms and Mission Focus
    3. Satisficing Rule Writers and the Shadow of State Farm
      1. Hard Look Review Basics
      2. Hard Look’s Effect on Agency Behavior
      3. Hard Look’s Secondary Effects on Regulatory Precision



A core concern of administrative law is constraining the discretion of agencies, given that they often operate under broad delegations of author­ity in statutes that contain vague standards and aspirations. As Professor Kenneth Culp Davis described it, administrative law primarily ought to encourage agencies to exercise their “rule-making power to replace vague­ness with clarity.” 1 Kenneth Culp Davis, Discretionary Justice: A Preliminary Inquiry 56–57 (1971). Much of  the development of  American admin­istra­tive law in recent  decades has aimed to  promote and fine-tune notice-and-comment rulemaking under the Administrative Procedure Act (APA) as a means of reducing discretion in the administrative  state and  making law  more certain. 2 See E. Donald Elliott, Re-Inventing Rulemaking, 41 Duke L.J. 1490, 1491–92 (1992) (observing that “most American  academic students of administrative law have been overly enamored of the formal beauty of the notice-and-comment  process” and have “reiterated  unanimously over the years that  agencies are  free to choose between rulemaking and other forms of agency action for making policy”); Reuel E. Schiller, Rulemaking’s Promise:  Administrative Law and Legal Culture in the 1960s and 1970s, 53 Admin. L. Rev. 1139, 1150 (2001) (noting that a shift from ad hoc adjudication to rulemaking  and policy statements was “the most common recommendation of all  the critics who examined the administrative state at the end of the 1950s and beginning of the 1960s”); Richard B. Stewart, The Reformation of American Administrative Law, 88 Harv. L. Rev. 1667, 1698 (1975) (suggesting that a “possible response to the problems created by broad legislative delegation is to acknowledge the large discretion enjoyed by agencies and to require that it be exercised in accordance with consistently applied general rules”).

Today, leading scholars and jurists view a core administrative law doc­trine known as Auer deference as an existential threat to this project and even an affront to fundamental constitutional separation of powers norms. Auer deference (also sometimes referred to as Seminole Rock defer­ence 3 The alternate moniker refers to a 1945 case, Bowles v. Seminole Rock & Sand Co., that initially used the “plainly erroneous or inconsistent” formulation in response to an agency’s “administrative construction” of its regulation. 325 U.S. 410, 414 (1945). A wave of recent scholarship suggests that the strong version of deference that exists today is of relatively recent vintage, mostly attributable to the Supreme Court’s unanimous restate­ment of the principle in Auer v. Robbins, 519 U.S. 452, 461 (1997). See infra section I.A. ) holds that a court reviewing an agency’s interpretation of its own regulations should defer to the agency’s construction so long as it is not “plainly erroneous or inconsistent with the regulation.” 4 Auer, 519 U.S. at 461 (internal quotation marks omitted) (quoting Robertson v. Methow Valley Citizens Council, 490 U.S. 332, 359 (1989)). This principle crucially allows agencies to avoid the impractical (and potentially debilitat­ing) need to make every slight adjustment in regulatory understand­ing of existing rules through cumbersome formal amend­ment of written rules. 5 See infra note 108 and accompanying text. Technically, as long as there is a gap in the meaning of an existing regulatory text, an agency can interpret that gap in nonbinding fashion without resorting to notice-and-comment rulemaking whether it receives deference or not. See 5 U.S.C. § 553(b)(A) (2012) (providing exemptions from the notice-and-comment rulemaking requirement for nonbinding interpretive rules and general statements of policy). But the promise of deference frees agencies to make that decision with less concern about being haled into court and subjected to probing review. With Auer deference available, agencies can instead issue guidance documents or policy statements that merely inter­pret existing rules in ways that achieve regulatory goals. However, the argument against Auer deference posits that if agencies know that they will win most cases involving their interpretations of previously promul­gated rules, they have strong incentives to write vague rules in the first instance. 6 See Talk Am., Inc. v. Mich. Bell Tel. Co., 564 U.S. 50, 68–69 (2011) (Scalia, J., concurring) (arguing that Auer “encourages the agency to enact vague rules which give it the power, in future adjudications, to do what it pleases”). That way, it is claimed, agencies will be able at a later time to package a more significant change as a regulatory interpretation and need only make a plausible argument that this new interpretation is loosely contemplated by the original rule’s capacious language. 7 See, e.g., Jennifer Nou, Regulatory Textualism, 65 Duke L.J. 81, 102 (2015) (“[A]fter an agency promulgates a legislative rule through notice and comment, it can then continuously revise its interpretations without meaningful notice to regulated entities and with little judicial accountability.”); Matthew C. Stephenson & Miri Pogoriler, Seminole Rock’s Domain, 79 Geo. Wash. L. Rev. 1449, 1453 (2011) (“[B]road judicial deference to an agency’s interpretation of its own regulations may enable an agency to enact binding rules without subjecting itself either to meaningful procedural safeguards or to rigorous judicial scrutiny.”). With such a strategy, agencies could systematically “self-delegate” by writing rules that subsequently allow them to circumvent the APA’s call for notice-and-comment rulemaking, effectively affording themselves the opportunity to make significant policy decisions merely by issuing guid­ance. 8 Kevin M. Stack, Preambles as Guidance, 84 Geo. Wash. L. Rev. 1252, 1254 (2016) (“[C]ritics contend that agencies rely on guidance documents in ways that circumvent the notice-and-comment rulemaking process. Their concern is that agencies are turning increasingly to guidance to establish norms that have significant de facto weight without the participation and accountability virtues of a notice-and-comment process.”). This general turn might feed a perceived trend away from the quasi-constitutional norms of the APA and its preference for traditional forms of agency policymaking toward less formal (and less accountable) forms of policymaking, such as guidance documents and policy statements. See Daniel A. Farber & Anne Joseph O’Connell, The Lost World of Administrative Law, 92 Tex. L. Rev. 1137, 1140 (2014) (“Our thesis is simple but powerful: the actual workings of the administrative state have increasingly diverged from the assumptions animating the APA and classic judicial decisions that followed.”); Mila Sohoni, The Administrative Constitution in Exile, 57 Wm. & Mary L. Rev. 923, 964 (2016) (“The aspects of administrative law that enable procedurally unfettered regulation have a far larger footprint now than they formerly did; they have attained a new level of importance in the allocation of power across various parts of the government.”).

Consider Title IX of the Civil Rights Act, which prohibits discrimina­tion “on the basis of sex” in public schools. 9 20 U.S.C. § 1681(a) (2012) (“No person in the United States shall, on the basis of sex, be excluded from participation in, be denied the benefits of, or be subjected to discrimination under any education program or activity receiving Federal financial assistance . . . .”). In 2000, under this statutory provision, the U.S. Department of Education promulgated a rule that allowed school districts to install sex-separate restrooms in public school buildings but required these separate facilities to be comparable. 10 34 C.F.R. § 106.33 (2018) (providing that schools “may provide separate toilet, locker room, and shower facilities on the basis of sex, but such facilities provided for students of one sex shall be comparable to such facilities provided for students of the other sex”). Six­teen years later, in response to questions schools were confronting over how to accommodate transgender students within this regulatory frame­work, the Department issued an opinion letter that further interpreted its own regulation to require that “[w]hen a school elects to separate or treat students differently on the basis of sex,” it must “treat transgender students consistent with their gender identity.” 11 G.G. ex rel. Grimm v. Gloucester Cty. Sch. Bd., 822 F.3d 709, 715 (4th Cir. 2016) (internal quotation marks omitted) (quoting James A. Ferg-Cadima, U.S. Dep’t of Educ., Office for Civil Rights, Opinion Letter (Jan. 7, 2015) (on file with the Columbia Law Review)). The Fourth Circuit upheld the Department’s policy, giving Auer deference to the agency’s interpretation contained in the opinion letter in part because the origi­nal regulation simply did not speak to how transgender students should be accommodated. 12 Id. at 722–23.

Suppose that the Department of Education had actually been clearer in its 2000 rule, explicitly stating how it expected transgender students to be treated. Suppose further that the Department in 2016 wanted to adopt a different stance with respect to the treatment of transgender stu­dents and issued an opinion letter to that effect. In a response to a properly framed challenge to the opinion letter, the Department could argue that Auer deference should apply to its new interpretation con­tained in the letter; however, there would be limits to how far the clear language in the original rule could be stretched before the subsequent interpretation would become “plainly inconsistent” with it. By speaking clearly and explicitly about transgender students in its original rule, the Department would have substantially limited its own flexibility down the road and very well may have been forced to pursue any changed view in 2016 by going through the full notice-and-comment rulemaking process.

Return now to what actually happened. Whether the Department actu­ally intended it or not at the time, the fact that it left out any men­tion of transgender students in the original rule in 2000 gave the Department of Education interpretive flexibility in 2016 when it decided to make clear that schools must treat transgender students in a manner consistent with their gender identity. Auer deference all but guaranteed that the exercise of this flexibility would not be substantially curtailed by reviewing courts.

If agencies see similar scenarios often enough, they may come to real­ize that there are potential benefits to making their rules in the first place more amenable to capacious interpretation down the road. For instance, perhaps on reflection the Department would realize that it could have even more discretion in the Title IX domain were it to write something like “schools should not discriminate on the basis of sex except when it is reasonable to fulfill the purposes of Title IX.” Such a woefully vague (even perhaps nonsensical) construction would only give more flexibility to the Department at a later time to pivot in and out of various specific approaches to protecting (or not protecting) transgender students.

Owing to concern about expansive agency discretion, critics of the Auer doctrine—including current Harvard Law School Dean John Manning 13 See infra notes 112–113 and accompanying text. and several current and former Supreme Court Justices 14 See infra section I.B. —claim that the doctrine creates perverse incentives for agencies to make their regulations unclear (or at least not worry too much if they are), know­ing that the courts will back agencies up if they later decide to issue informal opinion letters or other guidance documents interpreting those regulations as they please. 15 See infra section II.A. In other words, the critics’ claim, as Professors Cass Sunstein and Adrian Vermeule have characterized it, is that “agencies will issue vague, broad regulations, knowing full well that when the time comes, they will be able to impose the interpretation they prefer.” 16 Cass R. Sunstein & Adrian Vermeule, The Unbearable Rightness of Auer, 84 U. Chi. L. Rev. 297, 308 (2017) [hereinafter Sunstein & Vermeule, Unbearable Rightness]; see also Cynthia Barmore, Auer in Action: Deference After Talk America, 76 Ohio St. L.J. 813, 818 (2015) (“If Auer requires deference to an agency’s interpretation of ambiguous (but not unambiguous) regulations, agencies would maximize future flexibility and power by promulgating ambiguous regulations.”); John Yoo & James C. Phillips, With Kavanaugh, the Court Could Tame the Administrative State, Nat’l Rev. (Oct. 25, 2018), https:// [] (arguing that Auer “motivates agencies to maximize their power by promulgating ambiguous regulations”).

This argument, which might be called the “self-delegation” or “per­verse incentives” thesis, has powerfully captured the attention of judges and scholars alike because it cuts straight through normatively contested debates about whether any deference to administrative action is desirable as a policy matter. The critics posit that, at least when it comes to defer­ence to agencies’ interpretations of their own regulations, judicial defer­ence inexorably leads to agencies expanding the scope of their own discre­tion through a kind of regulatory sandbagging. Auer, its critics claim, perversely encourages agencies to promulgate indeterminate regula­tions with the intention that they can be molded through interpre­tive moves made at a later time without the need to go through the notice-and-comment process.

Of course, the possibility that agencies could interpret their regula­tions without going through notice-and-comment procedures is some­what consistent with the APA’s core compromises. 17 See Conor Clarke, The Uneasy Case Against Auer and Seminole Rock, 33 Yale L. & Pol’y Rev. 175, 179 (2014) (describing the APA’s, and the Court’s, “preference” for an administrative law that affords agencies great flexibility in how they proceed). Agencies thus have, under current law, vast discretion to choose from a wide range of policymaking forms, only some of which require extensive formal public engagement and purely prospective application, and they use these authorities liberally. See generally Robert A. Anthony, Interpretive Rules, Policy Statements, Guidances, Manuals, and the Like—Should Federal Agencies Use Them to Bind the Public?, 41 Duke L.J. 1311, 1313 (1992) (explaining that agencies can “make a substantive nonlegislative rulemaking document binding on private parties” through the issuance of an interpretive rule); Nina A. Mendelson, Regulatory Beneficiaries and Informal Agency Policymaking, 92 Cornell L. Rev. 397, 398 (2007) (noting that the volume of agencies’ general statements of policy and interpretive rules, as compared to legislative rules promulgated through notice-and-comment procedures, is “massive”); Todd D. Rakoff, The Choice Between Formal and Informal Modes of Administrative Regulation, 52 Admin. L. Rev. 159, 160 (2000) (discussing the “growing tendency of American administrative agencies to return to what we, in the United States, think of as informal administrative procedures”); Peter L. Strauss, Publication Rules in the Rulemaking Spectrum: Assuring Proper Respect for an Essential Element, 53 Admin. L. Rev. 803, 804 (2001) (“Publication rules—interpretative rules, statements of general policy, staff manuals, and the like—are an important element in the hierarchy of agency law.”). Reasonable people might well disagree about whether Auer strikes precisely the right balance between flexibility and fair notice when agencies seek to flesh out rules on the back end of the administrative process. 18 See Clarke, supra note 17, at 191 (noting that the question with regard to the propriety of deference doctrines is “always relative: what’s the best way to strike the pragmatic balance”). But it is much harder to disagree that a doctrine that causes agencies literally to augment their own authority—and purposefully to withhold their understanding of the law from those who bear its burdens—is flawed, both pragmatically and constitutionally. That is presumably why the self-delegation thesis has had such sway in the Supreme Court and among legal practitioners and why the Court now appears to be on the verge of overturning or signifi­cantly limiting Auer deference. 19 In December 2018, the Supreme Court granted a petition for writ of certiorari in Kisor v. Wilkie, No. 18-15, 2018 WL 6439837, at *1 (U.S. Dec. 10, 2018), a case involving a regulatory interpretation by the Department of Veterans Affairs and squarely presenting the question of whether Auer and Seminole Rock should be overruled. See Amy Howe, One New Grant, One Dissent from the Denial of Review, SCOTUSblog (Dec. 10, 2018), []. The grant, along with the Court’s decision to limit the review to the question of whether Auer and Seminole Rock should be overturned—the petitioner had presented the alternative question of whether Auer deference should be replaced with a canon of regulatory construction, Petition for a Writ of Certiorari at i, Kisor v. O’Rourke, No. 18-15 (U.S. June 29, 2018), 2018 WL 3239696—has led many Court watchers to predict that the Court will use the case as a vehicle to overturn the doctrine. See, e.g., Ellen M. Gilmer, Deference Rule May Be on the Chopping Block, E&E News (Dec. 10, 2018), []. The critical vote will likely come from Justice Kavanaugh, who predicted years ago that the Court would “someday” overrule Auer. Jimmy Hoover, ‘Auer Someday Will Be Overruled,’ Kavanaugh Said in 2016, Law360 (Dec. 17, 2018), (on file with the Columbia Law Review) (internal quotation marks omitted) (noting, however, that then-Judge Kavanaugh “was careful to point out in his speech that ‘these are not necessarily my wants, nor my fears, just my predictions’”).

All of this raises the question: Does the self-delegation thesis have any basis in reality? Recognizing how important evidence is to the future of the debate over Auer, this Article scrutinizes the self-delegation thesis from an empirical perspective. I test the thesis by analyzing the linguistic clarity of an original and extensive dataset of federal rules from 1982 to 2016. 20 This range of years was selected based on data availability. The White House Office of Information and Regulatory Affairs’ list of economically significant rules—the list from which the sample was generated—extended from 1981 to the present. It appeared that the records from 1981 were incomplete, so that year was excluded. Likewise, 2017 was excluded because data were available only for part of the year. The 1982 to 2016 range also makes for a roughly symmetric set of years before and after the 1997 Auer decision. See infra section III.A.1. My analysis reveals that agencies did not measurably increase the vagueness of their rules in response to Auer. If anything, rule writing became more specific over time despite Auer’s increasing prominence. 21 See infra sections III.B–.E. The findings persist across a range of measurement strategies and model­ing specifications.

These results are striking when considered in light of the promi­nence and influence of the criticisms that have been leveled at Auer in recent years. Yet, given what social scientists know about how agency offi­cials make decisions in the real world, the lack of empirical evidence of an “Auer effect” on rule writing should not be as surprising as it might be to Auer’s critics. 22 See infra sections IV.A–.B. The self-delegation thesis is based on a simplified model of the choices officials must make in determining the timing of clarifica­tion of their meaning. Regulators issue rules to address certain problems before them, and as a result they always face functional pressures in the short term for making rules clear. After all, it will presumably be difficult to ensure compliance if regulated entities cannot understand what a rule means, and changes in administration (and the political uncertainty that follows) likewise often favor locking in policy in ways that might bind poten­tially hostile future overseers. Moreover, Auer’s critics appear to over­look entirely how other core features of administrative law reinforce agencies’ short-term focus. The most immediate legal concern that agency rule writers face is the threat of vacatur under “hard look” review under the APA’s arbitrary and capricious clause. 23 See infra section IV.C. Hard look review, through its imperative that agencies thoroughly justify a chosen policy over its alternatives, creates incentives for clarity in rules—precisely the opposite of what Auer allegedly encourages. 24 See infra section IV.C.

These and other factors are abstracted away in the simplified model that implicitly stands behind the self-delegation thesis and that appears to assume that Auer is the only legal factor affecting agencies’ decisions. More generally, and overlaying all these factors favoring clarity, agencies have limited bandwidth to address all the considerations that might affect tradeoffs between providing clarity now or later. As with individuals more generally, agency officials can be expected to discount long-term benefits and overvalue short-term gains. 25 See infra section IV.A. Auer’s critics’ claims about the doctrine’s perverse incentives run counter to these well-documented fea­tures of agency behavior and human cognition that lead agencies to focus more on immediate outcomes at the expense of longer-term strate­gic gambits. 26 See infra section IV.B.

In short, recognizing the full range of institutional and cognitive fac­tors likely at play makes plain why no one should be at all surprised that the self-delegation thesis finds no support when subjected to careful empirical inquiry.

Beyond the obvious implication that the ongoing assault on Auer defer­ence needs revisiting, the faulty assumptions of the self-delegation thesis point more generally to the importance of an evidence-based under­standing of how administrative law doctrine and institutional struc­ture shape agency behavior. Claims about the behavioral effects of adminis­trative law are both pervasive and fundamental to the field, but they are too infrequently addressed empirically by administrative law scholar­ship. 27 See, e.g., Cary Coglianese, The Rhetoric and Reality of Regulatory Reform, 25 Yale J. on Reg. 85, 95 (2008) [hereinafter Coglianese, Rhetoric and Reality] (describing a “longstanding disjunction between the rhetoric and reality of regulatory reform” with regard to reform’s effects on “delay and paralysis” in administrative agencies). One contribution of this Article, then, is to show that, in at least one important context, a facially plausible intuition about behav­ioral effects has led us astray. The broader implication, though, is that this error could be pervasive in separation of powers jurisprudence and scholarship more generally, as it is much easier to generate abstract claims about agency incentives than it is to sustain a reasonable burden of proof through empirical analysis. 28 For a recent example of a court being persuaded by an argument that agency structure violated separation of powers principles, see PHH Corp. v. Consumer Fin. Prot. Bureau, 839 F.3d 1, 6–8 (D.C. Cir. 2016) (holding that the structure of the Consumer Financial Protection Bureau is unconstitutional in part because it is “an independent agency headed by a single Director and not by a multi-member commission,” which concentrates power in the one Director), rev’d en banc, 881 F.3d 75 (D.C. Cir. 2018). In short, a failure to check whether theory accords with reality can create a grossly distorted understanding of the effects of legal doctrine on administrative power.

Moreover, reckoning with administrative law’s linkages with agency behav­ior also bears on perennial debates about the role of administrative agencies within the American constitutional framework. 29 See infra notes 387–392 and accompanying text. Formalist separa­tion of powers theory resists many aspects of the administrative state because, as the argument goes, the administrative state often involves the breakdown of hard lines demarcating the constitutionally prescribed allocation of powers, which in turn creates dangerous combina­tions of powers that misalign incentives in a manner inconsistent with republican government. 30 See infra notes 383–386 and accompanying text. The criticism leveled against Auer is hardly the only manifestation of these larger concerns about vesting vari­ous combinations of legislative, executive, and judicial powers in the agen­cies that make up the administrative state. Indeed, claims about these kinds of combinations—and the opportunities they create for the “arrogation of power” 31 Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 620 (2013) (Scalia, J., concurring in part and dissenting in part). —are ubiquitous in contemporary criticisms of the administra­tive state. 32 See Gary Lawson, The Rise and Rise of the Administrative State, 107 Harv. L. Rev. 1231, 1248 (1994) (noting that “[a]dministrative agencies routinely combine all three governmental functions in the same body, and even in the same people within that body”); id. at 1248–49 (describing the Federal Trade Commission’s enforcement cycle as including the power to “promulgate[ ] substantive rules of conduct,” to “consider[ ] whether to authorize investigations,” to conduct investigations, to file administrative complaints, and to both prosecute and adjudicate the violation). But if agencies are often pulled in different directions by compet­ing doctrinal incentives and the inherent complex­ities of their task—not to mention that they might be institutionally and cognitively predisposed to err in predictable ways—then they are unlikely to invariably seek to maximize their discretion and power, and the formalist insist­ence on prophylactic divisions of powers to prevent agency tyranny loses much of its persuasive force.

The Article proceeds as follows. Part I introduces Auer deference and describes how it evolved from an earlier principle into a central fea­ture of litigation in the federal courts in the 1990s. This Part also details how Auer deference has more recently come under assault by some in the judiciary and the academy. Part II then presents the critique of Auer in greater detail, showing how the doctrine’s critics claim that it creates per­verse incentives for rule writing by relaxing the strict separation of rulemak­ing and rule-interpreting authority. That Part argues that the debate over Auer has reached an impasse because neither proponents nor critics have treated this claim as what it is: a testable empirical hypothe­sis.

Part III is the core of the Article. It provides an empirical linguistic analy­sis of agency rule writing, showing statistically that Auer’s incentives do not translate into any measurable changes in agencies’ propensity to produce vague rules. I used computer-assisted text analysis methods, some drawn from existing work in computational linguistics and psychol­ogy, to measure different facets of textual vagueness for some 19 million words of regulatory text from 1982 to 2016. This effort yielded nuanced, variegated, and comprehensive data on agency rule-writing behavior that I then used to test the self-delegation thesis. On these measures, Auer shows no sign of having undermined clarity in rule writing.

Part IV then shows that the results in Part III find firm support in a broader social science literature on individual and organizational deci­sionmaking. If it were not for the prominence and influence of the adher­ents of the self-delegation thesis, the results in Part III likely would not come as any news. After all, an extensive body of social science research points to cognitive and institutional tendencies that lead administra­tive agencies to give primacy to providing clarity and specificity in the here and now and to ignore any future purported benefits to the agency that might come from crafting vague rules. These tendencies rein­force and amplify the innumerable factors operating to induce clar­ity in rule writing, including the immediate possibility of judicial review under the arbitrary and capricious standard, which administrative law scholars widely accept as a threat that weighs heavily on agency officials’ minds in rulemaking.

Finally, the Article concludes with a discussion of the implications that this analysis has for the ongoing assault on Auer and for larger debates about the legitimacy of bureaucracy in a separation of powers framework. Before changing the law in response to formalist concerns about incentives, scholars and judges should insist that critics meet their burden of demonstrating actual perverse effects.

I. Auer’s Origins and Ascension

Auer stands for the principle that an agency’s interpretation of its own regulations is “controlling unless ‘plainly erroneous or inconsistent with the regulation.’” 33 Auer v. Robbins, 519 U.S. 452, 461 (1997) (quoting Robertson v. Methow Valley Citizens Council, 490 U.S. 332, 359 (1989)). In recent years, the Court has delineated several circumstances when Auer does not apply. See Christopher v. SmithKline Beecham Corp., 567 U.S. 142, 155 (2012) (holding that Auer is not to be afforded “when there is reason to suspect that the agency’s interpretation ‘does not reflect the agency’s fair and considered judgment’” (quoting Auer, 519 U.S. at 462)); Gonzales v. Oregon, 546 U.S. 243, 257 (2006) (holding that Auer does not apply when the rule’s text “parrots,” or mimics, the legislative ambiguity that justified Chevron deference in the first place—something widely known as the “anti-parroting” principle or canon). Together, these caveats have prompted some to consider whether there is an emerging “step zero” in Auer doctrine. See Clarke, supra note 17, at 189 (“But many of these doctrinal innovations do to Auer what United States v. Mead does to Chevron: they limit the ‘domain’ of deference by adding what is often described as a ‘step zero.’”). When courts apply Auer, they defer to an agency’s exercise of interpretive discretion. Reasonable interpretations are cred­ited, even if the court might actually believe that another possible interpreta­tion is a better one. 34 Decker, 568 U.S. at 613 (“It is well established that an agency’s interpretation need not be the only possible reading of a regulation—or even the best one—to prevail.”). Auer is thus sometimes compared to its more familiar doctrinal cousin, step two of Chevron deference, 35 Chevron U.S.A. Inc. v. Nat. Res. Def. Council, Inc., 467 U.S. 837, 842–43 (1984) (“[T]he court does not simply impose its own construction on the statute . . . . [I]f the statute is silent or ambiguous with respect to a specific issue, the question for the court is whether the agency’s answer is based on a permissible construction of the statute.”); see also Jeffrey A. Pojanowski, Revisiting Seminole Rock, 16 Geo. J.L. & Pub. Pol’y 87, 92 (2018) (noting that Auer is often viewed as “a kind of Baby Chevron doctrine”). and in some ways the comparison is apt—all the more so now that several empiri­cal studies have shown that the government’s win rate in Auer cases is consistent with that in Chevron cases. 36 Barmore, supra note 16, at 815 (“[C]ourts of appeals have responded to the Court’s recent Auer decisions by narrowing their application of the doctrine, leading to a steady decline from 2011 to 2014 in the rate at which courts grant Auer deference.”); Richard J. Pierce, Jr. & Joshua Weiss, An Empirical Study of Judicial Review of Agency Interpretations of Agency Rules, 63 Admin. L. Rev. 515, 519 (2011) (finding a 76% valida­tion rate for a set of cases decided by lower courts); William Yeatman, Note, An Empirical Defense of Auer Step Zero, 106 Geo. L.J. 515, 547 (2018) (finding that, from 1993 to 2013, the government won 74.4% of cases when a court invoked Auer and 68.6% when a court invoked Chevron). These findings show a noticeable drop in actual deference compared with the Supreme Court’s actual deference in the few Auer cases it has decided. According to a leading empirical study, the Supreme Court applied Auer in 1.1% of agency inter­pretation cases and deferred to the agency 90.9% of the time it applied Auer. William N. Eskridge, Jr. & Lauren E. Baer, The Continuum of Deference: Supreme Court Treatment of Agency Statutory Interpretations from Chevron to Hamdan, 96 Geo. L.J. 1083, 1099 tbl.1 (2008). Thus, although Auer’s “controlling ‘unless plainly erroneous or inconsistent with the regulation’” language might have suggested a kind of “super deference,” see Daniel Mensher, Seminole Rock in Environmental Law: A Window into Weirdness, Yale J. on Reg.: Notice & Comment (Sept. 15, 2016), [] (“[W]hat I find most perplexing about Auer is that it demands courts defer to nearly any agency interpretation of its regulations, regardless of where, when, or how the agency offers that interpretation. This leads to some bizarre results.”), Auer is in practice within the normal range of the continuum of deference. Auer deference, like Chevron defer­ence, has been justified on a number of grounds, including that interpretive deference broadly accords with notions of agencies as the most expert and most accountable body to implement statutory pro­grams. 37 See Stephenson & Pogoriler, supra note 7, at 1460 (describing a host of “prag­matic arguments” for Auer, including “expertise, efficiency, flexibility, and accounta­bility”); see also Clarke, supra note 17, at 179–81 (reviewing the evolution of justifications of Auer).

The difference between Chevron and Auer—and the impetus for the bulk of the skepticism of Auer deference—lies in the source of the interpre­tive discretion. With Chevron, the agency’s discretion, if it exists, comes from Congress’s statutory delegation of authority, express or implied. 38 See United States v. Mead Corp., 533 U.S. 218, 229 (2001) (noting that, even if Congress has not “expressly delegated authority” to an agency, “it can still be apparent from the agency’s generally conferred authority and other statutory circumstances that Congress would expect the agency to be able to speak with the force of law”). For a helpful explication of Chevron, see Cary Coglianese, Chevron’s Interstitial Steps, 85 Geo. Wash. L. Rev. 1339, 1347–51 (2017) (discussing in detail the ways that Congress can delegate inter­pretive questions to administrative agencies). With Auer, the discretion comes from lingering ambiguities in the agency’s own previously promulgated regulatory text. 39 Auer v. Robbins, 519 U.S. 452, 461 (1997) (applying Seminole Rock rather than Chevron because the “salary-basis” test at issue was “a creature of the Secretary’s own regulations”). When Auer comes up in litigation, it is most often because an agency has issued a guidance, policy statement, advisory letter, or manual—that is, a nonlegisla­tive rule—that clarifies a legislative rule 40 The difference between legislative rules and nonlegislative rules is that legislative rules “are designed to have binding legal effect on both the issuing agency and the public” and are therefore required to “undergo the expensive and time-consuming process known as notice-and-comment rulemaking before being promulgated.” David L. Franklin, Legislative Rules, Nonlegislative Rules, and the Perils of the Shortcut, 120 Yale L.J. 276, 278 (2010) (citing 5 U.S.C. § 553(b)–(c)(2012)). “Nonlegislative rules, by contrast, are not meant to have binding legal effect, and are exempted from notice and comment by the APA as either ‘interpretative rules’ or ‘general statements of policy.’” Id. (citing 5 U.S.C. § 553(b)(A)). previously promul­gated through notice-and-comment rulemaking under section 553 of the APA. 41 Of course, if the agency adds something too new, changing the substance of the rule in a way that binds the public, it risks a procedural challenge under the APA alleging that it should have gone through notice-and-comment rulemaking. See, e.g., Texas v. United States, 86 F. Supp. 3d 591, 677 (S.D. Tex. 2015) (enjoining the Department of Homeland Security from implementing the Deferred Action for Parents of Americans and Lawful Permanent Residents (DAPA) program following a challenge from twenty-six U.S. states that claimed DAPA had not been promulgated in compliance with the APA’s notice-and-comment procedures), aff’d, 809 F.3d 134 (5th Cir. 2015), aff’d mem. by an equally divided court, 136 S. Ct. 2271 (2016) (per curiam); Franklin, supra note 40, at 287–89 (discussing the various tests that courts employ in determining whether agency action is a “legislative rule” that should have gone through notice-and-comment procedures). Likewise, in certain cases an agency’s new interpretation in an enforcement context can raise issues of fair notice that will prevent post hoc application of the interpretation. See, e.g., Gen. Elec. Co. v. EPA, 53 F.3d 1324, 1329 (D.C. Cir. 1995) (outlining the “ascertainable certainty” requirement for fair notice in the enforcement realm (quoting Diamond Roofing Co. v. Occupational Safety & Health Review Comm’n, 528 F.2d 645, 649 (5th Cir. 1976))). Usually, the agency could have made an identical change by amend­ing the initial rule through a new round of notice-and-comment rulemaking, but issuing a guidance or other nonbinding document can be accomplished more quickly because the strictures of section 553 do not apply. 42 William Funk, Saving Auer, Jotwell (June 23, 2016) (reviewing Sunstein & Vermeule, Unbearable Rightness, supra note 16), [] (arguing that “it will be infinitely faster and easier for the agency to use [interpretive rulemaking techniques]” because it would “avoid the requirements for notice-and-comment rulemaking under the APA”). In effect, Auer blesses some procedural corner-cutting in the name of administrative efficiency. 43 See infra notes 104–111 and accompanying text for discussion of the advantages of policymaking by nonlegislative rule.

Auer is not as widely cited as Chevron, 44 See Thomas J. Miles & Cass R. Sunstein, Do Judges Make Regulatory Policy? An Empirical Investigation of Chevron, 73 U. Chi. L. Rev. 823, 824 n.2 (2006) (reporting that Chevron was cited a staggering 2,414 times in its first decade, 2,584 times in its next six years, and 2,235 times in the next five years). but it has become an estab­lished feature of litigation involving federal administrative agencies. 45 See Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 616 (2013) (Roberts, C.J., concurring) (“Questions of Seminole Rock and Auer deference arise as a matter of course on a regular basis.”). It was not always so. The first section of this Part chronicles the rise of Auer deference from its humble origins in Bowles v. Seminole Rock & Sand Co. 46 325 U.S. 410 (1945). to its unques­tioned endorsement by the Supreme Court in the late 1990s. The next section then turns to more recent years, which have seen both an increasing acceptance of Auer deference across the federal judiciary as well as some growing apprehension about the doctrine among some cur­rent and former Supreme Court Justices.

A. Early Origins—The Road from Seminole Rock to Auer

Although the Supreme Court decided Auer in 1997, the deference prin­ciple it represents actually stems from an unassuming case from the 1940s involving an interpretation issued by the Office of Price Administration (OPA) of its General Maximum Price Regulation under the Emergency Price Control Act. 47 Id. at 411–14. In Seminole Rock, the Court con­fronted an ambiguity in this regulation, which provided that “each seller shall charge no more than the prices which he charged during the selected base period of March 1 to 31, 1942.” 48 Id. at 413. Seminole Rock & Sand Company wanted to charge customers for crushed rock at a rate that it had formally contracted for during the month of March, but which it had not yet fulfilled. 49 Id. at 412. OPA believed that the “highest price charged” was to be determined by reference to sales that had actually been delivered; reference to contracted charges could be made only if there had been no delivery in the month of March, which was not the case. 50 Id. at 415. The delivery that had been made was simply at a lower price than Seminole Rock would have preferred. See id. In fact, at the same time it issued the General Maximum Price Regulation, OPA had issued a “bulletin” in which it used the more precise “actually delivered” language. 51 Id. at 417. Siding with OPA, the Supreme Court stated that, in interpreting an agency’s regulation, “the ultimate criterion is the admin­is­trative interpretation, which becomes of controlling weight unless it is plainly erroneous or inconsistent with the regulation.” 52 Id. at 414.

Little has changed in the black letter formulation of this principle between then and now. 53 See, e.g., Auer v. Robbins, 519 U.S. 452, 461 (1997) (holding that an agency’s interpretation of its own regulations is “under our jurisprudence, controlling unless ‘plainly erroneous or inconsistent with the regulation’” (citing Robertson v. Methow Valley Citizens Council, 490 U.S. 332, 359 (1989))). But, as several scholars have shown, it is not clear that the Justices were aware that they were creating such an endur­ing principle in administrative law with their decision in Seminole Rock. Professor Jeffrey Pojanowski, for instance, argues that the “‘plainly errone­ous’ verbiage” from Seminole Rock originated as a product of a rela­tively straightforward application of the Court’s Skidmore review, 54 See Skidmore v. Swift & Co., 323 U.S. 134, 140 (1944) (holding that, even when not controlling, an agency decision may be persuasive based on the “thoroughness evident in its consideration, the validity of its reasoning, its consistency with earlier and later pronouncements, and all those factors which give it power to persuade, if lacking power to control”). which then morphed over the years into a standalone deference doctrine. 55 Pojanowski, supra note 35, at 88–89 (arguing that this evolution makes Auer bad law in the sense that it misstates Seminole Rock’s true meaning); see also Aditya Bamzai, Henry Hart’s Brief, Frank Murphy’s Draft, and the Seminole Rock Opinion, Yale J. on Reg.: Notice & Comment (Sept. 12, 2016), [] (noting that the briefing in Seminole Rock seemed geared toward application of Skidmore, not toward creation of a new deference doctrine).

Consistent with this conclusion, Seminole Rock only slowly caught on over the first few decades after the decision. As Figure 1 makes clear, cita­tions to Seminole Rock first ebbed through the 1950s and then began to occur more regularly beginning in the 1960s. During this period, courts began to demonstrate an increasing willingness to apply Seminole Rock’s deference principle outside the price control context and to interpreta­tions that were not “official.” 56 See Sanne H. Knudsen & Amy J. Wildermuth, Unearthing the Lost History of Seminole Rock, 65 Emory L.J. 47, 69­–70 (2015) [hereinafter Knudsen & Wildermuth, Unearthing Lost History] (citing Boesche v. Udall, 303 F.2d 204 (D.C. Cir. 1961), aff’d, 373 U.S. 472 (1963)). In addition, whereas in the years immedi­ately following Seminole Rock courts tended to “engage[ ] in what looked like de novo interpretive analysis, only to cap off their decision with a reference and citation to Seminole Rock,” courts in the 1960s “began to articulate Seminole Rock as giving rise to a type of rebuttable presumption, a burden that would have to be overcome if the court were to adopt a contrary interpretation.” 57 Id. at 73–74. In the 1965 case Udall v. Tallman, the Supreme Court gave Seminole Rock some added authority as a general administra­tive law principle (rather than a provincial enclave of price-regulation law) 58 380 U.S. 1, 16–17 (1965). by “tying Seminole Rock to a broader body of well-accepted statutory interpretation doctrines.” 59 Knudsen & Wildermuth, Unearthing Lost History, supra note 56, at 78.

Figure 1: Total Circuit Court Citations to Auer and Seminole Rock, 1945–2016 60 The data reported were obtained through Lexis Advance by searching for cita­tions to the Shepard’s Citations footnotes concerning the principle of deference in both Auer and Seminole Rock and aggregating the citations by type and by year. Total citations include those defined by Shepard’s as neutral and routine citations. The vertical dashed line shows the year 1997, when Auer was decided.

Walters Figure 1

These changes cultivated a steady, but quite limited, institutionaliza­tion of Seminole Rock in the circuit courts through the 1970s, 1980s, and early 1990s, as Figure 1 demonstrates. While average total citations to Seminole Rock increased by more than 800% from the lowest period (1955–1965) to the period just before Auer was decided in 1997 (1990–1996), the frequency of citation remained low, averaging just twenty-one citations per year from 1970 through 1996. And while the Supreme Court occasionally cited Seminole Rock during the six years prior to 1997, it did so without closely examining the doctrine. 61 See, e.g., Shalala v. Guernsey Mem’l Hosp., 514 U.S. 87, 94–95 (1995) (“The Secretary’s position . . . is a reasonable regulatory interpretation, and we must defer to it.”); Thomas Jefferson Univ. v. Shalala, 512 U.S. 504, 512 (1994) (“We must give substantial deference to an agency’s interpretation of its own regulations.”); Martin v. Occupational Safety & Health Review Comm’n, 499 U.S. 144, 151 (1991) (“[W]e presume that the power authoritatively to interpret its own regulations is a component of the agency’s delegated lawmaking powers.”); Robertson v. Methow Valley Citizens Council, 490 U.S. 332, 358–59 (1989) (“This interpretation of the agency’s own regulation is not ‘plainly erroneous or inconsistent with the regulation,’ and is thus controlling.” (quoting Bowles v. Seminole Rock & Sand Co., 325 U.S. 410, 414 (1945))); Mullins Coal Co. v. Dir., Office of Workers’ Comp. Programs, 484 U.S. 135, 159–60 (1987) (“The agency’s interpretation . . . is deserving of substantial deference ‘unless it is plainly erroneous or inconsistent with the regulation’ . . . .” (quoting Seminole Rock, 325 U.S. at 414)); Lyng v. Payne, 476 U.S. 926, 939 (1986) (“[A]n agency’s construction of its own regulations is entitled to substantial deference.”).

The principle’s low profile changed abruptly in 1997 with the Court’s decision in Auer v. Robbins. 62 519 U.S. 452 (1997). Of course, as Figure 1 shows, citations to Seminole Rock itself did not change much, but insofar as Auer enunciated much the same principle, there was a significant change after 1997. In Auer, the question centered on the lawfulness of the Secretary of Labor’s interpretation of its “salary-basis test” regulations 63 Id. at 454–55. —a policy interpreting the Fair Labor Standards Act’s exemption from entitlement to overtime pay for “bona fide execu­tive, administrative, or professional” employees. 64 See 29 U.S.C. § 213(a)(1) (2012). Part of the salary-basis test was whether an employee was subject to disciplinary reductions in salary. If employees were “subject to” such requirements, they were not salaried employees and, therefore, not exempt. 65 See Auer, 519 U.S. at 461 (describing the Secretary of Labor’s interpretation of the salary-basis test); see also 29 C.F.R. § 541.118(a) (2018). The petitioners wanted to be classified as nonexempt, nonsalaried employees because they were, at least in theory, subject to disciplinary reductions in pay under their employee manual. 66 Auer, 519 U.S. at 459–60. But in an amicus brief requested by the Court, the Secretary of Labor argued that its salary-basis test required a showing of “an actual practice of making such deductions or an employment policy that creates a ‘significant likelihood’ of such deductions.” 67 Id. at 461 (quoting Brief for the United States as Amicus Curiae Supporting Affirmance at 21, Auer, 519 U.S. 452 (No. 95-897), 1996 WL 595843). Characteriz­ing the “controlling unless ‘plainly erroneous or inconsistent with the regulation’” language from Seminole Rock as a “deferential standard,” the unanimous Court, in an opinion written by Justice Scalia, concluded that the Secretary’s interpretation was valid. 68 Id. (citing Robertson v. Methow Valley Citizens Council, 490 U.S. 332, 359 (1989)). The critical inquiry, in the Court’s estimation, was whether the triggering language “subject to” in the salary-basis regulations could support a more restrictive interpreta­tion that would preclude the mere technical possibility of disciplinary reductions from preventing application of the Fair Labor Standards Act exemption. 69 Id.; see also 29 C.F.R. § 541.118(a). To this, the Court replied that the words “comfortably bear[ ] the meaning the Secretary assigns.” 70 Auer, 519 U.S. at 461. Nor did the fact that the inter­pretation was contained only in a brief filed in the litigation sway the Court’s conclusion that Seminole Rock demanded deference to the Secretary’s interpretation. 71 Id. at 462.

With the unanimous decision in Auer, the Supreme Court clarified that, whatever it had once meant, Seminole Rock now required substantial deference to agencies’ interpretations of their own regulations. The Secretary of Labor’s interpreta­tion was hardly inevitable, and its justifica­tion was far from a paragon of transparent reasoning, but that interpreta­tion was nonethe­less entitled to deference because it was not specifi­cally foreclosed by the regulatory text. As can be seen in Figure 1, this strong statement about the scope of the long-standing Seminole Rock prin­ciple led to a substantial increase in total citations to the doctrine. Total citations to Auer and Seminole Rock rose by 330% in 1997; positive citations likewise rose by over 2,300%. After Auer, it became possible to describe the princi­ple as a “full-blown and widely applied ‘axiom of judicial review.’” 72 Sanne H. Knudsen & Amy J. Wildermuth, Lessons from the Lost History of Seminole Rock, 22 Geo. Mason L. Rev. 647, 648 (2015) (quoting Allen M. Campbell Co. Gen. Contractors v. Lloyd Wood Const. Co, 446 F.2d 261, 265 (5th Cir. 1971)).

B. Recent Developments—Auer’s Second Revolution

Between 1997 and 2005, citations to Auer and Seminole Rock remained stable and overwhelmingly favorable to the doctrine, with total citations averaging about sixty-seven per year for the period, explicitly positive citations holding at about fifteen per year, and negative citations remaining rare, at about two per year. Beginning in 2006, however, Auer/Seminole Rock deference experienced a second revolution. Total cita­tions essentially doubled from 2006 to 2016. Much of this increase is actually attributable to the growth of positive treatments of the doctrine: Positive citations to Auer and Seminole Rock doubled over this period. But some of the change appears to be driven by growing concern about the doctrine, too, with the courts quadrupling the average yearly rate of explic­itly negative treatments during this period.

Figure 2: Circuit Courts’ Negative and Positive Citations to Auer and Seminole Rock, 1995–2016 73 The data reported were obtained through Lexis Advance by searching for cita­tions to the Shepard’s Citations footnotes concerning the principle of deference in both Auer and Seminole Rock and aggregating the citations by type and by year. The vertical dashed line shows the year 1997, when Auer was decided.


Walters Figure 2

The impetus behind this second revolution was a series of Supreme Court cases giving greater attention—sometimes sharply critical—to the doctrine. First, in early 2006, the Court decided Gonzales v. Oregon, hold­ing that the Attorney General’s interpretive rule prohibiting the use of controlled substances in carrying out otherwise state-sanctioned assisted suicides was invalid under the terms of the Controlled Substances Act. 74 546 U.S. 243, 274–75 (2006); see also id. at 249 (citing 21 U.S.C. §§ 801–971 (2012)). Although the Court nominally reaffirmed the validity of the Auer doc­trine, it ultimately concluded that the doctrine was “inapplicable” because “the underlying regulation [did] little more than restate the terms of the statute itself.” 75 Id. at 256–57. By endorsing a new “anti-parroting princi­ple,” the Gonzales decision suggested for the first time that there might be a “step zero” 76 Scholars often refer to a somewhat analogous “Chevron step zero,” where courts consider whether certain criteria—namely, whether Congress has expressly or implicitly delegated authority to the agency to make rules with the force of law—are met before even applying Chevron’s two steps. See Thomas W. Merrill & Kristin E. Hickman, Chevron’s Domain, 89 Geo. L.J. 833, 873 (2001) (describing “step zero” as “the inquiry that courts should undertake before moving on to step one of Chevron, or turning instead to Skidmore (or resolving the issue de novo)”); Cass R. Sunstein, Chevron Step Zero, 92 Va. L. Rev. 187, 191 (2006) (describing the “step zero” inquiry as the “inquiry into whether the Chevron framework applies at all”). requiring that certain criteria be fulfilled before an agency’s interpretation could be afforded Auer deference. The case drew little attention from observers at the time, but in light of subsequent devel­opments, it is clear that the Court’s power to “simply invent[ ] an exception to Auer 77 The Supreme Court, 2005 Term—Leading Cases, 120 Harv. L. Rev. 361, 365 (2006). undermined the stability of the doctrine and opened the door to more concerted questioning by federal courts. 78 See Stephen M. Johnson, Bringing Deference Back (But for How Long?): Justice Alito, Chevron, Auer, and Chenery in the Supreme Court’s 2006 Term, 57 Cath. U. L. Rev. 1, 3 (2007) (suggesting that “Gonzales seemed to signal a shift away from Auer deference for an agency’s interpretations of its regulations”). Professor Stephen Johnson noted two other cases, however, that signaled vitality for the doctrine. Id. (citing Nat’l Ass’n of Home Builders v. Defs. of Wildlife, 551 U.S. 644 (2007); Long Island Care at Home, Ltd. v. Coke, 551 U.S. 158 (2007)). One might add Coeur Alaska, Inc. v. Southeast Alaska Conservation Council to the list of cases that signaled the Court was still not quite fully convinced of the dangers of Auer. 557 U.S. 261, 274–75 (2009). The case involved an interpretation of EPA’s regulation determining which agency, EPA or the Army Corps of Engineers, had authority to issue a fill permit under the Clean Water Act. Id. at 265–66. The Court cited Auer in holding that EPA’s interpretation was acceptable. Id. at 274–75.

It was not until six years later, however, that any of the Justices explic­itly broached the possibility of jettisoning Auer deference. In 2011, the Court in Talk America, Inc. v. Michigan Bell Telephone Co. afforded Auer defer­ence to the interpretation that the Federal Communications Commission (FCC) advanced of its regulations governing what kinds of access to transmission facilities incumbent telecommunications providers must give to competitors. 79 564 U.S. 50, 53 (2011). The majority was satisfied that there was “no danger that deferring to the Commission would effectively ‘permit the agency, under the guise of interpreting a regulation, to create de facto a new regulation,’” 80 Id. at 63 (quoting Christensen v. Harris Cty., 529 U.S. 576, 588 (2000)). nor was there “any other ‘reason to suspect that the interpretation does not reflect the agency’s fair and considered judgment on the matter in question.’” 81 Id. at 63–64 (quoting Auer v. Robbins, 519 U.S. 452, 462 (1997)). In a concurrence, Justice Scalia agreed that FCC’s interpretation of the regulation controlled, but only because it was “the fairest reading of the orders in question.” 82 Id. at 67–68 (Scalia, J., concurring). With respect to Auer deference, Justice Scalia indicated that he had come to have second thoughts about the doctrine since his opinion for a unanimous Court in Auer—a change of heart that some have observed created “widespread confusion” in the circuits. 83 See, e.g., Kevin O. Leske, Splits in the Rock: The Conflicting Interpretations of the Seminole Rock Deference Doctrine by the U.S. Courts of Appeals, 66 Admin. L. Rev. 787, 832 (2014). Justice Scalia argued that, upon further reflec­tion, it is “contrary to fundamental principles of separation of pow­ers to permit the person who promulgates a law to interpret it as well,” and doing so in fact “encourages the agency to enact vague rules which give it the power, in future adjudications, to do what it pleases.” 84 Talk America, 564 U.S. at 68–69 (Scalia, J., concurring).

Justice Scalia’s about-face, and his theoretically driven argument that Auer has perverse effects on rule writing, provided a strong signal of grow­ing discontent with the doctrine. 85 See Aaron L. Nielson, Beyond Seminole Rock, 105 Geo. L.J. 943, 955 (2017) [hereinafter Nielson, Beyond Seminole Rock] (describing the concurrence as “the first shot” in the assault on Auer). Over the next few years, the depth of that discontent among the Court’s other Justices began to reveal itself. First, in Christopher v. SmithKline Beecham Corp., the Court clarified another requirement in Auer’s emerging “step zero”: that the agency’s interpretation cannot be one that creates a risk of “unfair surprise,” 86 567 U.S. 142, 155–56 (2012) (internal quotation marks omitted) (quoting Long Island Care at Home, Ltd. v. Coke, 551 U.S. 158, 170–71 (2007)). whether because it “conflicts with a prior interpretation” or because it appears to be “nothing more than a ‘convenient litigating position.’” 87 Id. (quoting Bowen v. Georgetown Univ. Hosp., 488 U.S. 204, 213 (1988)). Echoing Justice Scalia’s Talk America concurrence, Justice Alito’s majority opinion in Christopher invoked the self-delegation thesis but declined to reach the question of whether the doctrine ought to be discarded. 88 Id. at 158–59 (“Our practice of deferring to an agency’s interpretation of its own ambiguous regulations . . . creates a risk that agencies will promulgate vague and open-ended regulations that they can later interpret as they see fit . . . .”). The next term, the Court took its most serious collective look at the doctrine to date in Decker v. Northwest Environmental Defense Center. 89 568 U.S. 597 (2013). Justice Scalia penned a partial dissent to the application of Auer deference, urging his colleagues to recognize that “enough is enough.” 90 Id. at 616 (Scalia, J., concurring in part and dissenting in part). Chief Justice Roberts, joined by Justice Alito, also authored a concurring opinion that acknowl­edged their reticence about Auer, 91 Id. at 615 (Roberts, C.J., concurring) (“It may be appropriate to reconsider [Auer deference] in an appropriate case.”). but for the time, the emerging coali­tion against Auer again declined to force the issue, preferring to “await a case in which the issue is properly raised and argued.” 92 Id. at 616. Finally, although the question in Perez v. Mortgage Bankers Ass’n only tangentially raised ques­tions related to Auer, most of the Court’s conservatives again laid out their concerns with Auer in separate opinions. 93 135 S. Ct. 1199, 1210–11 (2015) (Alito, J., concurring in part and concurring in the judgment) (“I await a case in which the validity of Seminole Rock may be explored through full briefing and argument.”); id. at 1211–12 (Scalia, J., concurring in the judgment) (“By supplementing the APA with judge-made doctrines of deference, we have revolutionized the import of interpretive rules’ exemption from notice-and-comment rulemaking. Agencies may now use these rules not just to advise the public, but also to bind them.”); id. at 1213 (Thomas, J., concurring in the judgment) (questioning “the legitimacy of our precedents requiring deference to administrative interpretations of regulations”).

Prior to its grant of certiorari in Kisor v. Wilkie, 94 See supra note 19 and accompanying text. the Court had several opportunities to take up cases that would have squarely presented the question of whether to over­turn Auer. Until now, however, the Justices in favor of revisiting Auer had not appeared to have the necessary votes. 95 It is reasonably clear that the confirmation of Justice Kavanaugh will change the math. Although he never had occasion to weigh in on Auer deference in a live case during his time as a circuit judge, overall, he appears to be a deference skeptic. See Christopher Walker, Judge Kavanaugh on Administrative Law and Separation of Powers (Corrected), SCOTUSblog (July 26, 2018), [] (“Although Kavanaugh has not addressed the propriety of Auer deference . . . , his concerns about interpretive doctrines that turn on ambiguity, coupled with his views on separation of powers, seem to suggest he would be receptive to calls to eliminate—or at least further limit—Auer deference.”). In extrajudicial remarks before his nomination, Justice Kavanaugh opined that Justice Scalia’s dissent in Decker had “eviscerated” Auer deference and predicted that “some day Auer will be overruled, and that Justice Scalia’s dissent in Decker will be the law of the land.” Adam K. Lasky & Daniel Radthorne, Why Judge Kavanaugh’s Confirmation Could Be Good News for Government Contractors, Procurement Playbook (Sept. 4, 2018), []. That outcome appears to be in Justice Kavanaugh’s hands now. First, amidst a maelstrom of specula­tion that the Supreme Court would grant a petition for writ of certiorari in United Student Aid Funds precisely to overturn Auer, the Court instead denied the petition over a vigorous dissent from the denial by Justice Thomas. 96 United Student Aid Funds, Inc. v. Bible, 136 S. Ct. 1607, 1608 (2016) (Thomas, J., dissenting from the denial of certiorari). Then, after the Fourth Circuit ruled that a school district had to allow a transgender student to use the bathroom matching his gender identity, citing and deferring to a Department of Education guidance, 97 See supra notes 9–12 and accompanying text. the Supreme Court received a petition for certiorari squarely requesting that the Court abandon Auer. In a surprising development, the Court granted this petition, but not on the question of whether Auer should be overruled, 98 Gloucester Cty. Sch. Bd. v. G.G. ex rel. Grimm, 137 S. Ct. 369 (2016) (mem.) (granting certiorari to only two of the three questions presented by the petition); see also Amy Howe, Court Adds Five New Cases, Including Transgender Bathroom Dispute, to Docket, SCOTUSblog (Oct. 28, 2016), [] (“In granting review today, the justices sidestepped the most prominent issue they had been asked to take on: whether they should overrule their decision in Auer, which has been the target of criticism by conservative lawyers and jurists.”). prompting speculation that the Court would at most narrow Auer deference. 99 See Jonathan H. Adler, Supreme Court to Hear Transgender Bathroom Case, Wash. Post: The Volokh Conspiracy (Oct. 28, 2016),
news/volokh-conspiracy/wp/2016/10/28/supreme-court-to-hear-transgender-bathroom-case/ (on file with the Columbia Law Review) (“If Auer and Seminole Rock are to be reconsid­ered, it will not be in this case.”).
Even that possibility evaporated—at least for the time being—when the Trump Administration withdrew the guid­ance at issue, prompting the Supreme Court to return the case to the Fourth Circuit. 100 Gloucester Cty. Sch. Bd. v. G.G. ex rel. Grimm, 137 S. Ct. 1239, 1239 (2017) (mem.) (vacating and remanding the case to the Fourth Circuit “in light of the guidance document issued by the Department of Education and Department of Justice”); see also Amy Howe, Justices Send Transgender Bathroom Case Back to Lower Courts, No Action on Same-Sex Marriage Cake Case, SCOTUSblog (Mar. 6, 2017), [].

The effect of this second wave of Auer cases, as the data in Figure 1 show, has not been to stanch the flow of citations to Auer; it has seem­ingly been to bring Auer to unprecedented prominence. 101 See supra Figure 1. At the same time, the data show that the federal judiciary is becoming deeply divided about the desirability of the doctrine, and at least some of the Justices on the Court are clearly interested in revisiting Auer. 102 See supra notes 90–93 and accompanying text. The Court is poised to do so this term in Kisor. 103 See supra note 19 and accompanying text. The next Part of the Article examines the source of the deep disagreement developing in the federal courts: a theo­retical argument about Auer’s behav­ioral incentives.

II. Auer’s Perverse Incentives

The principle of deference to agency interpretations of regulations serves a fundamental purpose in administrative law. 104 See Scott H. Angstreich, Shoring up Chevron: A Defense of Seminole Rock Deference to Agency Regulatory Interpretations, 34 U.C. Davis L. Rev. 49, 92–93 (2000) (citing leading administrative law treatises endorsing Seminole Rock “as ‘based on common sense’” and noting that the Court often treated the rationale for deference as “self-evident” (quoting 1 Kenneth Culp Davis & Richard J. Pierce, Jr., Administrative Law Treatise § 6.10, at 282 (3d ed. 1994))); see also Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 616 (2013) (Roberts, C.J., concurring) (noting that Auer deference presents issues “going to the heart of administrative law”). Whenever there is rulemaking, there is an inescapable need for expert application of those rules to increasingly fine-grained, unforeseen situations that fall within the purview of the rules 105 The literature identifies two somewhat related bases for deference to agency interpretations: First, agencies are in the best position to resolve ambiguities according to the “regulative intent.” See Stephenson & Pogoriler, supra note 7, at 1456; Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 306–07. Second, agencies have exper­tise that allows them to more coherently “resolve any gaps, conflicts, or ambiguities” in the regula­tory scheme. See Stephenson & Pogoriler, supra note 7, at 1456. For a general account of the unique challenges of regulatory interpretation, as opposed to statutory interpre­tation, see Kevin M. Stack, Interpreting Regulations, 111 Mich. L. Rev. 355, 365–77 (2012). —even the most perspicacious of rule writers cannot anticipate every circumstance that might require application of a rule. 106 Nor do they necessarily want to. See Barmore, supra note 16, at 819 (“An agency could try to answer every potential interpretive question, but such clarity would increase the length and complexity of regulations until they were too opaque for regulated parties to understand.”). It is for precisely this reason that the APA provides for avenues other than notice-and-comment rulemaking, such as nonlegislative rule­mak­ing. 107 See 5 U.S.C. § 553(b)(A) (2012) (describing nonlegislative rulemaking); id. § 554 (describing formal adjudication); see also Franklin, supra note 40, at 303–04 (highlighting the benefits of nonlegislative rulemaking, including the relief it grants to resource-strapped agencies that might otherwise be forced to undergo cumbersome notice-and-comment rulemaking). It would be highly inefficient to require every needed clari­fication or interpretation to be formulated as an amendment to a rule—even as it would increase accountability to do so—yet that is what would be encouraged by deferring to rules and not to interpretations of rules. In essence, Auer deference attempts to balance out pragmatically a classic tension in administration—namely, “accommodating the need for agency flexibility while guarding against the specter of what Justice Jackson memora­bly described as ‘administrative authoritarianism’—the ‘power to decide without law.’” 108 Clarke, supra note 17, at 178 (quoting SEC v. Chenery Corp., 332 U.S. 194, 216 (1947) (Jackson, J., dissenting)) (describing these as “two classic concerns of the modern administrative state”); see also Stephenson & Pogoriler, supra note 7, at 1453 (“[W]hole­sale rejection of Seminole Rock would be quite disruptive, and would likely have serious disadvantages, including loss of regulatory flexibility and efficiency.”).

Auer vindicates core values of administrative flexibility and efficiency by giving rule writers some assurance that their agency’s good-faith applica­tion of the codified regulatory text to novel, but related, problems will be respected by the courts, in part because of the recognition that agencies are hardly inexpert when it comes to determining the intended meaning and purpose of regulatory text they drafted themselves. 109 See Martin v. Occupational Safety & Health Review Comm’n, 499 U.S. 144, 151–53 (1991) (“Because the Secretary promulgates these standards, the Secretary is in a better position than is the Commission to reconstruct the purpose of the regulations in ques­tion.”). Indeed, some scholars have argued that agencies’ unique ability to interpret regu­latory texts is a reason for making Auer deference “more robust than Chevron deference.” Stephenson & Pogoriler, supra note 7, at 1454–57 (acknowledging the argument but noting that it has been replaced by an ascendant “pragmatic” rationale analogous to the rationale for Chevron); see also Angstreich, supra note 104, at 92–93 (citing sources mak­ing the argument). In doing so, Auer gives agencies greater freedom to avoid undergoing cumber­some notice-and-comment rulemaking whenever they seek to make these adjustments. 110 See Barmore, supra note 16, at 819­–20 (“Auer allows agencies to apply their rules to unanticipated situations that fall within the interstices of the regulatory language.”). It bears noting, however, that agencies are apparently quite involved in revisiting existing rules through legislative rulemaking. As Professor Wendy Wagner and colleagues show, notwithstanding how cumbersome notice-and-comment rulemaking might be, agencies commonly make these workaday adjustments through this channel. Wendy Wagner et al., Dynamic Rulemaking, 92 N.Y.U. L. Rev. 183, 202, 211 (2017) [hereinafter Wagner et al., Dynamic Rulemaking] (estimating, based on the authors’ sample, that agencies revised 73% of rules at least once and engaged in notice-and-comment rulemaking for one-third of those revisions). Some might quibble with the precise balance struck by Auer, arguing that more interpretations should be made in the form of notice-and-comment rulemaking. 111 See Funk, supra note 42 (arguing that Auer plays a role in encouraging agencies to “cut corners” by adopting interpretive rules instead of amending rules through notice-and-comment rulemaking and that this occurs with sufficient regularity to justify some change to the doctrine). But these arguments, like arguments against the Court’s hands-off approach to the nondelegation doctrine, fail to persuade because they fail to explain why accountability must always be maximized at the expense of efficiency and flexibility.

The growing skepticism of Auer from vocal members of the federal judi­ciary, as detailed in section I.B, does not rely primarily on a fruitless second-guessing of the need for agency flexibility. Instead, the skeptical treatments from some Justices in cases like Talk America, Decker, Christopher, and Perez rely on a theoretical apprehension about the doc­trine’s incentives for rule writing. That is—according to Manning, who largely inspired the recent judicial questioning of the doctrine 112 See Nielson, Beyond Seminole Rock, supra note 85, at 954 (noting that “Manning’s analysis appears to have played a considerable role in flipping Justice Scalia’s views” on Auer); see also Leske, supra note 83, at 790. Auer has an “untoward effect upon [the agency’s] incentive to speak precisely and transparently when it promulgates regulations.” 113 See John F. Manning, Constitutional Structure and Judicial Deference to Agency Interpretations of Agency Rules, 96 Colum. L. Rev. 612, 655 (1996) [hereinafter Manning, Constitutional Structure]. For a more generalized critique of the perverse incentives that regulation by guidance creates, see Robert A. Anthony & Michael Asimow, The Court’s Deferences—A Foolish Inconsistency, Admin. & Reg. L. News, Fall 2000, at 10, 11. It is that appre­hen­sion that the Article now turns to in more detail.

A. The Critique of Auer—Incentives for Self-Delegation

Auer deference differs fundamentally from other deference doc­trines in administrative law that concern statutory interpretations by agen­cies. With a doctrine like Chevron, the agency itself has no control over the scope of the discretion embedded in the raw statutory text, and Congress retains no power to limit the exercise of interpretive discretion after the fact except through further legislation. 114 Talk Am., Inc. v. Mich. Bell Tel. Co., 564 U.S. 50, 68 (2011) (Scalia, J., concurring) (“When Congress enacts an imprecise statute that it commits to the imple­mentation of an executive agency, it has no control over that implementation (except, of course, through further, more precise, legislation).”). This institutional alloca­tion of law-writing and law-interpreting authority means that “Congress’s incentive is to speak as clearly as possible on the matters it regards as important,” 115 Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 620 (2013) (Scalia, J., concurring in part and dissenting in part). substantially limiting the scope of agencies’ interpre­tive authority. In contrast, with Auer, there is a fusion of legislative and interpretative authority within the same institution. 116 See id. Agencies write rules in one breath and interpret or enforce those rules in another. Affording deference to regulatory interpretations, as Auer does, makes agencies authoritative as to the meaning of regulatory text they have previ­ously written.

According to critics, this distinction makes all the difference and ren­ders Auer highly suspect. 117 This argument is often cast in terms of formalist separation of powers logic, see id. at 619 (arguing that there could be “no congressional implication that the agency can resolve ambiguities in its own regulations” because “that would violate a fundamental principle of separation of powers—that the power to write a law and the power to interpret it cannot rest in the same hands”), but it is ultimately based on, and influential because of, a consequentialist claim about the dangers of self-delegation that inhere in a breakdown of the strict separation of powers, see Thomas Jefferson Univ. v. Shalala, 512 U.S. 504, 525 (1994) (Thomas, J., dissenting) (“It is perfectly understandable, of course, for an agency to issue vague regulations, because to do so maximizes agency power and allows the agency greater latitude to make law through adjudication rather than through the more cumbersome rulemaking process.”). Because it “allows an agency to both write the law (the regulation) and determine its application,” 118 Barmore, supra note 16, at 817. Auer defer­ence creates the possibility that agencies will seek to maximize their interpre­tive discretion by writing vague, underspecified regulatory text in the first instance. 119 Decker, 568 U.S. at 620 (Scalia, J., concurring in part and dissenting in part) (“[W]hen an agency interprets its own rules—that is something else. Then the power to prescribe is augmented by the power to interpret; and the incentive is to speak vaguely and broadly, so as to retain a ‘flexibility’ that will enable ‘clarification’ with retroactive effect.”). That is, “[r]egulatory ambiguity, unlike statutory ambigu­ity, does not entail an implicit delegation to another institution”—and that difference “makes such ambiguity relatively more attractive.” 120 Stephenson & Pogoriler, supra note 7, at 1461 (emphasis added).

Recall the Title IX example: Instead of bearing the cost of specifying its meaning ex ante, as Auer critics claim it would have been more likely to do if another actor had the ultimate interpretive authority, the Department of Education allegedly had no incentive to specify in its rule how it would like to treat transgender students with regard to bathroom choice. Under Auer, Manning notes, “if an agency issues an imprecise or vague regulation, it does so secure in the knowledge that it can insist upon an unobvious interpretation, so long as its choice is not ‘plainly errone­ous.’” 121 Manning, Constitutional Structure, supra note 113, at 657 (emphasis added by Manning) (quoting Bowles v. Seminole Rock & Sand Co., 325 U.S. 410, 414 (1945)). In other words, an agency could simply leave more discre­tion for itself on the front end, when writing a rule, and leave for another day the task of specifying its meaning, as courts applying Auer will simply defer to any plausible construction of capacious or underspecified lan­guage.

Taken to its logical endpoint, the argument is not merely that Auer elim­inates the incentive to speak with clarity but that it actually creates an incentive to “maximize” opacity. 122 Barmore, supra note 16, at 818 (“If Auer requires deference to an agency’s interpretation of ambiguous (but not unambiguous) regulations, agencies would maximize future flexibility and power by promulgating ambiguous regulations. . . . Critics worry that the incentive to promulgate vague regulations would lead to predictably more ambiguous regulations . . . .”); see also Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 308 (explaining Auer critics’ concern that agencies “will issue vague, broad regulations, knowing full well that when the time comes, they will be able to impose the interpretation they prefer”). When agencies consider making pol­icy by nonlegislative rule (that is, by guidance document, policy state­ment, and the like), the most important parameter to consider is how much risk there is that the interpretation will stretch the meaning of the regulatory text too far, such that a reviewing court would reject that interpre­tation. 123 See generally M. Elizabeth Magill, Agency Choice of Policymaking Form, 71 U. Chi. L. Rev. 1383, 1395–96 (2004) (exploring how an agency’s choice of regulatory form will impact the judicial review that its policy will receive). But because agencies have control over the regulatory text as well as the interpretation, this key parameter is actually under their control in the Auer context. Agencies could promulgate texts that are so vague that virtually any interpretations would be at least arguably consistent with them. The optimal strategy, critics charge, would amount to a type of regulatory sandbagging: deliberately withhold some clarity now in order to augment flexibility down the road. 124 Nielson, Beyond Seminole Rock, supra note 85, at 955 (“[W]hen agencies promulgate regulations that do not tackle the hard problems, the agency does not ‘pay’ upfront, and when the agency later issues an interpretative rule to tackle those problems—even though interpretative rules are not subject to the same rigorous procedure—the agency does not ‘pay later’ either.” (citing Stephenson & Pogoriler, supra note 7, at 1464)). Indeed, critics might specu­late that agencies could even subvert the statutes they are administer­ing by first translating statutory language into vague terms, and then retranslating that translation into the agency’s preferred mean­ing. As Justice Scalia argued in Decker, if Auer functions in this way, then the doctrine gives to agencies a “dangerous permission slip for the arroga­tion of power.” 125 Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 620 (2013) (Scalia, J., concurring in part and dissenting in part). For Manning, the danger of self-delegation inher­ent in any breakdown of the strict separation of rule-writing and rule-interpreting authority is sufficiently serious that it justifies a constitu­tional presumption against any institution or doctrine with that feature. 126 See Manning, Constitutional Structure, supra note 113, at 631; see also Anthony & Asimow, supra note 113, at 10 (arguing that the Auer doctrine violates separation of powers principles because “[i]t authorizes the agencies to make law through an informal format where Congress has not delegated such power either explicitly or by implication”).

The key prediction of the self-delegation thesis, then, is that agen­cies operating in a world with Auer deference will systemically underspecify the rules they write relative to a counterfactual world where Auer does not exist. 127 Some have also argued that agencies would have incentives to issue more nonlegislative rules than they otherwise would because of the availability of Auer. See Funk, supra note 42. On the margins, this may be true, although it is always the case that agencies have strong incentives to cut corners around notice-and-comment rulemaking regardless of the level of deference. Whether Auer makes a difference in this calculation is an open question (and one that deserves further research), but the existing evidence is, notably, inconsistent with the notion that there has been an explosion of significant guidance (relative to significant rulemaking) in recent decades. See Connor N. Raso, Note, Strategic or Sincere? Analyzing Agency Use of Guidance Documents, 119 Yale L.J. 782, 813–15 (2010) (finding that significant guidances make up a negligible percentage of agency work and are often used to “clarify highly technical details”). That is, whereas under ordinary circumstances, agencies would balance various considerations to settle on some kind of “optimal” level of rule precision, 128 Colin S. Diver, The Optimal Precision of Administrative Rules, 93 Yale L.J. 65, 71–72 (1983) (arguing that such an “optimal” approach would require a normative balancing of various factors, including “transparency, accessibility, and congruence”); see also Adam I. Muchmore, Uncertainty, Complexity, and Regulatory Design, 53 Hous. L. Rev. 1321, 1327–28 (2016) (arguing that much legal uncertainty is irreducible and that agencies bal­ance many factors in designing regulatory programs). Auer would seem to promise a greater payoff for writing a relatively vague rule, and in fact could be thought to shift base­line rule-writing styles across the board. With vague rule texts, it is far easier for agencies to cut corners. They can avoid notice-and-comment rulemaking and make major policy changes by issuing guidance docu­ments, policy statements, opinion letters, and the like instead, as they may be more naturally inclined to do anyway. 129 See, e.g., James T. Hamilton & Christopher H. Schroeder, Strategic Regulators and the Choice of Rulemaking Procedures: The Selection of Formal vs. Informal Rules in Regulating Hazardous Waste, Law & Contemp. Probs., Spring 1994, at 111, 127 (theorizing that strategic agencies might choose to regulate by guidance as a means of avoiding the costs of notice-and-comment rulemaking). Notice-and-comment proce­dures can be costly and time-consuming. 130 See Kathryn E. Kovacs, Getting Back to the Basics with Agency Rulemaking, Reg. Rev. (Nov. 13, 2017), [] (“Although scholars argue about why the rulemaking process is ossifying and how much it is happening, there is little debate that rulemaking is a big pain for federal agencies.”). Indeed, by one study’s estimation, the average time to complete notice-and-comment rulemak­ing is between one and three years. 131 Anne Joseph O’Connell, Political Cycles of Rulemaking: An Empirical Portrait of the Modern Administrative State, 94 Va. L. Rev. 889, 958–59 (2008) [hereinafter O’Connell, Political Cycles of Rulemaking] (“The average duration of completed rulemakings, both significant and more routine, for the ten agencies in Charts 5 and 6 ranged from 243.74 days for the [Department of Commerce] to 760.93 days for the [Department of Health and Human Services].”). In contrast, agencies can, in princi­ple, issue a nonlegislative rule in a matter of days, as there is no pre­scribed procedure in the APA for such actions. 132 See 5 U.S.C. § 553(b)(3)(A) (2012). Executive actions have added some procedural gloss. For instance, the Office of Management and Budget issued a bulletin during the George W. Bush Administration encouraging agencies to develop standard procedures for the issuance of significant guidance documents. See Office of Mgmt. & Budget, Exec. Office of the President, OMB Bull. No. 07-02, Final Bulletin for Agency Good Guidance Practices (2007),
omb/memoranda/2007/m07-07.pdf [].
The cost savings on the back end can literally be accounted for in the decision about how specific to be on the front end.

B. The Defense of Auer

Although the self-delegation critique of Auer has been quite influen­tial, both in the courts and in the academy, a few voices supportive of the doctrine have emerged. Recently, Sunstein and Vermeule addressed the tenuous assumptions behind the separation of powers critique of Auer. 133 Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 310 (discussing the separation of powers critique—which suggests that it involves a “constitutionally suspect combination of the power to make law with the power to interpret law”—and rejecting that critique as “both unsound and too sweeping”). They note that, strictly speaking, agencies are not engaged in legislating or adjudicating when they engage in rule writing and rule interpretation, respectively. Rather, they are engaged in executing the law. 134 Id. at 311 (“When agencies make rules and interpret law in the course of exercising their statutory grant of authority, they are carrying out or completing a legislative plan . . . .”); see also Jack L. Goldsmith & John F. Manning, The President’s Completion Power, 115 Yale L.J. 2280, 2282 (2006) (defining the “completion power” as “the President’s authority to prescribe incidental details needed to carry into execution a legislative scheme, even in the absence of any congressional authorization to complete that scheme”). This task, with its mix of legislative and expository authority, is pervasive in the adminis­trative state. 135 See Lawson, supra note 32, at 1248 (discussing the extensive enforcement activities of a typical federal agency). Thus, the self-delegation argument that critics make against Auer has far-reaching implications not just for agency rule writing, but also for the majority of actions agencies undertake. 136 Adrian Vermeule, Law’s Abnegation: From Law’s Empire to the Administrative State 78 (2016) [hereinafter Vermeule, Law’s Abnegation] (“If the combination of lawmaking and law-interpreting functions in agencies really is constitutionally suspect as such, then there are much larger problems than Auer to discuss.”); Cass R. Sunstein & Adrian Vermeule, Auer, Now and Forever, Yale J. on Reg.: Notice & Comment (Sept. 19, 2016), [] [hereinafter Sunstein & Vermeule, Now and Forever] (“The constitutional critique of Auer rests on generalities about the separation of lawmaking from law-execution and law-interpretation. If those generalities were applied consistently, however, they would require declaring unconstitutional dozens of major federal agencies exercising combined functions.”). Indeed, Sunstein and Vermeule argue that such an amalgam of institu­tional functions is so common in the modern administrative state that those who advance a formalistic critique of Auer bear a heavy burden of explaining why Auer must go but modern administrative agencies may stay. 137 Sunstein & Vermeule, Now and Forever, supra note 136. Others have noted that Auer is part of a fragile equilibrium; change the ground rules, and different and potentially worse incentives might take hold. For instance, Professor Aaron Nielson argues that one possible effect of overturning or watering down the Auer doctrine would be to discourage agencies from writing rules in the first place, instead using their prerogative under the Chenery doctrine to make policy itera­tively through ad hoc adjudication. 138 Nielson, Beyond Seminole Rock, supra note 85, at 948 (noting that, under the Chenery doctrine, “agencies can choose to interpret the statutes they administer by promul­gating rules or, if they prefer, by simply enforcing the statutes directly through case-by-case adjudication”). Nielson argues that Auer’s infusion of back-end flexibility makes rule­making more attractive, with attendant benefits for notice of legal obligations even if gaps remain to be interpreted. Id. at 948–50. But if Auer were “off the table, agencies may rely more often on Chenery” and make policy in an iterative, unpredictable fashion through adjudication. Id. at 948. If the concern about Auer is in part about its failure to deliver fair notice and public accountability, a world in which agencies avoid prospective rulemaking in favor of retrospective and often-piecemeal adjudication would be a decidedly worse one. More­over, some have argued that the likely result of overturning Auer and hand­ing interpretive authority to courts would be to greatly undermine clarity in the law. 139 See Derek A. Woodman, Rethinking Auer Deference: Agency Regulations and Due Process Notice, 82 Geo. Wash. L. Rev. 1721, 1736 (2014) (noting the “disuniformity that could result if courts were to substitute the agency’s interpretation of the regulation with their own”). Instead of having agencies announce one authorita­tive (albeit ephemeral) interpretation, there would likely be circuit splits about the proper interpretation of important regulatory provisions, thereby increasing the costs of compliance for regulated entities. 140 See id. Then there is the fact that overturning Auer might throw the validity of count­less existing interpretations, many of which have induced substantial reli­ance interests, into question. 141 Clarke, supra note 17, at 193 (noting that there are substantial reliance interests in interpretive rules and that virtually every interpretation under the APA would be affected in some way since Seminole Rock predated the APA).

These arguments, as persuasive as they might be, elide the central ques­tion in the debate over Auer: that is, the self-delegation thesis. Indeed, the response to the self-delegation thesis itself has been strikingly muted. Responding to the self-delegation thesis, Auer’s rehabilitators have generally thought it sufficient to point out what Manning antici­pated 142 Manning, Constitutional Structure, supra note 113, at 655–56 (anticipating that “an agency may have various other reasons to draft clear, self-limiting rules, even when it possesses the right of self-interpretation,” including keeping enforcement costs low, main­taining hierarchical control of agency staff, and binding future administrations). —that there are cross-cutting incentives that actually encourage clarity in rule writing. 143 Sunstein & Vermeule, Now and Forever, supra note 136 (“And there is a cross-cutting incentive as well: agencies who want to bind their own successors, perhaps because a change of administration looms, are better off creating a binding rule, repealable only through the same relatively costly process.”). Agencies gain greater assurance that regulated entities will comply with the law, as the agency understands it, when they craft with clarity their understanding of a legislative rule. 144 See Diver, supra note 128, at 72 (arguing that clarity in regulations can streamline enforcement, or perhaps even make it less necessary if regulated entities can be relied on to comply). Likewise, both agencies and regulated entities benefit from certainty against future changes when issues are settled in legislative rules rather than in later guidance documents or policy statements, 145 See Anthony, supra note 17, at 1317 (noting the confusion and uncertainty that can result when agencies rely on guidances and policy statements rather than on legislative rules). as rules can be formally changed only through a new legislative rule. 146 See Nat’l Family Planning & Reprod. Health Ass’n v. Sullivan, 979 F.2d 227, 234 (D.C. Cir. 1992) (stating that an “agency issuing a legislative rule is itself bound by the rule until that rule is amended or revoked”). Citing these cross-cutting incentives, Sunstein and Vermeule argue that Auer’s critics have commit­ted what they call the “sign fallacy”—“identify[ing] the likely sign of an effect and then . . . declar[ing] victory, without examining its magnitude” and “without asking whether it is realistic to think that the effect will be significant.” 147 Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 299–300.

Some commentators have also noted that there is an absence of the kind of circumstantial evidence that we would expect if the self-delega­tion theory had any power. For instance, no former agency general coun­sel has come forward to say that any rule has been deliberately obfus­cated in response to the incentives. 148 Ronald M. Levin, Auer and the Incentives Issue, Yale J. on Reg.: Notice & Comment (Sept. 19, 2016), [] (“When I refer to [the lack of] evidence, I do not mean to insist on specific case citations or empirical studies. As yet, however, the critics of Auer have not even produced any good anecdotes to support their theory.”). Quite to the contrary, when Professor Christopher Walker asked agency rule drafters what they thought about Auer, one respondent answered, “I personally would attempt to avoid issuing ambiguous regulations that we would then have to interpret.” 149 Christopher J. Walker, Inside Agency Statutory Interpretation, 67 Stan. L. Rev. 999, 1065 (2015) [hereinafter Walker, Statutory Interpretation] (internal quotation marks omitted). Such sentiments are widely shared among regulators. 150 See, e.g., Aditi Prabhu, How Does Auer Deference Influence Agency Practices?, Admin. & Reg. L. News, Winter 2015, at 11, 12 (“[A]gencies have a strong interest in writing clear regulations. Agencies can effectively enforce only clear regulations; otherwise, they risk running afoul of fair notice and due process considerations.”). Indeed, as Walker further shows, most agency rule writers are apparently generally unaware of the doctrine, at least relative to other administrative law doctrines, such as Chevron deference. 151 See Walker, Statutory Interpretation, supra note 149, at 1019 fig.1 (reporting survey data showing that only 53% of agency rule drafters knew of Auer/Seminole Rock deference by name, whereas 94% knew of Chevron deference by name). The same study revealed that only 39% of agency rule drafters actually claimed to use Auer/Seminole Rock in shaping their drafting decisions, id. at 1020, and it was even unclear what respondents had in mind when they said they “use” the doctrine, id. at 1066. These findings are consistent with the data analyzed in Part III: Not one of the preambles among rules in the sample contained any reference to either Seminole Rock or Auer.

In sum, the response to the self-delegation critique has been to say that there is a “palpable lack of realism” and “lack of empirical ground­ing” behind the self-delegation thesis. 152 Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 309–10 (“We do not believe that agencies often preserve ambiguity on purpose—in fact we think that that is highly unusual—but when they do, Auer is hardly ever . . . part of the picture. The critics speak abstractly of possible abuses, but present no empirical evidence to substantiate their fears.”); see also Levin, supra note 148 (“[T]he factual basis of the critique of Auer isn’t paltry. It’s one hundred percent guesswork.”). But, notably, Auer’s rehabilita­tors have not offered any empirical analysis of their own to test the self-delegation thesis or assess the strength of any cross-cutting incentives they postulate. Instead, the response to the self-delegation thesis has been to make an argument based on empirical agnosticism. 153 See Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 309 (refusing to press the cross-cutting incentives argument too far for fear that doing so would “commit the sign fallacy (with a different sign)”).

C. The Need for Evidence in Assessing the Self-Delegation Critique

Despite the efforts of Auer’s rehabilitators, Auer today finds its future uncer­tain. The Court has taken up the question of whether Auer and Seminole Rock should be overruled, 154 See supra note 19 and accompanying text. and it is not difficult to count to five. Chief Justice Roberts 155 See Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 615–16 (2013) (Roberts, C.J., concurring) (“The bar is now aware that there is some interest in reconsidering [Auer and Seminole Rock], and has available to it a concise statement of the arguments on one side of the issue.”). and Justices Alito, 156 See Perez v. Mortg. Bankers Ass’n, 135 S. Ct. 1199, 1210–11 (2015) (Alito, J., concurring) (“The opinions of Justice Scalia and Justice Thomas offer substantial reasons why the Seminole Rock doctrine may be incorrect.”). Thomas, 157 See Garco Constr., Inc. v. Speer, 138 S. Ct. 1052, 1052 (2018) (Thomas, J., dissenting from denial of certiorari) (“Seminole Rock is constitutionally suspect.”). and Gorsuch 158 See id. Justice Gorsuch’s views about deference doctrines may well be the strongest on the Court. See, e.g., Kathryn M. Schroeder & Jason B. Hutt, Gorsuch May Further Tip Balance Against Deference to EPA, Law360 (Feb. 14, 2017), https:// (on file with the Columbia Law Review) (discussing Justice Gorsuch’s expressed skepticism of judicial deference to agency interpretation and discussing the possibility of Auer being overturned following Justice Gorsuch joining the Court); see also Gutierrez-Brizuela v. Lynch, 834 F.3d 1142, 1149 (10th Cir. 2016) (Gorsuch, J., concurring) (“Chevron and Brand X permit executive bureaucracies to swallow huge amounts of core judicial and legislative power and concentrate federal power in a way that seems more than a little difficult to square with the Constitution of the framers’ design. Maybe the time has come to face the behemoth.”).
have expressed strong reservations about Auer, and Justice Kavanaugh’s previous statements as a circuit judge suggest some antipathy toward Auer as well. 159 See supra note 95. In the circuit courts, Auer continues to be followed (at least for now), but it also finds itself sub­jected to stern criticism. For example, Judge William Jordan of the Third Circuit recently indicated that he believes Auer “deserves another look,” arguing that it is “contrary to the roles assigned to the separate branches of government” and “embed[s] perverse incentives in the operations of government.” 160 Egan v. Del. River Port Auth., 851 F.3d 263, 278 (3d Cir. 2017) (Jordan, J., con­curring in the judgment).

Even when courts nominally accept the premises of Auer deference, the tenor of some analysis has recently shifted in a less deferential direc­tion due in part to courts’ recognition that the critique of Auer has pur­chase with some members of the Supreme Court. For instance, in Perez v. Loren Cook Co., a majority of the en banc Eighth Circuit held that a Department of Labor (DOL) regulation requiring “machine guarding” as a means of protection from “hazards such as those created by . . . rotat­ing parts” did not require barrier guards to protect against ejection of rapidly rotating pieces of metal, contrary to DOL’s interpretation. 161 803 F.3d 935, 937, 943 (8th Cir. 2015) (en banc) (quoting 29 C.F.R. § 1910.212(a)(1) (2015)). But as the four dissenting judges noted, the majority’s “hypertechnical” pars­ing of the regulation was contrary to the spirit of Auer and risked undermin­ing the Secretary of Labor’s ability to “adjust its interpreta­tion . . . over time.” 162 Id. at 944–50 (Melloy, J., dissenting). Even if Auer survives Kisor v. Wilkie in some form, skepticism and avoidance like this will likely continue to pervade courts’ decisionmaking.

The reason for the persistence and influence of the critique of Auer is that its account of strategic agencies’ self-delegating authority has a certain theoretical appeal. If one assumes that agencies have preferences, and that those preferences are chiefly to maximize power and discretion in a self-interested manner, 163 See David B. Spence & Frank Cross, A Public Choice Case for the Administrative State, 89 Geo. L.J. 97, 99 (2000) (describing the early public-choice literature as “por­traying agency bureaucrats as shirking, self-interested budget-maximizers who thwart the will of the people and good government”). then there is little that stands in the way of the seemingly elegant model offered by Auer’s critics. It may be that there are cross-cutting incentives that make agencies consciously pass on the opportunity for self-delegation, 164 See Levin, supra note 148 (noting that Auer’s incentive to write vague rules “does not exist in a vacuum” and that many factors “militate toward specificity rather than vagueness”). but this more complex story may not seem as facially plausible or nearly as self-interested as the one offered by Auer’s critics. This is presumably why Manning, fully aware of cross-cut­ting incentives for agencies to craft clear rules, nevertheless argued that Auer’s perverse incentives are stronger than the benign or laudable ones created by other doctrines or operational exigencies. 165 See supra note 142 and accompanying text. Although Sunstein and Vermeule are right to point out that critics’ self-delegation theory commits the “sign fallacy,” 166 See supra note 147 and accompanying text. those critics can still appeal to an intuition that suggests that Auer must matter in some way by creating some kind of structural bias. The persuasiveness of the self-delegation critique of Auer boils down to the fact that it paints an intuitively plausi­ble picture of agency self-dealing that stands so long as no one puts for­ward any evidence.

The main obstacle to advancing the debate is a lack of evidence. As Professor Steve Johnson has written, we have two choices: “[A]ware of the possibility of abuse described by Manning and Scalia, we could disre­gard [Auer] deference entirely because of the possibility, or we could stay our hand until a convincing record has been established that the possibil­ity turns into actuality with sufficient frequency and consequence.” 167 Steve R. Johnson, Auer/Seminole Rock Deference in the Tax Court, 11 Pitt. Tax. Rev. 1, 40 n.213 (2013). In what follows, the Article turns to the project of filling out our understand­ing of Auer’s effects on agency rule writing—particularly, examin­ing whether Auer encourages an increase in the vagueness of agency rules.

III. Testing the Self-Delegation Thesis

Auer’s future would seem to hinge in no small part on what the evi­dence actually says about the potentially perverse incentives created by the doctrine. This Part makes an in-depth and systematic effort to assess whether Auer’s incentives do manifest in changed rule-writing behavior. In examining whether Auer has changed the ways agencies write rules, this Part draws on several different strands of textual-analysis methods used in computational linguistics, psycholinguistics, political science, and regulatory studies. Public law scholars are increasingly turning to corpus linguistics methods to bring systemic empirical evidence to bear on large bodies of legal text. 168 See, e.g., Thomas R. Lee & Stephen C. Mouritsen, Judging Ordinary Meaning, 127 Yale L.J. 788, 830–36 (2018) (detailing the range of corpus linguistic tools available to study “ordinary meaning”). In order to test the key predictions of the self-delega­tion thesis, it is necessary to use these kinds of sophisticated linguis­tic techniques to analyze over time the work product of agencies—that is, the text of their regulations.

A. Data

This section reviews the original data I collected to test the self-delega­tion thesis, including the sample of rules used for analysis and the specific measures of textual vagueness employed.

1. Sample. — Seeking to capture as broad a sample of regulatory texts as possible, I gathered every “economically significant” 169 Technically, economically significant rules are those identified by the agencies as having “an annual effect on the economy of $100 million or more or adversely affect[ing] in a material way the economy, a sector of the economy, productivity, competition, jobs, the environment, public health or safety, or State, local or tribal governments or communities.” Maeve P. Carey, Cong. Research Serv., R43056, Counting Regulations: An Overview of Rulemaking, Types of Federal Regulations, and Pages in the Federal Register 10–11 (2016), []. This definition originated in Executive Order 12,866 in 1993 and was the trigger for the requirement of a full regulatory impact analysis. See Regulatory Planning and Review, Exec. Order No. 12,866, 58 Fed. Reg. 51,735, 51,738 (Sept. 30, 1993). Before Executive Order 12,866, the trigger was a slightly broader category of “major” rules, of which one subset was what would today be considered economically significant rules. See Steven Croley, White House Review of Agency Rulemaking: An Empirical Investigation, 70 U. Chi. L. Rev. 821, 825 (2003) (citing Exec. Order No. 12,291, 3 C.F.R. § 128 (1981)). The Office of Information and Regulatory Affairs has logged in historical reports the reviews of economically significant rules back to 1981; these reports are what I used to identify a sample of rules. See infra note 170. final rule reviewed by the White House Office of Information and Regulatory Affairs (OIRA) and published in the Federal Register between 1982 and 2016. 170 These data were assembled in a two-step process, one involving rule identification and one involving text collection and cleaning. First, rules were identified using XML reports of data on rule reviews conducted by OIRA. These data are housed on OIRA’s website and go back to 1981. See Office of Information and Regulatory Affairs, XML Reports,, [] (last visited Sept. 12, 2018). A research assistant extracted all rules that fit certain criteria: Principally, to be included, an entry had to be a final rule, listed as economically significant, and listed as having been published. Almost all of the rules listed in the reviews conducted in 1981 were not published, so I made the decision to exclude 1981 from the analysis. For the rest of the years available, most reviews resulted in publication, and they were therefore included. Second, using data culled from the XML reports, I was able to manually identify the Federal Register notice publishing the rule for all but a handful of the rules in the sample. Using Lexis Advance’s Federal Register Library, I then bulk downloaded the entire text of the notice for each of the iden­tified rules. In all, there were 1,218 such rules published by twenty-eight differ­ent departments, agencies, and boards during the observation period. These entities ranged from the familiar—for example, the Environmental Protection Agency (EPA), the Department of Transportation (DOT), the Department of Health and Human Services (HHS), the Department of Agriculture (USDA), and the like—to the less familiar—for example, the Architectural and Transportation Barriers Compliance Board and the Emergency Oil and Gas Guaranteed Loan Board. 171 This sample excludes independent agencies, such as the Federal Energy Regulatory Commission, SEC, and FCC, because they are not subject to OIRA review, and therefore, their rules were not available in the dataset derived from The textual data assembled represent a broad swath of the adminis­trative state and capture the bulk of the important rulemaking activity con­ducted by it. 172 One might object that economically significant rules are critically different from other rules and that focusing only on these rules risks introducing sampling bias insofar as the goal is to say something about rulemaking writ large. This concern is addressed in section III.F.2.

2. Measures of Vagueness. — For each rule in the sample, I first parti­tioned the document into the preamble and the rule text. 173 The process in note 170 supra actually identified 1,318 rules, but some of the Federal Register notices associated with these rules failed to provide regulatory text distinct from the preamble. As such, they were excluded from the final analysis. I then sub­jected each rule’s text to computer-assisted content analysis to generate a number of measures of vagueness for each text. 174 For three of the measures of vagueness (legal vagueness, laxity, and cognitive complexity), I used the program Linguistic Inquiry and Word Count (LIWC), which allows the user to apply both built-in and user-provided dictionaries to selected texts, counting the instances of each dictionary term and generating statistics on the percentage of the document composed of terms in the dictionary. See Yla R. Tausczik & James W. Pennebaker, The Psychological Meaning of Words: LIWC and Computerized Text Analysis Methods, 29 J. Language & Soc. Psychol. 24, 25–28 (2010) (reviewing the history of LIWC and describing the software’s processing and dictionary features); see also James W. Pennebaker et al., Univ. of Tex. at Austin, The Development of Psychometric Properties of LIWC2015 2–7 (2015), (on file with the Columbia Law Review) (explaining how the latest version of the LIWC software analyzes words and phrases based on psychometric categories assigned to words in built-in or user-generated dictionaries). For the final measure (polysemy), I used Python to analyze the textual data. See generally Python Frequently Asked Questions, Python Software Found., [] (last visited Sept. 19, 2018). Relative to human cod­ing, computer-assisted content analysis is faster, cheaper, and orders of magnitude more reliable. 175 See Justin Grimmer & Brandon M. Stewart, Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts, 21 Pol. Analysis 267, 267–69 (2013) (discussing the methodology of automated language analysis and its advantages in validity); David S. Law, Constitutional Archetypes, 95 Tex. L. Rev. 153, 192 (2016) (same). Human coders tire and make computa­tional mistakes; computers do not. Thus, with over 1,200 rules and over 19 million words, computer-assisted content analysis is by far the best approach to measuring the vagueness of texts, although substantial care has to be taken to ensure that what the computer is measuring captures what we mean when we say that a regulatory text is vague. 176 See infra note 179 and accompanying text. This section describes the steps I took—including the deployment of multiple measures and validation with human readers—to provide that assurance.

The concern with Auer is a concern about an agency’s propensity to promulgate what the D.C. Circuit has called “mush.” 177 Paralyzed Veterans of Am. v. D.C. Arena L.P., 117 F.3d 579, 584 (D.C. Cir. 1997) (“It is certainly not open to an agency to promulgate mush and then give it concrete form only through subsequent less formal ‘interpretations.’”). Specifically, the concern is with language that is cast with such a degree of vagueness that it fails to meaningfully constrain or guide subsequent interpretation, giving agencies more room within which to operate. As the concept of “vague­ness” might be conceived in different ways, 178 Vagueness is generally defined as the “use [of] concepts that have indefinite application to particular cases,” Lawrence Solum, Legal Theory Lexicon: Vagueness & Ambiguity, Legal Theory Blog (June 28, 2015), [], and it usually is sharply distinguished from ambiguity by linguists, see, e.g., George Dunbar, Towards a Cognitive Analysis of Polysemy, Ambiguity, and Vagueness, 12 Cognitive Linguistics 1, 2 (2001) (“A word is vague with respect to an interpretative element of meaning if it does not supply it; it is ambiguous between two elements if it supplies one or the other.”). Moreover, linguists often describe vagueness as having specialized classes, including soritical vagueness, combinatorial vagueness, and pragmatic vagueness. See Geert Keil & Ralf Poscher, Vagueness and Law: Philosophical and Legal Perspectives, in Vagueness and Law: Philosophical and Legal Perspectives 1, 1–2 (Geert Keil & Ralf Poscher eds., 2016) [hereinafter Vagueness and Law] (discussing different categories of vagueness and the legal implications of their distinctions). to provide a robust assessment of the self-delegation thesis, I selected several complementary measures that serve as proxies for vagueness. My four measures—legal vagueness, laxity, cognitive complexity, and poly­semy—are described in the following paragraphs. In the empirical analy­sis, each measure is scaled so that greater values equate to greater vague­ness, and lower values equate to greater clarity. The discussion of each of these measures of vagueness provides examples from a validation exer­cise using real regulatory texts in Title 21 of the Code of Federal Regulations—governing food labeling and packaging—and human cod­ers’ evaluations of the vagueness of those texts. 179 I tested the validity of the measures by sampling sentences from Title 21, chapter 1, subchapter B, identifying the sentences that registered as extreme (both high and low) on each of my four measures, and giving those sentences to twelve human coders to rate on a seven-point Likert Scale for overall vagueness. A statistical analysis showed that raters were able to systematically distinguish between high-scoring sentences and low-scoring sentences, which in turn suggests that the measures are capturing much of what human readers consider to be vagueness. In addition, the fact that preambles scored significantly higher on average than the associated rule texts for three of the four measures (and basically identically in the fourth) reinforces the validity findings. After all, in Justice Thomas’s formulation of the self-delegation thesis, the problem is that rule texts affected by the incentives read more like preambles than rules. See Thomas Jefferson Univ. v. Shalala, 512 U.S. 504, 518–19 (1994) (Thomas, J., dissenting) (outlining his concern about affording deference to “precatory” language that “reads more like a preamble than a law”).

a. Legal Vagueness. — The first measure is an index based on a list of commonly used legal terms that fail to convey much practical guidance. Phrases like “reasonable precautions” and “prudent investor” use terms that do not by themselves convey criteria capable of deciding concrete cases. When these kinds of expressions appear in regulations, they add little precision and thus are prime candidates for identifying instances of an agency “enacting placeholder regulations and doing the real policymak­ing work in subsequent so-called interpretations.” 180 Stephenson & Pogoriler, supra note 7, at 1468–69. Indeed, Stephenson and Pogoriler use the example of an agency “announcing that regulated entities must behave ‘appropriately’” to show how agencies might use inherently vague terms to increase their discretion. Id. “Appropriate” is one of the terms included in the “vagueness” dictionary in this Article. See infra note 188. In Shalala v. Guernsey Memorial Hospital, for example, the Supreme Court dealt with vague language of this kind. 181 See 514 U.S. 87, 92 (1995) (noting that the case turned, in part, on the meaning of the vague phrase “[s]tandardized definitions, accounting, statistics, and reporting practices” (alteration in original) (internal quotation marks omitted) (quoting 42 C.F.R. § 413.20(a) (1994))). The question in that case hinged on the Secretary of Health and Human Services’ interpretation of a regulation outlining “principles of cost reimbursement” 182 Id. (quoting 42 C.F.R. § 413.20(a)). in the Medicare pro­gram, and whether it permitted the Secretary to depart from generally accepted accounting principles (GAAP) in determining reimbursement for “reasonable costs.” 183 Id. at 91 (internal quotation marks omitted) (quoting 42 U.S.C. § 1395x(v)(1)(A) (1994)). The Secretary maintained that the regulation only described the recordkeeping steps providers needed to take and remained silent on whether the Secretary could depart from GAAP. 184 Id. at 92–93. The Supreme Court seized on language from the relevant regulation provid­ing that “the methods of determining costs payable under Medicare involve making use of data available from the institution’s basis accounts, as usually maintained, to arrive at equitable and proper payment for services to beneficiaries” 185 Id. at 92 (emphasis added) (internal quotation marks omitted) (quoting 42 C.F.R. § 413.20(a)). to conclude that the Secretary’s interpreta­tion was correct, and that “a provider’s basic financial infor­mation is organized according to GAAP as a beginning point” from which the Secretary can make an equitable and proper reimbursement decision on grounds other than GAAP. 186 Id. at 92–93. Manning’s influential article in fact highlighted the vagueness in this regulation as a prime example of Auer’s alleged perverse incentives for rule writing. 187 Manning, Constitutional Structure, supra note 113, at 657–60.

While it would probably be difficult to capture all of the words and phrases in the law that might fit the billing of legal vagueness, I con­structed a core dictionary of paradigmatically vague terms, such as reasonable, prudent, appropriate, and feasible. 188 I constructed this list by reading a number of statutes and noting words that seemed inherently vague, given some level of legal sophistication. The complete list of words used in the legal vagueness measure is as follows: reason*, prudent, best, available, possible*, optimal *, appropriate*, feasible*, acceptable, unreason*, careful *, proper*, undue*, unavailable, impossible*, infeasible*, unacceptable*, caution. Asterisks indicate that I included all derivatives from suffixation (for example, appropriate*ly). Using the program Linguistic Inquiry and Word Count (LIWC), I then calculated the total percentage of the words in each document falling in the dictionary. The resulting percentage is the measure referred to as legal vagueness in my analysis.

The measure is quite effective at distinguishing high-vagueness text from low-vagueness text. An example of a regulatory sentence that scored extremely high on this measure reads as follows: “Reasonable deficien­cies of calories, total sugars, added sugars, total fat, saturated fat, trans fat, cholesterol, or sodium under labeled amounts are acceptable within cur­rent good manufacturing practice.” 189 21 C.F.R. § 101.9(g)(6) (2018). In contrast, the following sentence scored extremely low on the measure: “The calorie declaration for a pack­aged food must include the total calories present in the packaged food, regardless of whether the packaged food contains a single serving or multiple servings.” 190 Id. § 101.8(c)(2)(C).

b. Laxity. — The second measure relies on the distinction between legal terms that are inherently binding and those that are inherently lax. 191 To some extent, the resulting measure is related to the Mercatus Center’s RegData measures of regulatory constraint, see Omar Al-Ubaydli & Patrick A. McLaughlin, RegData: A Numerical Database on Industry-Specific Regulations for All United States Industries and Federal Regulations, 1997–2012, 11 Reg. & Gov. 109, 112 (2017), but it was necessary to construct my own measure because the RegData are only available for all regulations that have already been promulgated and incorporated into the Code of Federal Regulations. My sample identification process thus precluded use of RegData. For instance, words like shall and must indicate a clear command; regulated parties ignore these kinds of commands at their peril. But at the same time, words like may and should leave regulated parties with greater uncertainty and leave law enforcers with a greater degree of discre­tion. Capturing the balance of these two categories of terms in a given legal text could reveal just how clear a regulator has been about the legal duty the text imposes. Lest a regulator do serious damage to its pro­grams, it will most likely not issue a stern command unless it has itself confirmed that the accompanying requirements are spelled out to cap­ture precisely the conduct that the regulator means to capture.

To construct the laxity measure, I started by developing a simple list of common permissive legal terms and a list of common compulsory legal terms and ran the dictionary on the rule texts as with the measure above. 192 The complete list of words in the “compulsory” dictionary is as follows: must, shall, will, cannot, never. The complete list of words in the “permissive” dictionary is as follows: could, might, can, probably, may, should. I then subtracted the percentage of compulsory terms in a text from the percentage of permissive terms in a text to generate an aggre­gate index for laxity that measures the degree to which a given text fails to give relatively binding instructions. Again, the measure of laxity seems to capture intuitive distinctions between sentences. An example of a high-laxity sentence reads: “Such representations may be made either by statements, photographs, or vignettes.” 193 21 C.F.R. § 101.13(l)(2). By contrast, an example of a low-laxity sentence is the following: “A common fraction shall be reduced to its lowest terms; a decimal fraction shall not be carried out to more than two places.” 194 Id. § 101.7(d).

c. Cognitive Complexity. — Psychologists have argued that writing can be scaled in terms of its degree of “cognitive complexity.” 195 See Tausczik & Pennebaker, supra note 174, at 35 (“Cognitive complexity can be thought of as a richness of two components of reasoning: the extent to which someone differentiates between multiple competing solutions and the extent to which someone integrates among solutions.”). For the third measure, I employed a widely used measure of cognitive complexity 196 I followed the lead of a number of recent studies of Supreme Court opinion writing in employing the measure of cognitive complexity. See, e.g., Pamela Corley & Justin Wedeking, The (Dis)Advantage of Certainty: The Importance of Certainty in Language, 48 Law & Soc’y Rev. 35, 43 (2014) (using the LIWC program to measure the degree of certainty in Supreme Court majority opinions); Frank B. Cross & James W. Pennebaker, The Language of the Roberts Court, 2014 Mich. St. L. Rev. 853, 866 (using the LIWC program to analyze cognitive processes in opinions from the first five years of the Roberts Court); Deborah H. Gruenfeld, Status, Ideology, and Integrative Complexity on the U.S. Supreme Court: Rethinking the Politics of Political Decision Making, 68 J. Personality & Soc. Psychol. 5, 9–10 (1995) (comparing cognitive complexity in majority and dissenting Supreme Court decisions and in decisions written by liberal and conser­vative Justices); Ryan J. Owens & Justin P. Wedeking, Justices and Legal Clarity: Analyzing the Complexity of Supreme Court Opinions, 45 Law & Soc’y Rev. 1027, 1039–40 (2011) [hereinafter Owens & Wedeking, Justices and Legal Clarity] (describing a method for measuring cognitive complexity in Supreme Court opinions using the LIWC program); Ryan J. Owens & Justin Wedeking, Predicting Drift on Politically Insulated Institutions: A Study of Ideological Drift on the United States Supreme Court, 74 J. Pol. 487, 491–93 (2012) [hereinafter Owens & Wedeking, Predicting Drift] (measuring the cognitive complexity of Supreme Court Justices over their lifetime, both before and after being nominated to the Court, using a LIWC program); Philip E. Tetlock, Jane Bernzweig & Jack L. Gallant, Supreme Court Decision Making: Cognitive Style as a Predictor of Ideological Consistency of Voting, 48 J. Personality & Soc. Psychol. 1227, 1229–31 (1985) (describing the linguistic analysis required to measure integrative complexity in Supreme Court opinions). For a useful summary of this literature, see Ryan C. Black et al., U.S. Supreme Court Opinions and Their Audiences 40–46 (2016).
The built-in dictionaries I used, following the literature above, contained categories of words that are theorized to be proxies for cognitive complexity. These dictionaries include: causation words (for example, because, effect, hence, and depend), insight words (for example, think, know, and consider), discrepancy words (for example, should, would, and could), tentativeness words (for example, maybe, perhaps, and fairly), certainty words (for example, always, never, and absolutely), difference words (for example, hasn’t, but, and else), negation words (for example, no, not, and never), and six-letter words (that is, words with more than six letters, which are thought to be correlated with sophistication and complex­ity). After obtaining raw percentage scores for each of these categories in each text, I standardized the scores by obtaining z-scores for each category in each text. The z-score is simply z = , where x is the observed raw score, m is the mean score for all texts, and s is the standard deviation. Finally, again following the method outlined in prior studies, see Owens & Wedeking, Justices and Legal Clarity, supra, at 1039–40, I summed the standardized scores that indicated greater cognitive complexity and subtracted the stan­dardized scores associated with simplicity (that is, certainty and negation words) to create an aggregate measure of cognitive complexity for each text in the dataset.
on the assumption that regulations may be more malleable when they implic­itly (or explicitly) acknowledge different layers of a policy prob­lem. These layers might provide the policy justification for an interpre­tive change at a later time. Indeed, cognitive complexity seems to have been the focus of Justice Thomas’s concerns in Thomas Jefferson University v. Shalala, in which he lamented that a rule that he asserted had succumbed to the perverse incentive thesis looked more like preamble text than law proper. 197 512 U.S. 504, 519 (1994) (Thomas, J., dissenting). Preambles contain more of the justification for, and purposes behind, a rule, often highlighting competing consider­ations at play in the agency’s thinking, responding to various critiques of a proposed action, and seeking to build a complete record for judicial review. They are inherently cognitively complex.

Cognitive complexity, therefore, is well suited to capture at least this dimen­sion of vagueness in rule texts. Rule texts that score high on cogni­tive complexity tend to use “multidimensional terms and to integrate a variety of evidence,” 198 Tetlock, Bernzweig & Gallant, supra note 196, at 1228. More specifically, cognitive complexity consists of two related concepts called “differentiation” and “integration.” The former “represents the degree to which an individual can see multiple perspectives or dimensions in an issue,” and the latter indicates “the degree to which a person recognizes relationships and connections among these perspectives or dimensions and integrates them into their decision or judgment.” Black et al., supra note 196, at 44. while low cognitive complexity denotes the kind of “conceptual organization of decision-relevant information” charac­teris­tic of codified legal text. 199 Gruenfeld, supra note 196, at 5. There are obviously important parallels with the literature on rules versus standards. See, e.g., Louis Kaplow, Rules Versus Standards: An Economic Analysis, 42 Duke L.J. 557, 586–90 (1992) (discussing the varying degrees of complexity that can be written into rules and standards); Russell B. Korobkin, Behavioral Analysis and Legal Form: Rules vs. Standards Revisited, 79 Or. L. Rev. 23, 32 (2000) (noting that dispute resolution through standards is usually, but not always, more complex than dispute resolution through rules). Indeed, Owens and Wedeking find that Justice Scalia, a well-known enthusiast of rules, wrote opinions that displayed comparatively low scores on measures of cognitive complexity. Owens & Wedeking, Justices and Legal Clarity, supra note 196, at 1044. When an agency writes in a cognitively complex style, it is in effect leaving itself room to emphasize different facts or different considerations in the next round of decisionmaking. For instance, Professors Ryan Owens and Justin Wedeking find that mea­sures of cognitive complexity in the opinions authored by Supreme Court Justices predict ideological drift over the course of their careers. 200 See Owens & Wedeking, Predicting Drift, supra note 196, at 498–99. This is precisely what critics of Auer see agencies doing—using under­specification in the first round of play to sandbag for greater author­ity or discretion in future situations. In other words, cognitive com­plexity proxies for the capacity for plausible drift.

The expectation is that prose low in cognitive complexity will appear crisp and simple, whereas prose high in cognitive complexity will read as muddled and complicated. Indeed, an example of a high-scoring sentence on the measure reads: “Fat and/or oil ingredients not present in the product may be listed if they may sometimes be used in the prod­uct.” 201 21 C.F.R. § 101.4(b)(14) (2018). In contrast, a low-scoring example is: “Label serving size for ice cream cones, eggs, and breath mints of all sizes will be 1 unit.” 202 Id. § 101.12 tbl.2 & n.8. The first sentence takes time to process because it involves contingencies, whereas the second sentence is a straightforward command.

d. Polysemy. — A final measure relates to what philosophers of lan­guage call “combinatorial vagueness,” which arises when a speaker or writer “ha[s] not formed an intention with regard to the inclusion or exclusion of certain objects.” 203 See Ralf Poscher, An Intentionalist Account of Vagueness: A Legal Perpsective, in Vagueness and Law, supra note 178, at 65, 85. This kind of categorization problem is likely the source of the lion’s share of disputes about federal statutes, and it extends well beyond the kind of paradigmatic legal vagueness discussed above. 204 See Lawrence M. Solan, The Language of Statutes: Laws and Their Interpretation 35 (2010) (“Cases involving linguistic ambiguity are far less common than those involving conceptual vagueness . . . .”). For instance, courts have had to grapple with such questions as whether a statute defining “motor vehicle” to include “an automobile, automobile truck, automobile wagon, motorcycle, or any other self-pro­pelled vehicle not designed for running on rails” includes airplanes. 205 McBoyle v. United States, 283 U.S. 25, 25–26 (1931) (quoting National Motor Vehicle Theft Act, ch. 89, § 2(a), 41 Stat. 324, 324 (1919) (current version at 18 U.S.C. § 2311 (2012))). They have also had to grapple with whether the Voting Rights Act’s use of the term “elect[ed] representatives” includes judges who are elected. 206 Chisom v. Roemer, 501 U.S. 380, 385 (1991) (quoting Voting Rights Act Amendments of 1982, Pub. L. No. 97-205, sec. 3, § 2(b), 96 Stat. 131, 134 (current version at 52 U.S.C. § 10301 (2012))). In essence, vagueness arises when a legal text chooses a word that is gen­eral enough to elide costly determinations about what is and what is not included.

In an effort to account for this kind of vagueness in the rules in the sam­ple, I employed the concepts of hyponymy and hypernymy, which provide a way of hierarchically mapping the relationships between words on the dimension of specificity. 207 See George A. Miller, WordNet: A Lexical Database for English, Comm. of the ACM, Nov. 1995, at 39, 40,
48b536b78bfeafba.pdf (on file with the Columbia Law Review) (defining and explaining “hyponymy” and “hypernymy”).
A hyponym is a subordinate of a hyper­nym: For instance, the word orange is a hyponym of fruit, and fruit is a hyponym of food. Conversely, a hypernym is a superordinate of another word. 208 Id. Thus, the word fruit is a hypernym of orange, but so is color. These concepts, when applied to the words that are actually used in a text, capture quite cleanly at least one key dimension of what concerns Auer’s crit­ics, which is systematic evasiveness in providing the details of what trig­gers a legal obligation. If the count of hyponyms associated with the words in a text is relatively large (that is, the words actually used are hyper­nyms of a relatively high number of hyponyms), this suggests for­gone specificity, because a “hyponym inherits all the features of the more generic concept and adds at least one feature that distinguishes it from its superordinate and from any other hyponyms of that superordinate.” 209 George A. Miller et al., Introduction to WordNet: An On-line Lexical Database 8 (1993), []. If an agency is withholding clarity strategically, we would expect that agen­cies would choose more words that decline to specify that addi­tional, clarifying feature. Likewise, a high count of hypernyms could cre­ate vagueness because a large number of superordinate concepts associ­ated with a given word would require the reader to differentiate that word’s true subordinate relationship from those distinct concepts. Of course, hypernymy is also associated with specificity, as the agency could have used a more overarching term. The end goal is to use the concepts of hypernymy and hyponymy to place each word in a hierarchy of more general terms to more specific terms. In essence, the higher a word is in the hierarchy—and the greater the number of distinct words below it—the more potentially vague the word is.

Using a lexical database called WordNet, 210 WordNet, [] (last visited Sept. 14, 2018). For background on WordNet and its potential applications, see gen­erally Richard Beckwith et al., WordNet: A Lexical Database Organized on Psycholinguistic Principles, in Lexical Acquisition: Exploiting On-Line Resources to Build a Lexicon 211 (Uri Zernik ed., 1991). which is widely used in natu­ral language processing and artificial intelligence, I was able to count the number of hyponyms and hypernyms linked to each word in the sample rules and generate document-level averages of these counts. 211 WordNet uses a Directed Acyclic Graph (DAG) to hierarchically graph the links between words in terms of hypernymy and hyponymy. For a visual example of a WordNet DAG, see Beckwith et al., supra note 210, at 216 fig.9.1. The result generated two measures of possible vagueness—nondis­tinction of subordinate concepts (that is, average rate of hypo­nyms for the words in a given text) and nondistinction of entirely different word senses, or lexical ambiguity (that is, average rate of hypernyms for a given text)—which I then added together to form an index labeled “polysemy.” Again, examples from the FDA’s food-labeling regulations help demon­strate what distinguishes text under this measure. A low-scor­ing sentence reads: “The gram (mL) quantity between 2 and 5 g (mL) should be rounded to the nearest .05 g (mL) and the g (mL) quantity less than 2 g (mL) should be expressed in .01-g (mL) increments.” 212 21 C.F.R. § 101.9(b)(7)(ii) (2018). A high-scoring example stands in sharp contrast for its verbosity: “Sub­stance means a specific food or component of food, regardless of whether the food is in conventional food form or a dietary supplement that includes vitamins, minerals, herbs, or other similar nutritional sub­stances.” 213 Id. § 101.14(a)(2). This latter sentence employs general categories, such as food, minerals, and herbs, that leave the agency with ample room to make argu­ments that a particular item is really outside the ambit of the overarch­ing category. The polysemy measure aims to pick up differences between regulatory texts on just this tendency to employ broad catego­ries instead of specific terms.

B. Aggregate Trends

My starting point for testing whether Auer’s incentives translate into effects on agency behavior was to look at aggregate trends on the core measures of linguistic “mush” described in section III.A.2. Figure 3 pre­sents the time series trends of the median values of each of the four vague­ness measures for rules from 1982 through 2016. If there is a sys­temic perverse effect of the doctrine on rule writing, there should be a generally positive trend on the measures of vagueness following Auer, given that positive citations to Auer have continued to grow over time. At the very least, if Auer created strong perverse incentives, there should be some indication that the post-Auer period has seen higher levels of vague­ness than the pre-Auer period.

Even at this general level, it is clear that there are some difficulties for the self-delegation thesis. First, for the entire sample, only two of the measures show statistically significant trends over time. Legal vagueness decreased by an average of .004 per year from 1982 through 2016 (p>|t| = .000), while laxity increased by an average of .005 per year over the same time period (p>|t| = .048). For the other two measures, agencies have been static in terms of the specificity of their writing over a nearly forty-year period that saw the rise of Auer deference. 214 See supra section I.A & Figure 1 (describing and showing the increasing promi­nence of Auer and Seminole Rock).

Figure 3: Standardized Indices of Rule-Text Vagueness, 1982–2016  215 The solid lines represent median values, shown as z-scores, for each year on the selected index across all agencies. The vertical dashed lines show the year 1997, when Auer was decided.


Walters Figure 3

Second, none of the four measures of vagueness show any statisti­cally significant trend when looking only at rules published in 1997 or later. Two measures—cognitive complexity and polysemy—show increases in the propensity for vagueness but fall far short of statistical significance (p>|t| = .587 and .926, respectively) during this period. The other two measures—legal vagueness and laxity—indicate decreases in vagueness since 1997, even though those changes also fall short of statisti­cal significance (p>|t| = .184 and .115, respectively). In contrast, the period before Auer did see statistically significant movement on at least one of these variables. From 1982 through 1996, agencies’ rules trended downward (statistically significantly so, at the .01 level) on the measure of legal vagueness. Two of the other variables—laxity and cognitive complex­ity—showed a negative trend, but not statistically distinguishable from zero. Polysemy trended upward but, again, not at a statistically signifi­cant rate.

Table 1: Difference of Means, Pre- and Post-Auer 216 P-values are from a two-tailed test of the null hypothesis that the difference of means is zero. Statistical significance is denoted as follows: p<.1 = ^, p<.05 = *, p<.01 = **, p<.001 = ***. Statistically significant p-values indicate that the null hypothesis of zero difference can be rejected in favor of the alternative hypothesis that there was a significant change not attributable to natural variation.

Table 1

In terms of simple differences in means between the two periods, most of the data are, again, inconsistent with the self-delegation thesis. As Table 1 shows, the post-Auer mean for two of the measures—legal vague­ness and polysemy—was significantly lower than the pre-Auer mean. The difference with cognitive complexity is not statistically distinguishable from zero. Only one piece of evidence possibly squares with the self-delega­tion thesis’s basic predictions: Laxity does show a statistically signi­fi­cant increase post-Auer.

While this comparison of means offers just an initial overview of the data, it is a highly suggestive one. Most of the evidence suggests that the post-Auer period has been marked by abnormally precise regulatory lan­guage or that it has not been substantially different than it was before Auer. 217 The analysis to this point does not allow one to say whether the post-Auer rules changed the trend, or slope, relative to the pre-Auer trend. This possibility is explored in greater detail in sections III.D and III.E using interrupted time series methods.

C. Agency-Specific Trends

As informative as aggregate trends can be, collapsing the data into one time series for all agencies sacrifices much of the interesting varia­tion. Although there is significant coordination between agencies in rulemak­ing, 218 See Jody Freeman & Jim Rossi, Agency Coordination in Shared Regulatory Space, 125 Harv. L. Rev. 1131, 1136–37 (2012) (discussing efforts to promote coordination across government agencies). agencies are largely independent of one another and come to the rule-writing task with drastically different baselines. 219 Some agencies may deal with subject matter that is inherently more vague compared to the matters others deal with (imagine, for instance, the difference between technical air pollution regulations and antidiscrimination regulations), and some variation might also exist because of the level of vagueness in the governing statutes that the regulations seek to implement. Consequently there are significant differences in patterns in the data across agencies. For instance, there is great variation in the frequency with which each of the agencies produced rules. Only DOL, DOT, EPA, HHS, and USDA published more than twenty economically significant final rules during the period of observation. That means the rest of the twenty-three agencies that published at least one rule during this entire period averaged less than one final rule per year—and often much less than that.

If agencies produce rules at starkly different rates, it might also stand to reason that agencies would produce rules in very different ways. That intuition is supported even when focusing just on agencies that produce larger numbers of rules annually. Figure 4 plots the measures of vague­ness used for this study for the six agencies—the Department of the Interior (DOI), DOL, DOT, EPA, HHS, and USDA—that produced the greatest number of economically significant final rules. The measures are standardized (adjusted so that they all have means of zero and standard deviations of one) for ease of presentation. As Figure 4 makes clear, the agencies vary significantly in terms of the vagueness of their rules, and the patterns do not necessarily correspond to the aggregate trends dis­cussed in section III.B.

This heterogeneity is both a feature and a problem from the stand­point of studying Auer’s incentives and effects. On the positive side, it is possible to see a more nuanced picture of how individual agencies might have responded to Auer. For instance, Figure 4 shows that, on the whole, EPA and HHS have more or less steadily increased the clarity of their rule texts over time. In contrast, DOL, DOT, and USDA show more vola­tile patterns, with major upticks and downticks in vagueness both before and after Auer. Statistically speaking, the data suggest that there may not be a one-size-fits-all determinant of rule vagueness across agencies, and any statistical analysis should control for agency differences.

Figure 4: Standardized Indices of Vagueness by Major Rulemaking Agency, 1982–2016  220 Each panel displays agency-specific median values, shown as z-scores, for all four measures of vagueness. The vertical dashed lines show the year 1997, when Auer was decided.

Walters Figure 4

That said, the variation in these data suggests a problem because the self-delegation thesis posits a systematic effect that, in theory, should apply across the board to all agencies. If certain agencies have highly unique trends, outlier values, or just more rules than other agencies, statisti­cal methods that do not take account of agency-level effects will be biased. Fortunately, it is possible to take account of this variation by exploit­ing the panel structure of the data. 221 Most empirical research on administrative behavior thus includes fixed effects at the agency level. See O’Connell, Political Cycles of Rulemaking, supra note 131, at 932, 940. Fixed effects allow one to “control for all stable characteristics of the [entities] in the study,” even without measuring these characteristics, by “completely ignor[ing] the between-[entity] variation and focus[ing] only on the within-[entity] variation.” Paul D. Allison, Fixed Effects Regression Methods for Longitudinal Data Using SAS 3–4 (2005). Thus, each agency serves as its own control. The next two sections turn to statistical tests to determine whether, accounting for this agency-level heterogeneity, there is any effect from Auer.

D. Searching for a Link Between Incentives and Effect in Auer

In order to test more fully whether Auer might have had the effect of encour­aging agencies to write rules more vaguely, I used a quasi-experi­mental method called an interrupted time series (ITS) design. 222 See generally Andrea Cann Chandrasekher, Empirically Validating the Police Liability Insurance Claim, 130 Harv. L. Rev. Forum 233, 239–40 (2017), [] (“With interrupted time series, the researcher compares the level of some outcome variable . . . right before and right after the imposition of some policy that affects the availability (or level) of the treatment . . . .”). An ITS design  allows me  to  measure  the pre-Auer trend, or slope, using ordinary least squares (OLS) regression, and then to estimate the  post-Auer  trend on the same variable. The measure of interest is the difference between the pre-Auer trend—which serves as a  counterfactual  prediction  of  what  would have occurred without  any  change  in  agency behavior — and the observed post-Auer trend. 223 See generally Ariel Linden, Conducting Interrupted Time-Series Analysis for Single – and Multiple-Group Comparisons,  15 Stata J. 480, 481–82 (2015)  (providing back­ground  on  the methodology and usefulness on  ITS analysis);  A.K. Wagner et al.,  Segmented Regression Analysis of Interrupted Time Series Studies in Medication Use Research, 27 J. Clinical Pharmacy & Therapeutics 299, 301 (2002) (same). This approach  has  been used to  address  treat­ment  effects  in  a  wide range  of  legal  and  policy fields,  from  criminal justice to education. 224 See, e.g., Paul G. Cassell & Richard Fowles, Handcuffing the Cops? A Thirty-Year Perspective on Miranda’s Harmful Effects on Law Enforcement, 50 Stan. L. Rev. 1055, 1071–74 (1998) (using interrupted time series analysis to determine if crime clearance rates declined after Miranda v. Arizona); Michael Heise, Equal Educational Opportunity, Hollow Victories, and the Demise of School Finance Equity Theory: An Empirical Perspective and Alternative Explanation, 32 Ga. L. Rev. 543, 602–04 (1998) (measuring change in education spending and centralization in school districts following school-finance litigation); Malcolm D. Holmes et al., Plea Bargaining Policy and State District Court Caseloads: An Interrupted Time Series Analysis, 26 Law & Soc’y Rev. 139, 143–45 (1992) (examining the relationship between plea bargaining policies and state court caseloads); Brian J. Ostrom et al., Congress, Courts and Corrections: An Empirical Perspective on the Prison Litigation Reform Act, 78 Notre Dame L. Rev. 1525, 1546 & n.71 (2003) (determining whether 28 U.S.C. § 1983 cases were filed less often in each circuit court as a result of passage of the Prison Litigation Reform Act); Jonathan W. Penney, Chilling Effects: Online Surveillance and Wikipedia Use, 31 Berkeley Tech. L.J. 117, 143–45 (2016) (exploring whether internet traffic to “privacy-sensitive” Wikipedia articles decreased following the June 2013 revelation of the NSA/PRISM surveillance program).

Table 2 presents the results of an ITS model using the entire time series and a single treatment period starting in 1997. For each core meas­ure of vagueness, Table 2 presents first the pre-Auer trend and then the change to that trend. Three separate specifications are reported. All three models include a fixed effect for the issuing department. Model 2 controls for the differences in baseline vagueness that might exist between different rulemakings with very different subject matter. For this, I used the preamble score for the corresponding measure of vague­ness. If, for example, a rule on the meaning of “waters of the United States” in the Clean Water Act deals with an inherently vague subject mat­ter, both the rule text and the preamble likely reflect that baseline subject-matter vagueness. By including the control for preambles, there is less risk that differences are driven by chance variation in the subject mat­ter that agencies choose (or are forced) to tackle. Finally, I included standard measures about the political environment in Model 3. Agencies might adjust their rule-writing strategy in response to their perceptions about the likelihood of political oversight or the preferences of their politi­cal principals. Thus, I included measures of whether the rule was promulgated during (a) divided or unified government and (b) a Republican or Democratic presidency.

The results in Table 2 provide no support for the hypothesis that Auer caused agencies to write rules that are more “mushy” than the norm. In fact, the strongest finding, from a statistical perspective, is that Auer seems to have encouraged agencies to write less vaguely than before, by one measure at least. During the post-Auer period, the average meas­ure of rule laxity decreased by between .31 and .40 per year relative to the pre-Auer trend—a total reversal for the agencies from their pre-Auer trend toward more vagueness, and one that is statistically significant. For the rest of the variables, the findings are not statistically significant. That is, while there are directional changes indicated—negative for polysemy and positive for legal vagueness and cognitive complexity—these findings are not statistically distinguishable from the null hypothesis of no change in trend.

Table 2: Interrupted Time Series Models of the Effect of Auer on Rule Vagueness 225 Estimates are OLS with Newey-West Standard Errors. Observations are at the department-year level (N=362). Statistical significance is denoted as follows: p<.1 = ^, p<.05 = *, p<.01 = **, p<.001 = ***.

Table 2

These findings are also robust to adding a one-year lag to the treat­ment. 226 Using a lag accounts for a possible delayed effect. I do not report the results because they are substantially similar to the results reported in Table 2. Even after account­ing for a possible delayed onset by adding a one-year lag, there are no statistically significant results. As an additional robustness check, I also tested for short-term effects on agencies that were only observable in the first few years immedi­ately following the decision. I used two windows of time: one with an observation period of two years before and after Auer and a second with an observation period of three years before and after Auer. For these two observation windows, I estimated the same regressions as in Table 2. Limiting the analysis to these windows, I found that none of the ITS esti­mates of post-Auer change were statistically significant. Thus, even if the hypothesized effect of Auer is merely a short-term effect, there is no evidence to sup­port it.

E. Testing the Effect of the Second Revolution

As a further effort to evaluate the self-delegation thesis, I tested whether developments since 2005—namely, the Supreme Court’s signifi­cant Auer-related decisions in Perez, Decker, Christopher, Talk America, and Gonzales 227 See supra notes 74–93 and accompanying text. —have had any effect on agency rule writing. It is not exactly clear as a matter of theory what kind of effect these developments might have had on agency rule writing. On one level, the growing apprehen­sion about Auer—as signaled by statements of disapproval from Chief Justice Roberts, Justices Thomas and Alito, and the late Justice Scalia 228 See supra notes 85–93 and accompanying text. —suggests that the perverse incentive might have lost some of its luster for agencies. If the Supreme Court is on the precipice of overturning Auer, any rule intentionally obscured now would have an uncertain payoff were there a change to the doctrine. At the same time, Auer is still good law. Moreover, as Figures 1 and 2 suggest, Auer has never been more popular in the circuit courts, as measured both by explicitly positive citations and by total citations. Additionally, given that some scholars have viewed recent decisions such as Perez 229 See supra note 93 and accompanying text. and the certiorari decisions in United Student Aid Funds and Gloucester County 230 See supra notes 96, 98 and accompanying text. as a signal that Auer was safe, at least temporarily, 231 See supra notes 96–100 and accompanying text. The Court’s decision to revisit Auer deference this term in Kisor v. Wilkie suggests that the doctrine might no longer be safe. See supra note 19 and accompanying text. there might be reason to suspect that Auer’s “second revolu­tion” had positively reinforced its perverse incentives.

Teasing out which, if any, of these hypotheses is correct involved apply­ing the same methodology from section III.D to these later treat­ment dates. As a first cut, and taking my cue from Figures 1 and 2, which show spikes in total citations, positive citations, and negative citations in 2005, I analyzed whether trends in agencies’ rule clarity changed appre­cia­bly relative to the pre-2005 trend. I also tested whether there was any change to the pre-2011 trend after the citations to Auer rose sharply in 2011. These dates roughly correspond to the litigation in Gonzales and Talk America, two significant cases in which several Justices expressed reservations about Auer. 232 See supra notes 74–84 and accompanying text. Table 3 presents the results of this ITS analysis.

Table 3: Interrupted Time Series Models of Recent Auer-Related Supreme Court Cases  233 Estimates are OLS with Newey-West Standard Errors. Observations are at the department-year level (N=362). Statistical significance is denoted as follows: p<.1 = ^, p<.05 = *, p<.01 = **, p<.001 = ***.

Table 3

The results in Table 3, again, provide no support for the self-delega­tion hypothesis. As with the analysis of Auer’s effect, both the post-2005 and the post-2011 periods show virtually no statistically significant changes in agency rule writing, despite the fact that these years marked the beginning of major upticks in citations to the doctrine. And, again, the one statistically significant finding that does stand out is the statisti­cally significant negative change in the trend for laxity post-2005.

The results in Table 3 show changes in trends relative to the entire pre-2005 and pre-2011 periods (that is, 1982 onward), and that is proba­bly the appropriate baseline, given that section III.D found that Auer had no effect. Nevertheless, as an additional robustness check, Table 4 pre­sents the results of models looking at just the segment of time between each of these salient treatment periods. That is, the models look at the effect on the preexisting trend from 1997 up to 2005, and then at the effect on the preexisting trend from 2005 up to 2011. The results in Table 4 provide no evidence in favor of either possible effect. On the whole, these data show that the recent flurry of cases has had no effect on agency rule writing one way or the other.

Table 4: Segmented Interrupted Time Series Results 234 Estimates are OLS with Newey-West Standard Errors. Observations are at the department-year level (N=362). Models 1 through 3 show change in trends from 2005 onward relative to the approximate period of time between Auer and Gonzales (1997–2005). Models 4 through 6 show change in trends from 2011 onward relative to the approximate period of time between Gonzales and Talk America (2005–2011). Statistical significance is denoted as follows: p<.1 = ^, p<.05 = *, p<.01 = **, p<.001 = ***.

Table 4
Arguably, agencies may still be waiting for the Court to make a move before committing to any particular strategy. But this seems unlikely given that the lack of any change in Table 4 simply fits the pattern of all the other analyses. Agencies’ rule clarity appears quite strikingly to remain unaffected by Auer, falling short of the expectations of the self-delegation thesis.

F. Other Robustness Checks

The most notable statistically significant finding from all of the anal­yses is the complete opposite of what the self-delegation thesis would have predicted—clarity over legal duty actually increased following Auer, as shown in Table 2. 235 Specifically, Table 2 reported a statistically significant shift in the trend line for the laxity measure after Auer. See supra Table 2. Otherwise, the findings from the ITS analysis are null. Although null results mean that the analysis has yielded evidence that the Auer effect is statistically indistinguishable from zero, they are not, by themselves, sufficient to infer that the effect of Auer is truly zero. There are thus limits to what can be inferred from this kind of evidence, but null results are not meaningless and ignoring them can be problem­atic. 236 See Robert Rosenthal, The “File Drawer Problem” and Tolerance for Null Results, 86 Psychol. Bull. 638, 640 (1979) (discussing the problem that statistically nonsignificant studies are not generally published, which could make erroneously significant studies appear more credible than they actually are). This is particularly the case when other attempts to analyze the problem also fail to find any effect.

To that end, I supplemented the main analysis with a number of robust­ness checks, and across the board the results are similar: Auer appears to be unassociated with any systematic trends in agency rule writ­ing. If numerous efforts to identify an effect fail to yield results, we can be more confident that the null results are meaningful in their own right. If nothing else, the burden of proof surely should shift decisively to adher­ents of the self-delegation thesis to put forward evidence in support of their assertions.

1. Alternate Treatment. — One potential concern might be that the ITS methodology employed above fails to capture a true treatment effect—that is, it might fail to capture the extent to which agencies were aware of and likely to act on Auer’s incentives. That seems unlikely, as the data in Part I indicated that Auer became widely cited and widely per­ceived only in 1997, 237 See supra Figures 1 & 2. but perhaps there are reasons to believe that an agency’s “hot hand” (that is, recent success in deploying Auer) is more likely to activate the perverse incentives than any kind of general aware­ness about Auer.

Using data collected by William Yeatman, I measured this hot-hand treatment by computing agency-level win rates and Auer-related litigation rates in the U.S. Courts of Appeals from 1993 through 2013. 238 Yeatman, supra note 36, at 523–45 (discussing the methodology employed in assembling the dataset). Because Yeatman’s data ran for only a subset of the years in my data, and because many agencies never experienced Auer-related litigation during this period, the models include only fifteen agencies. Moreover, because even agencies that did experience Auer-related litigation did not litigate every year, the data were further limited to just those agency-year pairs for which there was observable litigation. Do experi­ences with Auer deference in real-world litigation affect agency rule writ­ing down the road? In order to answer this question, I estimated models of the effect of Auer case win rates, total Auer cases per year, and their interaction on the indices of rule vagueness. Since rule writing takes time—and the effects of one year’s litigation successes would not likely appear until the next year at the earliest—I estimated the models with lags of one and two years. Table 5 reports the results.

Overall, the results show that Auer-related litigation successes do not produce increases in rule vagueness in subsequent years, or at least not increases that are statistically distinguishable from zero. Only two of the models—the two-year lagged models of cognitive complexity and poly­semy—show any statistically significant relationship, and only one of these (polysemy) is consistent with the self-delegation thesis’s prediction that greater exposure to Auer would trigger more vague­ness. Moreover, it is notable that the interaction of the agency’s win rate and the number of cases in a given year—which represents the strong­est possible measure of “hot-hand” exposure to Auer’s incentives—is almost uniformly negatively signed. That is, when agencies are winning a great deal of cases, they are, if anything, writing more clearly in the year or two after. While these findings on the interaction are not statistically significant, the negative sign is inconsistent with the basic intuition of the self-delegation thesis.

Table 5: The Effect of Auer-Related Litigation Success on Agency Rule Writing 239 Estimates are OLS with robust standard errors and agency-level fixed effects. “Lag 1” rows give coefficients from models estimated with one-year lags. “Lag 2” rows give coefficients from models estimated with two-year lags. Statistical significance is denoted as follows: p<.1 = ^, p<.05 = *, p<.01 = **, p<.001 = ***.

Table 5
2. Sample Bias. — There are major advantages to limiting the sam­ple to economically significant rules. These are the rules that most affect business and the public, 240 See Carey, supra note 169, at 12 (noting that economically significant rules might be of “higher salience and political importance”). and they are often new regulatory programs rather than amendments to existing programs. Both of these factors mean that there is greater potential for agencies to act on Auer’s perverse incentives—the agency is often writing on a blank slate in a policy arena that is almost certainly important enough that the agency will have to revisit it time and again. Beyond these substantive reasons for focusing on economically significant rules, identifying and collecting the entire population of economically significant rules for several decades is relatively tractable, given that the White House Office of Information and Regulatory Affairs posts these data on its website. 241 See supra notes 169–170 and accompanying text.

Notwithstanding the importance of rules deemed economically signifi­cant, they are admittedly only a subset of all the rules agencies promul­gate every year. 242 See Carey, supra note 169, at 11 tbl.4 (comparing the total number of OIRA reviews of economically significant rules and non-economically significant reviews from 1994 through 2015). One notable difference between these two catego­ries of rules is that economically significant rules are vetted extremely carefully relative to other significant rules because they require a Regulatory Impact Analysis quantifying, to the extent possible, the costs and benefits of both the regulation and its alternatives. 243 Office of Mgm’t & Budget, Exec. Office of the President, Circular A-4, Regulatory Analysis (Sept. 17, 2003),
a004/a-4.html [].
It might be reason­able to ask whether economically significant rules are less likely to make it out of this process with unresolved ambiguities—the stakes are simply too high, and the scrutiny too probing, to allow agencies to act on even the strongest incentives from Auer.

To address this possibility, in a separate analysis I used a different sam­pling strategy that brought so-called “other significant” rules into the fold. 244 See Daniel E. Walters, Auer’s Incentives 26 (June 28, 2017) (unpublished manu­script) (on file with the Columbia Law Review). This separate analysis was a pilot study for the more extensive analysis in this Article. But the results were similar to the null results for the economi­cally significant rules. 245 Id. at 50. In this additional robustness analysis, I selected three agencies with a high guidance-to-rulemaking ratio 246 Connor Raso’s work on guidance issuance in agencies allowed me to identify this ratio. See Raso, supra note 127, at 813 tbl.3. —HHS, DOL, and EPA—and collected all significant rules and proposed rules issued from 2010 through 2016. I then tested whether these three agencies altered their rule clarity following a series of Supreme Court cases address­ing Auer deference during this period. For each of these cases, there were no statistically significant effects. 247 Walters, supra note 244, at 35. Not only do these findings confirm that economically significant rules are not a grossly distorted subset of rules to look at for the main analysis, but they also reinforce the findings above: There is no evidence that agencies have changed rule-writing practices in response to Auer.

G. Summary of Empirical Findings

This Part has presented an extensive empirical analysis of the self-delega­tion thesis. The empirical analysis of over 1,200 rules and 19 mil­lion words of regulatory text fails to support the claim that agencies respond to Auer by deliberately obfuscating regulations. The effects of Auer on rule vagueness are either absent or, at least in the case of one particular kind of vagueness (laxity), exactly the opposite of what the self-delegation thesis would suggest. Moreover, these findings remain robust even after looking at the data from a number of angles. The analysis started with a basic statistical breakdown of the aggregate patterns, find­ing that the aggregate trend appears to be toward greater rule clarity—a challenge, at the very least, to the notion that agency behavior is becom­ing more unconstrained. More nuanced empirical analyses of whether salient moments in Auer’s development were pivot points, slowing down or accelerating preexisting trends, also showed no evidence of any per­verse effect. Finally, more agency-specific exposure to Auer deference, as measured by win rates and number of cases invoking Auer, did not corre­late with any decrease in clarity down the road. These results, if nothing else, shift the burden in the debate over Auer. The self-delegation theory has played a critical role in elevating the assault on Auer, but there simply is no solid evidence to support it, and there even is some evidence to refute it.

IV. Problematizing the Self-Delegation Thesis

As discussed in Part II, Auer’s critics see agencies as having a self-inter­ested motive to maximize future interpretive discretion through the kind of strategic sandbagging that Auer supposedly makes possible or more attractive. 248 This assumption that agencies will maximize their own self-interest (often against the public interest) animates much economic, or “public choice,” scholarship on the bureaucracy. See Jerry L. Mashaw, Greed, Chaos, and Governance: Using Public Choice to Improve Public Law 131–57 (1997) (“[A]gencies may appear to behave more like inde­pendent entrepreneurs seeking funding from the Congress for projects of their own than like well-instructed agents implementing their principal’s orders.”); André Blais & Stéphane Dion, Are Bureaucrats Budget Maximizers? The Niskanean Model & Its Critics, 22 Polity 655, 656–63 (1990) (providing an overview of the public-choice model and the major criti­cisms that have been raised in response); Spence & Cross, supra note 163, at 99 (sum­marizing early public-choice scholarship that characterizes “agency bureaucrats as shirking, self-interested budget-maximizers who thwart the will of the people and good govern­ment”). Indeed, some argue that this assumption is the sine qua non of public-choice scholarship. See Jeffrey J. Rachlinski & Cynthia R. Farina, Cognitive Psychology and Optimal Government Design, 87 Cornell L. Rev. 549, 563 (2002) (“Common to all analyses labeled ‘public choice’ is the core concept, taken from economic thought, of instrumental ration­ality: The individual will order his behavior so as to maximize the likelihood of achieving his individually defined goals.”). In economic parlance, Auer is said to change the expected utility of writing vague rules by making the payoff from reserv­ing interpretive clarity for subsequent, less formal pronouncements more likely to materialize. The analysis in Part III, however, suggests that, if agen­cies are indeed acting to maximize their self-interest, they are not doing it by creating rules that are more vague, for there is no evidence that Auer has perversely affected agency rule writing.

This seems to present a puzzle: If the incentives are so clear, why do we not see any evidence that agencies follow through on them? In fact, though, the incentives vis-à-vis the timing of interpretive clarity are not clear, which is why the findings in Part III should not really come as any surprise. The assumption that agencies invariably seek to maximize discre­tion, while central to the self-delegation thesis, distorts the task of rulemaking. 249 Cf. George A. Krause, Agency Risk Propensities Involving the Demand for Bureaucratic Discretion, in Politics, Policy, and Organizations: Frontiers in the Scientific Study of Bureaucracy 41, 48, 64 (George A. Krause & Kenneth J. Meier eds., 2003) (arguing that the demand for discretion on the part of agencies is highly situational and that in many instances, particularly those with endemic uncertainty, it would be expected that agencies would reject additional discretion). In this task environment, discretion at some point down the road will not always be perceived by agencies as a good on par with certainty in the present moment. 250 Id. at 48–50. In short, the self-delegation thesis overstates the case by oversimplifying the model parameters 251 The model behind the self-delegation thesis is grounded in classical economic models of rational choice that assume comprehensive rationality and “ends-means reason­ing.” See Bryan D. Jones, Bounded Rationality and Political Science: Lessons from Public Administration and Public Policy, 13 J. Pub. Admin. Res. & Theory 395, 397 (2003) [hereinafter Jones, Bounded Rationality] (describing how advances in behavioral sciences in the 1950s influenced political theorists who pushed back against the notion that organi­zations engage in “fully rational decision making”). Although there is wide variation in the rational-choice tradition across disciplines, scholarship in this vein tends to share common features, including the assumptions that “decision-makers hold stable, ranked and ordered preferences for outcomes” and that they possess the “necessary information to connect choices to outcomes.” Bryan D. Jones et al., Behavioral Rationality and the Policy Processes: Toward a New Model of Organizational Information Processing, in Handbook of Public Policy 49, 51 (B. Guy Peters & Jon Pierre eds., 2006). and ignor­ing the real-world cognitive limits and institutional pressures that affect agencies’ decision calculus and push them into present-mindedness. The real-word complexities and uncertainties to which I refer are myriad, but two are particularly noteworthy.

First, a longstanding research tradition—which the self-delegation the­sis sells short—shows that  agencies, like individuals,  are not likely to make perfectly rational decisions  all  the time. 252 See, e.g., Herbert A. Simon, Administrative Behavior: A Study of Decision-Making Processes in Administrative Organizations 93–97 (1997) (describing the limitations of “objective rationality” for developing models of administrative agency decisionmaking). Much  institutionally grounded  research has shown  that agencies  are often boundedly rational, meaning that  their leaders and  workers have cognitive biases and tendencies that put a thumb on the scale in favor of attention to the more immediate effects of any decision. 253 Empirical scholarship on the bureaucracy—accumulated over the course of decades—demonstrates that administrative agencies are boundedly rational “satisficers.” See, e.g., Jonathan Bendor & Terry M. Moe, An Adaptive Model of Bureaucratic Politics, 79 Am. Pol. Sci. Rev. 755, 756 (1985) (describing a model for analyzing bureaucratic politics, based on computer simulations, that assumes decisionmakers “mov[e] in directions that appear to promise them greater utility,” without assuming that such decisionmakers “optimize” or “carry out complex calculations”); Charles E. Lindblom, The Science of “Muddling Through,” 19 Pub. Admin. Rev. 79, 80–81 (1959) [hereinafter Lindblom, Muddling Through] (arguing that the optimization model of agency behavior “assumes intellectual capacities and sources of information that men simply do not possess” and ignores that the “time and money that can be allocated to a policy problem is limited”); see also James G. March & Herbert A. Simon, Organizations 169–71 (1958) (arguing that organizations develop and use structural mechanisms to adapt to the bound­aries of rationality). Existing research in cognitive psychology  and behavioral economics indicates that agencies are predis­posed  to be present-minded  and to undervalue long-term opportuni­ties. 254 See infra notes 272–274 and accompanying text. Indeed, agency officials  have an aversion to  planning long sequences of  behavior because of  the  uncertainty that deferral creates. 255 See infra notes 275–279 and accompanying text. Yet planning a long sequence of actions is just what the self-delegation thesis asserts that agencies do: They choose to forgo clarity now in order to have greater discretion to be clearer and more specific at some later time. The choices agencies make about the timing of interpretive clarity are likely to be  substantially colored by the cognitive limitations  of agency officials responsible for developing rules. Even if it is assumed that an agency  generally aims to  maximize its  discretion  when  writing rules, 256 Mark Seidenfeld, Cognitive Loafing, Social Conformity, and Judicial Review of Agency Rulemaking, 87 Cornell L. Rev. 486, 488 (2002) (noting the widespread assump­tion that “a rational agency will promote its interests by formulating a rule that comports with the agency’s objectives to the greatest extent possible without going so far as to incite the judicial or political branches to countermand it”). clear-eyed  realism about the  complexity  and uncertainty of this  task makes it unsurprising that Auer has had no  discernible effect on rule writ­ing. 257 Indeed, agencies often bind themselves to rules and norms that seem inexplic­able from a comprehensively rational, self-interested perspective. John Brehm & Scott Gates, Working, Shirking, and Sabotage: Bureaucratic Response to a Democratic Public 3 (1999) (finding that, despite being basically unconstrained by superiors, rank-and-file bureaucrats “for the most part are hard workers, motivated principally by . . . ‘functional’ preferences, the extent to which bureaucrats feel rewarded by performing their job duties well”); see also Elizabeth Magill, Foreword, Agency Self-Regulation, 77 Geo. Wash. L. Rev. 859, 899–900 (2009) (describing the many ways that agencies take steps to constrain their discre­tion, and arguing that these decisions to self-regulate are potentially revealing of what makes agencies “tick”—that is, what motivates agencies). For a wonderful volume with variations on this theme, see generally Administrative Law from the Inside Out: Essays on Themes in the Work of Jerry L. Mashaw (Nicholas R. Parrillo ed., 2017). To be sure, the rational-choice tradition does not act as if uncertainty does not exist. More recent rational-choice scholarship folds uncertainty into an overall expected utility function wherein “individuals form strategic preferences probabilistically” and then “compare the probability that their most preferred outcome will occur against the possibility that their less pre­ferred outcome will occur, and both against the cost of making a decision.” Jones et al., supra note 251, at 51. The difference with bounded rationality is simply one of degree: The theory holds that “[b]ecause of human cognitive architecture, uncertainty is far more funda­mental to choice than expected utility theory admits.” Id. at 55.

Second, Auer’s critics have also overlooked the many other factors that further put a thumb on the scale in favor of front-loading regulatory clarity, reinforcing a tendency that bounded rationality would already lead agencies to have. 258 Judge Jane Stranch of the Sixth Circuit made this exact point in a recent concurrence:
Finally, I am perplexed by the argument that Auer has led agencies to regulate in a way that is broad and vague with, apparently, the goal of creating maximum leeway to define the meaning of a regulation some­where down the road. That claim assumes a world of political continuity and agency longevity that we would be hard pressed to find today. It also ignores multiple incentives and constraints. Consider the internal pres­sures within the agency and throughout the governing executive branch to implement the agency’s program and the external pressures from those regulated and their lobbyists to obtain predictability, both of which encourage clear regulations.
United States v. Havis, No. 17-5772, 2018 WL 5117187, at *7 (6th Cir. Oct. 22, 2018) (Stranch, J., concurring).
For instance, future changes in politics and resources cannot be predicted and create substantial pressure to lock in meaning in clear legislative rules. While there are innumerable uncertain­ties like these, one in particular should have stood out to adminis­trative law theorists. Hard look review under the APA’s arbitrary and capricious standard is widely viewed as having had far-ranging effects on how agencies write rules. 259 See infra section IV.C.2. Since the Supreme Court’s endorse­ment of so-called hard look review in Motor Vehicle Manufacturers Ass’n v. State Farm Mutual Auto Insurance Co., 260 463 U.S. 29, 43 (1983). agencies by almost every account have had strong incentives to build thorough rulemaking rec­ords that vet every serious alternative and to support every choice with evi­dence. 261 See infra section IV.C.2. As a result, agencies also have the incentive to speak with clarity. After all, it is presumably close to impossible to be thorough in justification and explanation yet vague in prescription. In some instances there are even hard look cases that take issue with agencies’ failure to match rule precision with a demonstrably complex problem. 262 See infra section IV.C.3. Hard look’s incen­tives thus stand in considerable tension with many of Auer’s purported incentives, and it should not be surprising that hard look review’s more immediate incentives would tend to win out when agency officials confront the problem of the timing of interpretive clarity.

This Part explains why the failure of any of the perverse effects asserted by Auer’s critics to materialize should have been entirely pre­dictable based on prior empirical institutional scholarship on bureaucratic decisionmaking. Section IV.A reviews research in cognitive psychol­ogy, behavioral law and economics, and public administration, highlight­ing the ways that agency decisionmakers are generally thought to per­ceive intertemporal uncertainties in deciding whether to maximize discre­tion in decisionmaking. Section IV.B brings these insights to bear on the particular problem that Auer presents for agencies (and that prompted such widespread concern about Auer’s allegedly perverse incen­tives): the question of the timing of interpretive clarity. On this ques­tion, the presentist bias of bounded rationality suggests one compel­ling account for the results reported in this Article. Finally, section IV.C provides an additional explanation for why these results should not be at all surprising: The threat of hard look review creates immediate incen­tives for clarity in rule writing that can be expected to overwhelm any of Auer’s purported incentives.

A. The Limits of Comprehensive Rationality in Administrative Agencies

The theory of bounded rationality “emerged as a critique of fully rational decision making” and was motivated by Nobel laureate Herbert Simon’s efforts to ground a theory of choice in “scientific principles of observation and experiment rather than the postulation and deduction characteristic of theoretical economics.” 263 Jones, Bounded Rationality, supra note 251, at 397. Acknowledging the over­whelming complexities that decisionmakers confront, as well as their funda­mentally limited capacity to understand and eliminate uncertain­ties related to these complexities and to make tradeoffs among incom­mensurable goods, 264 Id. at 397–99; see also Jones et al., supra note 251, at 55 (“[B]ounded rationalists argue that people find it difficult to trade off one goal against another when forming pref­erences and making choices.” (citations omitted)). bounded rationality predicts that decisionmakers will often fail to maximize or optimize what might be objectively in their long-term interest. 265 See Herbert A. Simon, Rational Choice and the Structure of the Environment, 63 Psychol. Rev. 129, 136 (1956) (“Since the organism . . . has neither the senses nor the wits to discover an ‘optimal’ path . . . we are concerned only with finding a choice mechanism that will lead it to pursue a ‘satisficing’ path, a path that will permit satisfaction at some specified level of all of its needs.”). Instead of maximizing their utility, decisionmakers “satisfice”—“choos[ing] alternatives that are ‘good enough.’” 266 Jones, Bounded Rationality, supra note 251, at 399. More­over, decisionmakers develop heuristics, or shortcuts, that help them pro­cess the complexities of real-world decisionmaking but also often lead them astray. 267 See generally Daniel Kahneman, Thinking, Fast and Slow (2013) (providing an overarching summary of research on heuristics). For applications of this research tradition to administrative law, see Rachlinski & Farina, supra note 248, at 579 (arguing that the detrimental effect of heuristic decisionmaking on administrative agencies may be muted as compared to the effect on other governmental actors, due to the high level of expertise in administrative bodies); Seidenfeld, supra note 256, at 529–43 (analyzing how the group-based context in which administrative decisionmaking occurs may either attenuate or amplify heuristic biases). We know, for instance, that decisionmakers filter and priori­tize incoming information, selectively activate central and periph­eral processing systems, and turn to previously discovered solutions before initiating a “search” for new solutions. 268 Jones, Bounded Rationality, supra note 251, at 400. If a search is not acti­vated, existing solutions, norms, and routines are usually sufficient, as they offer “simplified models that capture the main features of a prob­lem without capturing all its complexities.” 269 March & Simon, supra note 253, at 169. These kinds of adaptive cognitive infrastructures mean that “objective” information will rarely be sufficient to predict behavior. As political scientist Bryan Jones has explained, “People must adapt not just to the objective circumstances in which they find themselves, but also to their own inner cognitive and emo­tive constitutions.” 270 Bryan D. Jones, Politics and the Architecture of Choice 10 (2001) [hereinafter Jones, Architecture of Choice].

Research in psychology and behavioral economics shows that a key chal­lenge for boundedly rational decisionmakers is intertemporal uncer­tainty. 271 See generally David L. Weimer, Behavioral Economics for Cost-Benefit Analysis: Benefit Validity When Sovereign Consumers Seem to Make Mistakes 106–09 (2017) (providing an overview of psychological research into deviations from the neoclassical model of exponential discounting for intertemporal tradeoffs). Numerous studies have shown that decisionmakers engage in “hyperbolic discounting,” wherein the discount applied to the value of a long-term outcome grows nonlinearly as the time to reward increases. 272 David Laibson, Golden Eggs and Hyperbolic Discounting, 112 Q.J. Econ. 443, 445 (1997) (summarizing research on “[h]yperbolic discount functions” for the value of an outcome which shows that decisionmaking is “characterized by a relatively high discount rate over short horizons and a relatively low discount rate over long horizons”). In other words, decisionmakers do not perceive long-term benefits the same way that they perceive short-term benefits; typically, they overvalue short-term benefits relative to equally attractive long-term payoffs. Were it consistent and predictable, hyperbolic discounting might be easily recon­ciled with comprehensive rationality through a uniform discount rate, but the sheer number of behavioral anomalies that have been empirically documented and the “spectacular variation” in discounting behavior sug­gest a larger failure of economic modeling. 273 See Shane Frederick, George Loewenstein & Ted O’Donoghue, Time Discounting and Time Preference: A Critical Review, 40 J. Econ. Literature 351, 352, 362–65 (2002) (detailing “empirical regularities” that contradict the classical discounted-utility model). Some of the better expla­nations of the pervasive economic anomalies that economists have docu­mented lie in cognition. That is, “temporal distance influences individu­als’ responses to future events by systematically changing the way they construe those events.” 274 Yaacov Trope & Nira Liberman, Temporal Construal, 110 Psychol. Rev. 403, 403–04 (2003) (showing that “individuals form more abstract representations, or high-level construals, of distant-future events than near-future events,” with the practical effect that “the value of outcomes is discounted or diminished as temporal distance from the outcomes increases”).

Organizations—both the institutions as a whole and the individuals who compose them—have well-documented tendencies toward bound­edly rational decisionmaking, and the evidence overall suggests that govern­ment officials deal with the challenge of intertemporal uncer­tainty, in particular, by focusing on short-term goals. For instance, policymak­ers have been shown to rely on a “toolkit of loosely connected heuristics” that collectively result in incremental policy adaptation. 275 Jonathan Bendor, Incrementalism: Dead yet Flourishing, 75 Pub. Admin. Rev. 194, 195 (2015). That is, they learn how to manage uncertainty and goal ambiguity by proceed­ing in small, controllable steps. 276 See id. at 194–95 (explaining the applications and enduring legacy of Professor Charles Lindblom’s theory of disjointed incrementalism); Charles E. Lindblom, Still Muddling, Not Yet Through, 39 Pub. Admin. Rev. 517, 517–18 (1979) (describing the different analytical strategies and preferences associated with “incrementalism”). Drawing on past experience, organizations adopt rules, routines, and processes that help reduce the costs of analysis and further reinforce incremental adjustment. 277 James G. March, A Primer on Decision Making: How Decisions Happen 58–59 (1994) (“Individuals and social systems depend on rules and on the standardization, routinization, and organization of actions that they provide.”); Barbara Levitt & James G. March, Organizational Learning, 14 Ann. Rev. Soc. 319, 323 (1988) (“[L]earning produces increasing returns to experience . . . and leads an organization, industry, or society to persist in using a set of procedures or technologies that may be far from optimal.” (citation omitted)). Bureau­cratic organizations, in particular, put great faith in rules and established practices even as new evidence suggests a need to adapt. 278 James Q. Wilson, Bureaucracy: What Government Agencies Do and Why They Do It 59–65 (1989) (discussing the difficulty of enacting meaningful reform in “established bureaucracies with strong professional traditions”). These characteris­tics suggest a distinct tendency, both at the individual and organi­zational level, for incremental adaptation over “plan[ning] long behavior sequences.” 279 Jones, Architecture of Choice, supra note 270, at 61.

In sum, the theory of bounded rationality suggests that there are lim­its to the capacity of individuals and organizations to specify and pur­sue their goals, especially over the long term, when faced with more immedi­ate consequences of their decisions. 280 March, supra note 277, at 9 (“Although decision makers try to be rational, they are constrained by limited cognitive capabilities and incomplete information, and thus their actions may be less than completely rational in spite of their best intentions and efforts.”). Within this environment of limited capacity and scant resources to adapt, choice is structured by cognitive heuristics, professional norms, and routines, some of which are barely perceptible to the decisionmaker, let alone consciously chosen. 281 Amos Tversky & Daniel Kahneman, Judgment Under Uncertainty: Heuristics and Biases, 185 Science 1124, 1124 (1974) (“[P]eople rely on a limited number of heuristic principles which reduce the complex tasks of assessing probabilities and predicting values to simpler judgmental operations.”); see also Rachlinski & Farina, supra note 248, at 555–56 (describing how “people rely on two primary strategies to make the most of their cogni­tive abilities,” namely relying on “mental shortcuts” and on “organizing principles” that “consist of a scripted set of default information and organizational themes that help people focus on the information most likely to be relevant”). Particularly when the choices involve complex tradeoffs over time, organiza­tions, including administrative agencies, can be expected to engage in incremental decisionmaking that, by definition, sacrifices some long-term benefits in order to secure immediate benefits.

The point of this discussion is not to demonstrate that bounded rational­ity perfectly explains real-world behavior, or even necessarily to dismiss nonbounded rationality at all. 282 Neither is the aim to defend bounded rationality and incrementalism as a pre­scriptive theory. Much of the debate over Lindblom’s The Science of “Muddling Through” concerned whether Lindblom’s theory was defensible as a prescriptive theory, rather than a merely descriptive one. See Lindblom, Muddling Through, supra note 253. The domi­nant prescriptive theory in administrative law has historically been a more rational-comprehensive theory emphasizing cost-benefit analysis and other analytical methods and requirements. But, citing Lindblom, some administrative law scholars are now urging the benefits of an incrementalist approach. See Sidney A. Shapiro & Robert L. Glicksman, Risk Regulation at Risk: Restoring a Pragmatic Approach 22–27 (2003) (arguing that admin­istrative law has shifted toward a paradigm of “comprehensive analytical rationality,” and that this framework should be rejected in favor of a pragmatic attitude characterized by incrementalist experimentation); Wagner et al., Dynamic Rulemaking, supra note 110, at 229–32 (noting that incrementalism “has fallen out of favor as a prescriptive model for policymaking in recent decades,” but that it “may nevertheless be advantageous under conditions of limited knowledge and political conflict”). In some situations, particularly those in which complexity and uncertainty are low, agency officials can surely behave more like the archetypal homo economicus—specifying a goal ex ante, considering large quantities of information and numerous alternatives, and coming to a decision that maximizes goal attainment. The point is simply that the widely acknowledged constraints of bounded rationality cannot be ignored when theorizing about administrative behav­ior. As Professor Mark Seidenfeld has observed, the assumption that agencies are comprehensively rational utility maximizers is “not so much wrong as incomplete”: “[I]ncentives, whether applied to an institu­tion like an agency or to individuals, matter,” but they “are not the only things that matter.” 283 Seidenfeld, supra note 256, at 488. The next section explains why the choice that agen­cies face in deciding whether to provide rule clarity now or later is one that is particularly understandable from the standpoint of bounded rationality.

B. Bounded Rationality’s Impact on the Timing of Interpretive Clarity in Administrative Action

After Congress delegates to agencies the authority to interpret stat­utes and make policy, agencies must decide not only the what of policy, but also the when. Most obviously, they have to decide when to act rather than demur. 284 See, e.g., Cary Coglianese & Daniel E. Walters, Agenda-Setting in the Regulatory State: Theory and Evidence, 68 Admin. L. Rev. 93, 93–95 (2016) (synthesizing “an in-depth set of discussions” with “scholars, practitioners, and government officials to discuss agenda-setting in the regulatory process”); Michael D. Sant’Ambrogio, Agency Delays: How a Principal-Agent Approach Can Inform Judicial and Executive Branch Review of Agency Foot-Dragging, 79 Geo. Wash. L. Rev. 1381, 1388 (2011) (reconceiving agency delay “as a principal-agent problem between congressional enacting coalitions and the agency”); Cass R. Sunstein & Adrian Vermeule, The Law of “Not Now”: When Agencies Defer Decisions, 103 Geo. L.J. 157, 158 (2014) (“Every day of every year, administrative agencies must decide what and whether to decide.”). But the when also comes into play even when agencies decide to act. That is because agencies can act now with the expecta­tion that it will be their final statement, or they can save some interpretive clarity or specificity for later. 285 The Chenery doctrine affords agencies broad authority to choose when to provide clarity by giving agencies the prerogative to make policy via prospective rulemaking or iter­ative adjudication. SEC v. Chenery Corp., 332 U.S. 194, 203 (1947) (holding that “the choice made between proceeding by general rule or by individual, ad hoc litigation is one that lies primarily in the informed discretion of the administrative agency”). The self-delegation thesis can be under­stood simply as a hypothesis that the existence of Auer defer­ence alters the expected utility of aiming for greater clarity in the future rather than in the present.

The framework of bounded rationality can  help us to  understand why, contrary to the expectations of the  self-delegation thesis, agencies might exhibit a distinct bias toward front-loading interpretive clarity, Auer notwithstanding. 286 Sunstein and Vermeule might suggest that the analysis in this section  commits  the sign fallacy. See Sunstein  &  Vermeule, Unbearable Rightness, supra note 16,  at 309. If  the  theoretical account offered here were without the empirical analysis in Part III, this would perhaps be a problem. Part III, though, shows not only that  there is no evidence to sup­port the self-delegation thesis  but also that,  on one measure,  agencies have increased the  clarity of their writing. This is entirely consistent with the account of bounded rationality in Part IV. Choices about  the timing of interpretive  clarity pose an extraordinarily complex set of questions for agencies, and consciously opting to play the long  game introduces yet more complexity  and uncer­tainty. If, as bounded  rationality would suggest,  agencies do not  have the capacity to comprehensively trace out every future contingency involved in deferring some questions and to optimize  a  strategy to maximize their utility,  they are more likely to satisfice by doing the best they can in the present to pursue their goals. 287 Note, as well, that the predictions from a comprehensive-rationality framework are observationally equivalent when uncertainty is high—in both cases the agency will choose to forgo some future discretion. See Krause, supra note 249, at 48 (modeling the “agency preference for more, less, or the same level of discretion in response to a change in policy outcome uncertainty at a given (fixed) level of agency utility”). Bounded rationality helps explain why agencies may overestimate the uncertainty, feeding the impulse to speak clearly in rules.

As agencies navigate the choice between providing more clarity now or later, they also cannot avoid confronting a task environment of uncertainty and complexity. The major uncertainties that exist—goal ambigu­ity and tactical uncertainty—each increase the relevance of more immediate and concrete concerns and deter agencies from planning long sequences of behavior as part of a comprehensive strategy.

1. Goal Ambiguity. — One form of uncertainty is goal or task ambigu­ity. Agency officials may not actually have a clear idea of what it is that they want to accomplish in the long run when they act. 288 See Bendor & Moe, supra note 253, at 771 (offering a dynamic model of bureau­cratic goal adaptation). At a mini­mum, it will probably be much more challenging to have clarity over longer-term goals than shorter-term ones. Goal uncertainty is already pervasive in administrative rulemaking and can seriously affect the rulemak­ing task. 289 See, e.g., Christopher Carrigan, Structured to Fail? Regulatory Performance Under Competing Mandates 52–82 (2017) (discussing goal ambiguity and its relationship with regulatory performance); Jung Wook Lee, Hal G. Rainey & Young Han Chun, Goal Ambiguity, Work Complexity, and Work Routineness in Federal Agencies, 40 Am. Rev. Pub. Admin. 284, 284–86 (2010) (collecting sources documenting the presence and effect of goal ambiguity in federal agency rulemaking). Some of this ambiguity stems from multiple, conflict­ing delegations, 290 See Eric Biber, Too Many Things to Do: How to Deal with the Dysfunctions of Multiple-Goal Agencies, 33 Harv. Envtl. L. Rev. 1, 7 (2009) (discussing the “ubiquity” of multiple-goal agencies—for example, the National Park Service, which is “required both to protect the natural resources of the parks and to develop facilities for visitors”). as well as from jurisdictional overlaps and the constant pull of multiple political principals. 291 See Jason Marisam, Duplicative Delegations, 63 Admin. L. Rev. 181, 184 (2011) (discussing the problem of duplicative delegations and describing the “antiduplication insti­tutions” that have emerged endogenously to help smooth over potential conflicts between agencies); Michael M. Ting, A Theory of Jurisdictional Assignment in Bureaucracies, 46 Am. J. Pol. Sci. 364, 366 (2002) (outlining a positive explanation of the existence of dupli­cative delegations). Much of it, however, is simply inher­ent in the highly complex process of policymaking, in which agency officials are forced to act before they have fully formed and specified objec­tives, 292 See Biber, supra note 290, at 9–13 (deriving insights into agency incentives from economic models of principal–agent relationships, under which agencies are forced to produce results at the principal’s pace). all in the face of significant public pressures to pursue particu­lar goals. 293 See Susan Webb Yackee, Sweet-Talking the Fourth Branch: The Influence of Interest Group Comments on Federal Agency Rulemaking, 16 J. Pub. Admin. Res. & Theory 103, 117–19 (2005) (analyzing the impact of public comments on the content of agency rules).

In some sense, agency officials who are completely uncertain of their goals cannot even begin to think about acting strategically with respect to the timing of interpretive clarity. 294 Justin R. Pidot, Governance and Uncertainty, 37 Cardozo L. Rev. 113, 164 (2015) (“If the goals of governance are up for grabs[,] . . . this renders the project of governance inherently unstable. Where a policy’s meta-goal includes modification of its first order priorities, it risks proceeding without a compass.” (footnote omitted)). Less radical uncertainty about overarch­ing goals might clear the way for more strategic consideration of the timing question, but as it is, we would expect boundedly rational agen­cies—torn between these sometimes confusing pushes and pulls—to err on the side of what they can control and measure with greater cer­tainty. 295 See Biber, supra note 290, at 12 (“Where an agency is faced with multiple goals, it will tend to overproduce on the goals that are complements and the goals that are easily measured, and it will tend to underproduce on the goals that are substitutes and the goals that are hard to measure.”); Avinash Dixit, Incentives and Organizations in the Public Sector: An Interpretive Review, 37 J. Hum. Resources 696, 715–16 (2002) (noting that “incentives will be generally weaker” in situations of multiple goals and multiple princi­pals, and that “in their day-to-day operations, agencies will think not in terms of the multi­ple and vague ultimate goals, but in terms of a smaller number of immediate and mea­surable tasks”). What to do in the short term will almost inevitably be more cer­tain, particularly in the constantly changing world of executive-branch politics. With new administrations or changes in the balance of political power in Washington come new goals that agencies can only faintly imag­ine, let alone act on with any precision.

2. Tactical Uncertainty. — Assuming that goal ambiguity is overcome and an agency has determined that more discretion down the road (and less clarity now) will better advance its goals, an agency still faces signifi­cant tactical complexities in executing a self-delegation strategy that will unfold over time. Many of these complexities are also captured by mod­els of how government decisionmakers deal with uncertainty by choosing whether to build “dynamic” elements into their rules. Facing a choice between “static law,” which “intend[s] for the intervention to remain fixed, making no special allowances that could facilitate a future modifica­tion,” 296 Pidot, supra note 294, at 131. and “dynamic law,” which involves deliberately setting the stage for future revision, 297 See id. at 140–41. agencies are forced to cope with the often-excessive cost of planning for, and monitoring over, the longer term.

The conscious attempt to save some opportunity for greater clarity or specification at some later time is analogous to a choice about how many contingencies to account for in a rule, and it is therefore subject to many of the same costs and risks. 298 See Diver, supra note 128, at 67–79 (arguing that the “optimal” precision of rules is determined by tradeoffs between three incommensurable goals). For instance, agencies that engage in strategic deferral or self-delegation would have to commit, ex ante, to a number of subsequent actions. But there is no guarantee that there will be resources to carry out these subsequent actions 299 David L. Markell & Robert L. Glicksman, A Holistic Look at Agency Enforcement, 93 N.C. L. Rev. 1, 14 n.33 (2014) (noting that “lack of clarity leads to delays and increases in transaction costs” when it comes to enforcement). or that responsibil­ity for those subsequent actions will not be given to other actors who do not share the same long-term vision. This is a particularly salient consideration for agency leaders, as the average tenure of political appoin­tees is around two years. 300 Raso, supra note 127, at 803; see also Anne Joseph O’Connell, Vacant Offices: Delays in Staffing Top Agency Positions, 82 S. Cal. L. Rev. 913, 919 n.23 (2009) (collecting sources measuring the average tenure of political appointees). Similarly, even if the tenure of an agency leader lasts longer than that, external political circumstances may change, making it more difficult to do in the future precisely what that leader might have intended in the present. 301 Elections can fundamentally change the enforcement priorities, leaving an agency engaged in a self-delegation strategy high and dry. See Daniel T. Deacon, Note, Deregulation Through Nonenforcement, 85 N.Y.U. L. Rev. 795, 812 (2010) (describing the ways that an incoming administration can, with relative ease, reverse the previous administration’s enforcement policies).

Finally, there are back-end risks related to potential vetoes of subse­quent enforcement actions. For instance, in certain cases courts have blocked subsequent enforcement when an agency’s lack of clarity failed to provide fair notice from the rule itself that conduct was prohibited. 302 See, e.g., Gen. Elec. Co. v. EPA, 53 F.3d 1324, 1333–34 (D.C. Cir. 1995) (holding that—given the “lack of clarity in the regulations” promulgated by EPA, combined with a subsequent agency policy statement that did not adequately address the contested issue—General Electric was not fairly put on notice that its chemical distillation process violated EPA regulations). Interest group mobilization could also block implementation and updat­ing. 303 See Frank R. Baumgartner et al., Lobbying and Policy Change: Who Wins, Who Loses, and Why 40 (2009) (describing the disproportionate ability of entrenched interests to mobilize to protect the regulatory status quo); Pidot, supra note 294, at 118–19 (arguing that “political and legal forces invested in the status quo” often obstruct progress in agency policymaking); Susan Webb Yackee, The Politics of Ex Parte Lobbying: Pre-Proposal Agenda Building and Blocking During Agency Rulemaking, 22 J. Pub. Admin Res. & Theory 373, 374 (2012) (theorizing that interest group lobbying “play[s] a key role during the pre-proposal stage” of agency rulemaking and that these groups’ influence “manifests itself through both agenda building and agenda blocking”). In sum, as Professor Justin Pidot explains, “Creating dynamic regula­tion is an inherently complex task that requires lawmakers to con­sider more than the immediate government action at hand.” 304 Pidot, supra note 294, at 175. Front-end clarity, like a decision to write “static law,” is in fact best understood as a default strategy for coping with these costs and uncertainties.

Front-end clarity has its own costs, to be sure, and deferral of such clar­ity might yield certain benefits, even if they are somewhat hard to pre­dict. 305 See, e.g., Joel A. Mintz, “Running on Fumes”: The Development of New EPA Regulations in an Era of Scarcity, 46 Envtl. L. Rep. 10,510, 10,516 (2016) (discussing how program offices at EPA display a “preference for satisfying stakeholders,” and that, as a result, they “sometimes display a ‘passion for ambiguity’” as a means to “serve a program office’s goal of playing conflicting interests against one another and avoiding outside pressures and subsequent legal challenges” (quoting an anonymous former EPA official)). Bounded rationality does not deny that the costs and benefits of agency decisions are, in principle, measurable. The key takeaway, though, is that the simple act of thinking through these costs and benefits is taxing, and agency officials likely do not have the resources or the time to thoroughly study each factor that might bear on the decision. If agencies navigate the choice of the timing of interpretive clarity in a boundedly rational manner, it becomes understandable that, on balance, “ambiguities are a threat at least as much as they are an opportunity” for agencies. 306 Sunstein & Vermeule, Unbearable Rightness, supra note 16, at 309.

3. Professional Norms and Mission Focus. — Agency staff may also be disin­clined to take on the unnecessary risk of failure that accompanies a self-delegation strategy. This disinclination follows partly from the find­ings on risk-averse behavior in prospect theory: “[I]n the domain of gains people value certain gains over possible gains.” 307 Jones et al., supra note 251, at 57. That is, the possi­ble payoff of the self-delegation strategy is hard for agencies to value properly because it is uncertain to materialize. 308 See Daniel Kahneman & Amos Tversky, Prospect Theory: An Analysis of Decision Under Risk, 47 Econometrica 263, 265–69 (1979) (developing a theory of behavior under which people systematically overweight outcomes that will certainly occur over outcomes that are uncertain to occur); see also supra notes 272–274 and accompanying text (review­ing the literature on hyperbolic discounting and intertemporal uncertainty). Beyond this general fea­ture of individual and organizational decisionmaking under uncertainty, a great deal of research on the bureaucracy shows that career-level bureau­crats are motivated by professional norms and the agency’s mis­sion, which likely leads bureaucrats to overvalue the certain win gained by having promulgated a comprehensive rule. 309 For instance, Herbert Kaufman’s classic study of the Forest Service documents one particularly strong instance of a mission orientation in a federal agency, prompted in part by Forest Service Chief Gifford Pinchot’s brilliant decision to institute an agency-wide education program to inculcate shared values. See Herbert Kaufman, The Forest Ranger: A Study in Administrative Behavior 85–86 (1960). Political scientists John Brehm and Scott Gates show that “organizational culture” develops, in essence, from the ground up, as “uncertain individuals look[ ] to fellow subordinates for appropriate responses.” 310 Brehm & Gates, supra note 257, at 194. This organic development of organizational culture is generalizable as well: In an extensive survey of bureaucratic preferences, Brehm and Gates find that bureaucrats “prefer work and serving the public” over sabotage, shirking, and self-interested utility maximization. 311 Id. at 196. For some, these features of the career civil ser­vice—the “neutral expertise” promoted by professional commitments and the “other-regarding” motivations of bureaucrats—are part of what generates administrative legitimacy “from the inside-out.” 312 See, e.g., Sidney Shapiro, Elizabeth Fisher & Wendy Wagner, The Enlightenment of Administrative Law: Looking Inside the Agency for Legitimacy, 47 Wake Forest L. Rev. 463, 465, 470–71 (2012) (arguing for a move to a “deliberative-constitutive paradigm” that “embraces, rather than rejects, the professionalism of agency staff,” viewing it as a “positive attribute that helps ensure the integration of technical expertise in rulemaking and serves as a buffer against undue influence by highly interested stakeholders”); see also Emily Hammond & David L. Markell, Administrative Proxies for Judicial Review: Building Legitimacy from the Inside-Out, 37 Harv. Envtl. L. Rev. 313, 354 (2013) (noting that “[a]gency culture and professionalism can serve as internal legitimizers for agency behavior” due in part to their promotion of “neutral expertise”); Sidney A. Shapiro & Ronald F. Wright, The Future of the Administrative Presidency: Turning Administrative Law Inside-Out, 65 U. Miami L. Rev. 577, 578 (2011) (arguing that administrative law scholarship often ignores research in public administration that displays the “other-regarding motives” of career-level bureaucrats). On the other hand, as others have noted, one result of this institutionalization of professional norms in agency policymaking is that agencies “at times fix­ate on particular missions, even when the principal has expanded the number of goals the agency is supposed to take into account.” 313 Biber, supra note 290, at 17.

The professional commitments and investment in the mission of the agency are widely thought to interact with an uncertain task environment to produce fundamentally conservative, risk-averse behavior on the part of bureaucrats. Rule writers that are invested in the purposes of rulemak­ing are more likely to focus on the relatively certain gains for their pro­gram that can come from rule clarity today than on the nebulous pros­pect of enhancing their discretion down the road.

C. Satisficing Rule Writers and the Shadow of State Farm

Boundedly rational actors’ aversion to long-term strategy offers a ready explanation for why Auer would never be expected to have discerni­ble behavioral consequences for rule writing. But these tenden­cies are only reinforced by the weight of other core features of administra­tive law. From the standpoint of the typical government lawyer, an agency’s primary short-term concern is crafting a legally defensible rule. This means crafting a rule that can survive hard look review in fed­eral court, where the court will review the rule and its underlying record to ensure that the decision is thoroughly analyzed and responsive to a wide range of perspectives. 314 The literature is replete with treatments of hard look review, with most focusing on whether hard look review serves to improve the regulatory process by requiring agencies to furnish reasoned analysis to support decisions. For a comprehensive overview of both the doctrine and the debates about it, see generally Patricia M. Wald, Regulation at Risk: Are Courts Part of the Solution or Most of the Problem?, 67 S. Cal. L. Rev. 621 (1994). These requirements make it necessary to seek to provide clarity in regulatory language.

1. Hard Look Review Basics. — Hard look review developed in the 1970s out of the Supreme Court’s interpretation of section 706(2)(A) of the APA in Citizens to Preserve Overton Park, Inc. v. Volpe. 315 401 U.S. 402, 416 (1971) (describing an agency’s duty under the arbitrary and capricious standard to conduct a “searching and careful” inquiry). Elaborating on the Court’s reading of the arbitrary and capricious standard in Overton Park, 316 Id. D.C. Circuit Judge Harold Leventhal wrote that a court should exercise its role with “particular vigilance if it ‘becomes aware, especially from a combination of danger signals, that the agency has not really taken a “hard look” at the salient problems, and has not genuinely engaged in reasoned decisionmaking.’” 317 Harold Leventhal, Environmental Decisionmaking and the Role of the Courts, 122 U. Pa. L. Rev. 509, 511 (1974) (emphasis added by Leventhal) (quoting Greater Bos. Television Corp. v. Fed Commc’ns Comm’n, 444 F.2d 841, 851 (D.C. Cir. 1970)).

The precise contours of hard look review are difficult to pin down, 318 Richard J. Pierce, Jr., The Special Contributions of the D.C. Circuit to Administrative Law, 90 Geo. L.J. 779, 786 (2002) (describing hard look review as “notoriously malleable”). in part because it developed iteratively in the D.C. Circuit over many years, and also because it is composed of a “concert” of goals, including promoting detailed explanations from agencies, encouraging agencies to respond to salient comments and perspectives, requiring consistency over time, and requiring adequate reasons to justify an agency deci­sion. 319 See Cass R. Sunstein, In Defense of the Hard Look: Judicial Activism and Administrative Law, 7 Harv. J.L. & Pub. Pol’y 51, 52 (1984) (outlining courts’ “efforts to police and discipline the administrative process”). The most definitive statement came from the Supreme Court’s ruling in State Farm, in which the Court vacated the decision of the National Highway Traffic Safety Administration (NHTSA) to rescind a previously promulgated seat-belt standard applicable to new motor vehi­cles. 320 Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 56–57 (1983). The standard had required the installation of either automatic seat belts or airbags in all new vehicles. 321 Id. at 37. In rescinding the standard, NHTSA argued that it could not “reliably predict[ ] that the Standard would lead to any significant increased usage of restraints,” and in light of the costs of compliance, decided to change course. 322 Id. at 39. The Court found NHTSA’s evidence wanting. NHTSA had not explained the basis of its belief that passengers would disengage automatic seat belts, 323 Id. at 51–57. and it did not provide any analysis of an obvious alternative to requiring auto­matic seat belts: simply mandating the installation of airbags. 324 Id. at 46–51. In reject­ing NHTSA’s reasoning, the Court articulated particularly broad con­tours of arbitrariness review. Agency action is invalid

if the agency has relied on factors which Congress has not intended it to consider, entirely failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implau­sible that it could not be ascribed to a difference in view or the product of agency expertise. 325 Id. at 43.

The State Farm decision—both in terms of the standard enunciated and the “rather strict judicial scrutiny” 326 Stephen Breyer, Judicial Review of Questions of Law and Policy, 38 Admin. L. Rev. 363, 387 (1986). the Court actually exercised in probing the agency’s justifications—indicated the Court’s strong approval of much of the D.C. Circuit’s substantive hard look doctrine. Not surprisingly, State Farm was then followed by a number of high-profile court decisions vacating agency rules under the hard look doctrine. 327 See William S. Jordan, III, Ossification Revisited: Does Arbitrary and Capricious Review Significantly Interfere with Agency Ability to Achieve Regulatory Goals Through Informal Rulemaking?, 94 Nw. U. L. Rev. 393, 412 (2000) (discussing the impact of hard look review on the outcomes of challenges to agency regulations and noting that the D.C. Circuit vacated a majority of challenged rules after State Farm). Today, courts subject agency decisions to extremely stringent analysis, often almost stochastically so, 328 See, e.g., Jody Freeman & Adrian Vermeule, Massachusetts v EPA: From Politics to Expertise, 2007 Sup. Ct. Rev. 51, 53–54 (2008) (arguing that Massachusetts v. EPA, 549 U.S. 497 (2007), which expanded the scope of hard look review to agencies’ “discretionary decisions not to promulgate regulations,” could be “State Farm for a new generation”). At least one study has shown that arbitrariness review is often driven more by the political preferences of the reviewing panel of judges than by anything substantive. See, e.g., Thomas J. Miles & Cass R. Sunstein, The Real World of Arbitrariness Review, 75 U. Chi. L. Rev. 761, 767 (2008). meaning that agencies have reason to make their rules hard look–proof even though most will not be thor­oughly audited.

2. Hard Look’s Effect on Agency Behavior. — By most accounts, the threat of hard look review has had a rippling effect on every aspect of the rule-formation process. 329 See Thomas O. McGarity, Some Thoughts on “Deossifying” the Rulemaking Process, 41 Duke L.J. 1385, 1419–20 (1992) [hereinafter McGarity, Deossifying] (noting the “resource-intensive” and “time-consuming” processes agencies must engage in under the threat of hard look review). Agencies cannot be sure of the “precise scope or intensity of that review process,” but they can loosely predict that poten­tial challengers, in order to maximize the chances of successful chal­lenge, will search for “issues of such importance that the agency argua­bly should have discussed them more thoroughly or in greater detail.” 330 Richard J. Pierce, Jr., Seven Ways to Deossify Agency Rulemaking, 47 Admin. L. Rev. 59, 69 (1995). Agencies acting in the shadow of State Farm are said to “make every effort to ensure a thorough record that can withstand review the first time around” despite this substantial uncertainty. 331 Emily Hammond Meazell, Super Deference, the Science Obsession, and Judicial Review as Translation of Agency Science, 109 Mich. L. Rev. 733, 750–51 (2011) (citing McGarity, Deossifying, supra note 329, at 1400–01). Thus, the incen­tives widely accepted as having been created by hard look review should be clear to agencies: Leave no rock unturned in the administrative rec­ord; address every argument and counterargument, however small; hew closely to statutory criteria; and, perhaps most importantly, leverage agency expertise by populating the record with substantial scientific and technical evidence.

There is no shortage of evidence that agencies have taken such steps in response to the threat of hard look review. First, as a burgeoning litera­ture on rulemaking “ossification” has posited, the stringency of review may prompt agencies to issue fewer rules than they otherwise might. 332 Jerry L. Mashaw & David L. Harfst, Regulation and Legal Culture: The Case of Motor Vehicle Safety, 4 Yale J. on Reg. 257, 262–63 (1987) (arguing that judicial review of motor vehicle safety standards prompted NHTSA to abandon rulemaking in favor of a recall-based strategy of regulation). Note also that ossification is really a two-fold concept, and some scholarship views ossification as primarily slowing the promulgation of rules rather than stopping it entirely. See, e.g., Richard J. Pierce, Jr., Response, Rulemaking Ossification Is Real: A Response to Testing the Ossification Thesis, 80 Geo. Wash. L. Rev. 1493, 1493 (2012) [hereinafter Pierce, Jr., Ossification Is Real] (defining ossification to mean “that it takes a long time and an extensive commitment of agency resources to use the notice and comment process to issue a rule”). Instead of employing rulemaking as a first-best option, agencies are thought to have shifted their energies to nonrulemaking channels such as adjudications and nonlegislative rules to avoid costly vacaturs. 333 See, e.g., Mashaw & Harfst, supra note 332, at 263 (“Since the mid-1970’s, NHTSA has instead concentrated on its alternative statutory power to force the recall of motor vehicles that contain ‘defects’ related to safety performance. It has thus retreated to a traditional form of case-by-case adjudication . . . .”); McGarity, Deossifying, supra note 329, at 1385–96 (noting that “[f]rustrated agencies” have begun to explore methods of “avoiding notice-and-comment rulemaking altogether,” opting instead for adjudications and “nonrule rulemaking”); Pierce, Jr., Ossification Is Real, supra note 332, at 1502 (explaining how, during the 2008 financial crisis, the Treasury Department and Federal Reserve Board identified “old statutory provisions” that would allow them to take “expe­ditious actions” without the burden of “comply[ing] with mandatory decisionmaking procedures”). The evidence of such an effect is concededly somewhat equivocal, as a num­ber of recent empirical studies of rulemaking activities suggest that hard look review is not responsible for any change in the volume or pace of rulemaking. 334 See Cary Coglianese, Empirical Analysis and Administrative Law, 2002 U. Ill. L. Rev. 1111, 1127 (“The empirical evidence for a retreat from rulemaking in the face of stringent judicial review is not nearly as clear as has been generally supposed.”); O’Connell, Political Cycles of Rulemaking, supra note 131, at 963–64 (arguing that the relatively high number of notices of proposed rulemakings from 1983 to 2002 suggests that the costs to rulemaking are not as high as feared); Jason Webb Yackee & Susan Webb Yackee, Testing the Ossification Thesis: An Empirical Examination of Federal Regulatory Volume and Speed, 1950–1990, 80 Geo. Wash. L. Rev. 1414, 1467 (2012) (“[A]gencies—and in particular, DOI agencies—do not appear to have abandoned notice and comment wholesale, either by failing to regulate entirely, or by embracing surreptitious forms of regulation.”). Second, while there is, to date, no systematic empirical analysis of agen­cies’ record-building behavior after State Farm, “it seems a matter of com­mon sense that agencies are mindful of the possibility of judicial review for major rulemakings and would therefore approach rulemaking more deliberately.” 335 Meazell, supra note 331, at 751. As Professor Wendy Wagner has argued, the hard look doctrine would seem to encourage agencies to “load their rule and rec­ord with details and defensive statements” to the point of “defensive over­kill.” 336 Wendy E. Wagner, Administrative Law, Filter Failure, and Information Capture, 59 Duke L.J. 1321, 1357–59 (2010). Of course, in principle, the information-forcing role of hard look review is often cited as a positive one, notwithstanding Wagner’s cogent argu­ment that the effects are often overkill. See Catherine M. Sharkey, State Farm “with Teeth”: Heightened Judicial Review in the Absence of Executive Oversight, 89 N.Y.U. L. Rev. 1589, 1605 (2014) (concluding that “[t]here is little doubt that judicial review plays a significant information-forcing role”); see also Matthew C. Stephenson, A Costly Signaling Theory of “Hard Look” Judicial Review, 58 Admin. L. Rev. 753, 755–56 (2006) (“[J]udicially-imposed explanation requirements can help reviewing courts overcome their comparative informa­tional disadvantage for reasons that are independent of the (in)ability of courts to under­stand or verify the substantive content of the justifications advanced by government decisionmakers.”).

Agencies that write a lot of rules have generally invested in an institu­tional infrastructure that helps them facilitate a response to the prospect of judicial review. Professor Thomas McGarity, more than any­body else, has given significant attention to the way rulemaking processes are structured, and what he describes as the “team model” that agencies usually employ seems tailor-made to respond to the threat of hard look review. 337 See generally Thomas O. McGarity, The Internal Structure of EPA Rulemaking, Law & Contemp. Probs., Autumn 1991, at 57, 90–94 [hereinafter McGarity, Internal Structure] (discussing the advantages and disadvantages of a model of agency policymaking in which the basic “decisionmaking unit” is composed of “representatives from all of the institutional subunits within the agency”); Thomas O. McGarity, The Role of Government Attorneys in Regulatory Agency Rulemaking, Law & Contemp. Probs., Winter & Spring 1998, at 19, 20–21 [hereinafter McGarity, Role of Government Attorneys] (describing that, in the team model, all tasks associated with rule drafting, promulgation, and legal defense “are delegated to a ‘team’ or ‘work group’ composed of representatives from all of the agency subunits that have an interest in the outcome of the rulemaking process”). Rather than allowing one insulated program office to draft a rule in its entirety and elevate it for final approval (what McGarity calls the “assembly line” model), agencies employing the team model seek to draw out a “bureaucratic pluralism” that “transcends the knowledge and experience of any individual person or office within the agency.” 338 McGarity, Internal Structure, supra note 337, at 61. This process aims to ensure, in accordance with hard look review’s manifest goals, that every possible angle receives some attention in the process. 339 See id. at 90–91 (noting that an advantage of the team model is that it encourages “innovative, cross-disciplinary thinking” from people with “divergent professional perspectives”). These processes are a prime example of the kind of rules, norms, and processes that boundedly rational agencies employ to cope with uncertainty. See supra section IV.B.2.

According to McGarity, lawyers have come to play an especially impor­tant role in this process, “advising the relevant agency decision­mak­ers on the many aspects of the rule that might be challenged in court.” 340 McGarity, Role of Government Attorneys, supra note 337, at 22; see also Mintz, supra note 305, at 10,517 (noting that Office of General Counsel attorneys at EPA “some­times attempt to eliminate unenforceable language in proposed regulations,” but con­cluding that their concern with enforceability is not sufficient to eliminate all ambigu­ities); Rosemary O’Leary, The Impact of Federal Court Decisions on the Policies and Administration of the U.S. Environmental Protection Agency, 41 Admin. L. Rev. 549, 566 (1989) (noting that lawyers often “have the last word” in rulemaking). Enforcement staff, for their part, have perhaps the strongest interest in promoting clarity in the regulatory text, as that can greatly improve the enforceability of the rule down the road (with or without the benefit of Auer deference). 341 McGarity, Internal Structure, supra note 337, at 62 (noting that enforcement professionals “are primarily concerned with the degree to which . . . regulated [entities] adhere to agency commands,” and therefore push the agency to “articulate its rules in unambiguous ways” that put regulated parties on notice and minimize their ability to “avoid compliance through interpretational loopholes”). In their role as “scrivener,” agency lawyers are said to be pivotally involved in the process of “achiev[ing] clarity in the wording of the rule,” in “provid[ing] adequate references to the rec­ord in support of the agency’s resolution of major issues,” and in “main­tain[ing] a consistent line of reasoning throughout.” 342 McGarity, Role of Government Attorneys, supra note 337, at 26. The stochasticity of hard look review has apparently increased this role for lawyers in the agency at the expense of program and policy offices. 343 Id. at 26–27. Indeed, Professors Emily Hammond and David Markell recently documented law­yerly meticulousness in a context practically devoid of judicial review—EPA’s processing of petitions to withdraw states’ authority to administer environmental programs that had been delegated as part of cooperative federalism—prompting them to query “whether EPA has so internalized the expectation of judicial review that it treats even informal matters accord­ing to the norms resulting from hard-look review.” 344 Hammond & Markell, supra note 312, at 355.

If agencies “are constantly ‘looking over their shoulders’ at the review­ing courts,” 345 McGarity, Deossifying, supra note 329, at 1412. it stands to reason that these same agencies do not have their eyes on a very different ball: that is, the long-term possibility of gaining strategic advantage by self-delegating through vague rules. The costs of vacatur or remand of a rule are tangible; the benefits of self-delega­tion are speculative. If agencies really are satisficers, and if they are listening to the voices of their legal advisors who tell them that judicial review is likely, they will focus most of their attention on addressing that risk. In addressing the risk of hard look review, agencies will of necessity seek to reduce vagueness. In short, hard look review “requires federal agencies to fully explain their decisions at the outset, favoring a front-loaded decision process that culminates in a single record of decision that allows for judicial review.” 346 Pidot, supra note 294, at 170 (footnote omitted).

3. Hard Look’s Effects on Regulatory Precision. — Apart from inducing a generalized institutional response, the threat of hard look review can be expected to have systemic effects on rule-writing style, presumably favoring clarity over vagueness. 347 See Blake Emerson & Cheryl Blake, Admin. Conference of the U.S., Plain Language in Regulatory Drafting 14–19 (2017), [] (reporting interview-based evidence that agencies aim for clarity in part due to the threat of judicial review). Hard look review’s pres­sure on agencies to seek regulatory clarity would seem to stem from at least three possible factors.

First, hard look challenges sometimes come packaged as claims that rule text is not clear enough. 348 See, e.g., Copar Pumice Co. v. Tidwell, 603 F.3d 780, 800 (10th Cir. 2010) (“We also reject Copar’s argument that by including the term ‘verifiable proof’ in the Notice of Noncompliance, the [Forest Service] imposed a vague standard.”); Ariz. Cattle Growers’ Ass’n v. U.S. Fish & Wildlife, 273 F.3d 1229, 1251 (9th Cir. 2001) (“Based upon the lack of an articulated, rational connection between Condition 1 and the taking of species, as well as the vagueness of the condition itself, we hold that its implementation was arbitrary and capricious.”). In an early hard look case in 1971, Environmental Defense Fund, Inc. v. Ruckelshaus, the D.C. Circuit held that “[c]ourts should require administrative officers to articulate the stan­dards and principles that govern their discretionary decisions in as much detail as possible.” 349 439 F.2d 584, 598 (D.C. Cir. 1971). But see PDK Labs. Inc. v. DEA, 438 F.3d 1184, 1194 (D.C. Cir. 2006) (holding that an agency “proceeding on a case-by-case basis” need only “pour ‘some definitional content’ into a vague statutory term” to survive an arbi­trariness challenge (quoting Pearson v. Shalala, 164 F.3d 650, 660 (D.C. Cir. 2003))). While this emphasis on clarity may not appear as frequently in more recent cases, the principle in Ruckelshaus has never been rejected. 350 See Lightfoot v. District of Columbia, 448 F.3d 392, 400 (D.C. Cir. 2006) (stating in dicta that Ruckelshaus’s statement about the vagueness of rules “may well be an overly broad statement of judicial review, even under the APA, and inconsistent with our more modest jurisprudence in subsequent decades,” but stopping short of rejecting the principle outright). And it does still arise. More recently, the Ninth Circuit, in vacating a decision to delist the Yellowstone grizzly bear population as a threatened species under the Endangered Species Act, rejected “out of hand” the suggestion that a vague promise to relist the species if circum­stances change could “operate as a reasonable justification for delisting” without the benefit of any specific “management responses” and “spe­cific triggering criteria.” 351 Greater Yellowstone Coal., Inc. v. Servheen, 665 F.3d 1015, 1029 (9th Cir. 2011).

Litigants often pair the claim that a rule violates the arbitrary and capri­cious standard with a claim that the rule violates due process by being “too vague to warn the industry of the scope of prohibited con­duct.” 352 E.g., CPC Int’l Inc. v. Train, 515 F.2d 1032, 1052 (8th Cir. 1975). Typically, these challenges are brought after an agency attempts to enforce a vague rule, and in these cases, the courts occasionally vacate agency action based either on Fifth Amendment due process grounds or on arbitrary and capricious grounds. For instance, in Pearson v. Shalala, the D.C. Circuit agreed with the challengers that “the APA requires the agency to explain why it rejects their proposed health claims,” and “do[ing] so adequately necessarily implies giving some definitional con­tent to the phrase ‘significant scientific agreement.’” 353 164 F.3d at 660 (quoting 21 C.F.R. § 101.14(c) (1998)). For the court, “this proposition is squarely rooted in the prohibition under the APA that an agency not engage in arbitrary and capricious action.” 354 Id. (citing 5 U.S.C. § 706(2)(A) (1994)). How­ever, with striking frequency, these kinds of claims are also brought in preenforcement review of rules. 355 A back-of-the-envelope calculation, achieved by searching Westlaw’s database of appellate briefs from the D.C. Circuit using the search terms “rule /p vague & ‘arbitrary and capricious,’” yielded 350 briefs (as of October 2018) mentioning vagueness in close proximity to a challenge to a rule under the arbitrary and capricious standard. These kinds of claims are surely far more numerous than they are successful, 356 Preenforcement challenges, notwithstanding the principle articulated in Ruckelshaus, are often stopped, as a practical matter, by the fact that an agency is usually “entitled to proceed case by case or, more accurately, sub-regulation by subregulation.” Pearson, 164 F.3d at 661. Thus, an agency is not “necessarily required to define” specific terms in its “initial general regulation.” Id. but the claims are not frivo­lous enough that they do not warrant a response, which means agencies attuned to the possibility of undergoing probing review by courts will pre­sumably take few chances and will have short-term reasons to try to make their rules clear. After all, with the availability of preenforcement review of agency rules, the shadow of review need not be very long and can be formative for agency behavior.

Second, the very same record-building strategies that agencies use to defend themselves against the threat of hard look review make it very difficult to leave major ambiguities in the rule text itself. Even though hard look review often focuses on the evidentiary record and the agency’s reasoning, 357 See Sidney A. Shapiro & Richard W. Murphy, Arbitrariness Review Made Reasonable: Structural and Conceptual Reform of the “Hard Look,” 92 Notre Dame L. Rev. 331, 347 (2016) (noting that “as the judicial transformation of informal rulemaking took hold, agencies had to offer far more detailed notices and explanations for their rules, which meant that courts had a much richer set of ‘record’ materials to review”). For a thorough overview of the dynamics (and pathologies) of record building in notice-and-comment rulemaking, see generally Wagner, supra note 336. a certain level of rule clarity in the rule text is neces­sary just to support the kind of thorough consideration of analysis that courts expect. There would be a jarring disconnect between a rule with a lengthy preamble, filled to the brim with studies, estimates, and counterarguments, and a rule text that says not much more than that a regulated entity should take “reasonable” steps to ensure the safety of its plant. 358 Cf. Pearson, 164 F.3d at 660 (“To refuse to define the criteria it is applying is equivalent to simply saying no without explanation.”).

An example can make the point clearer. In Greater Yellowstone Coalition, the agency’s formal plan for delisting the Yellowstone grizzly bear population as a threatened species avoided specifying particular risks that might justify relisting, instead offering a vague promise to care­fully monitor the grizzly population. 359 Greater Yellowstone Coal., Inc. v. Servheen, 665 F.3d 1015, 1029 (9th Cir. 2011). The court demanded greater explana­tion of why the decline in whitebark pine, which had been linked with grizzly mortality, would not threaten the species. 360 Id. Understandably, the agency had not built a record about an issue that its open-ended “man­agement and monitoring framework” did not “even specifically dis­cuss.” 361 Id. If the agency’s position is that the evidence does not support a more specific standard, it will be expected to furnish that evidence, not simply invoke uncertainty as a reason for a vague standard. 362 See Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 52 (1983) (“Recognizing that policymaking in a complex society must account for uncer­tainty, however, does not imply that it is sufficient for an agency to merely recite the terms ‘substantial uncertainty’ as a justification for its actions.”); see also All. for Nat. Health U.S. v. Sebelius, 775 F. Supp. 2d 114, 133–34 (D.D.C. 2011) (rejecting the kind of arbitrariness-as-vagueness challenge described above in part because the agency had offered evidence that it “could not ‘predict with mathematical precision’” what would be the appropriate standard (quoting 72 Fed. Reg. 34,752, 34,787–88 (June 25, 2007))). In effect, if an agency were to elide clarity and specificity, it would leave itself in a bind: Its rulemaking record would likely be spotty, superficial, or both, leaving the rule vulnerable to hard look vacatur.

Finally, agencies actually perceive clarity as a way to reduce the risk of hard look vacatur. As part of its project on plain language in regula­tory drafting, the Administrative Conference of the United States (ACUS) interviewed officials at seven agencies and specifically asked them what impact judicial review has on their incentives to use plain lan­guage in rule writing. 363 See Emerson & Blake, supra note 347, at 2. The study reports that agency officials favored clear drafting as a means of reducing the risk of judicial reversal. 364 See id. at 15. As one agency official noted, the agency could “defend regulations better when we’ve developed the record and made the regulation clear and under­standable to the public.” 365 Id. Another agency official stated, “[I]f regula­tions just aren’t understandable, or they can be misconstrued, you are a lot more vulnerable legally.” 366 Id. (internal quotation marks omitted). As the authors of the ACUS report note, agencies are aware of their audience: Article III judges who are general­ists. 367 See id. While these judges are, by virtue of their expertise, more tolerant of complexity and more willing to “forgive certain lapses in linguis­tic clarity” than the general public, judges also appear to be skepti­cal of anything that seems like avoidable obfuscation. 368 Id.

The hard look tradition thus stands in direct tension with any sup­posed perverse incentives that might be created by Auer deference, and the bulk of the evidence would seem to suggest it is a far greater concern to agency officials as they take on the task of trying to write rules. 369 See supra section IV.C.2; see also Walker, Statutory Interpretation, supra note 149, at 1019 fig.1 (reporting survey data revealing an unimpressive awareness of Auer def­erence among agency respondents). Even if agencies are also aware of Auer’s purported incentives, the considerable short-term benefits of specificity and clarity as a means of avoiding costly vacaturs are more likely to determine the behavior of boundedly rational agency officials than the remote possibility of boot­strap­ping some addi­tional discretion down the road.


For Supreme Court watchers, one of the surest bets today might be to bet against Auer deference. With at least five of the Justices on the rec­ord as opposing Auer deference or other forms of deference to agency action, 370 See supra notes 155–159 and accompanying text. and with “continuing controversy” over Auer raging in the acad­emy, it seems likely that the Court will use Kisor v. Wilkie as a vehicle to do away with Auer deference, possibly replacing it with a lesser form of deference like Skidmore deference or eliminating deference to agency regulatory interpretations altogether. 371 See supra note 19 and accompanying text; see also Gillian E. Metzger, The Supreme Court, 2016 Term—Foreword: 1930s Redux: The Administrative State Under Siege, 131 Harv. L. Rev. 1, 26 (2017) (noting, presciently, “that the Court will likely address Auer’s scope and propriety in coming Terms”). Yet this assault on Auer is unwar­ranted, even on the terms laid out by Auer’s critics.

At the core of the emerging assault on Auer lies a claim about the doc­trine’s effect on agency officials’ incentives to promulgate vague rules that expand agency discretion. 372 Christopher J. Walker, Attacking Auer and Chevron Deference: A Literature Review, 16 Geo. J.L. & Pub. Pol’y 103, 106–07 (2018) (noting that one “predominant” argument against Auer is the claim that the “combination of law-making and law-executing authority creates inappropriate incentives for agencies to draft vague regulations and interpret those regulations through less-formal means after the fact”). Yet until now, that claim has never been tested. Despite looking carefully for any trace of changed rule-writing behavior in the aftermath of Auer and other Auer-related cases, I have found no empirical evidence that agencies respond to Auer’s rule-writing incentives in any systematic way. Further, I uncovered some limited evidence that agencies have actually shifted toward greater rule clarity over the years. This Article also identifies pressing, short-term consid­erations, such as the risk of hard look vacatur, that explain why any supposed long-term incentives created by Auer would hold little sway. The research presented here under­mines the self-delegation thesis upon which the critique of Auer has been grounded. If nothing else, it clearly shifts the burden of proof to Auer’s critics if they are to continue citing Auer’s perverse incentives as the basis for overturning or limiting the doctrine. As this Article has shown, it is not simply that Auer’s critics have committed the “sign fal­lacy,” as Sunstein and Vermeule put it, 373 See supra note 153 and accompanying text. but that the self-delegation the­sis is, as far as the best available evidence goes, empirically unfounded and in tension with institutional research on administrative agencies. For all of the attention it has received, the self-delegation thesis turns out just to be a false alarm.

The find­ings reported in this Article should give the Justices signifi­cant pause as they consider what to do in Kisor. Auer deference purports to serve important purposes in administrative law—most notably, ensuring that agencies have flexibility to clarify the law they have written. 374 See supra notes 104–111 and accompanying text. Critics of Auer believe that much of this need for ex post clarification is unnecessary if rules are written clearly in the first instance. They claim that agency officials operating in a world without Auer would embrace ex ante clarity to a much greater degree than they do now. 375 See supra section II.A. The evidence amassed in this Article offers a window into this counterfactual world by examining rule writing before Auer’s ascension and comparing it with rule writing afterward. There is no evidence that agencies were any more likely to front-load specificity and clarity before the Court decided Auer. 376 See supra sections III.B–.E. Consequently, were the Court to scale back Auer deference in Kisor, it would presumably only sacrifice Auer’s current benefits with no guarantee of any offsetting or countervail­ing benefits. 377 Cf. Nielson, Beyond Seminole Rock, supra note 85, at 950 (noting that overturning Auer would have unintended consequences, and that this “cuts in favor of stare decisis”).

In addition, the fact that agencies are not self-delegating en masse by increas­ing the vagueness of their rules suggests that there exist outer bounda­ries on their ability to stretch the meaning of regulatory texts in a manner that would be unfair to regulated parties. If the existing rule texts being interpreted have not been systematically skewed toward vague­ness by Auer’s alleged incentives, and if agency officials must still be able to make a plausible argument that these rules in some way support the agency’s interpretation, there will remain inherent limits on agen­cies’ ability to avoid accountability by couching every policy change as an interpretation of a preexisting rule. 378 Cf. Vermeule, Law’s Abegnation, supra note 136, at 80 (“It is a simple confusion to suggest an agency could ever ‘delegate power to itself.’ Agencies just have whatever quantum of power they have, under relevant statutory grants of authority . . . . Judges can always enforce the outer boundaries of the agency’s grant of authority, however exercised.”). Agencies in this world—that is, the world we live in—only have whatever amount of interpretive discretion they have, given the difficult tradeoffs involved in determining the opti­mal precision of rules in the first instance. 379 See Diver, supra note 128, at 71–72 (framing a discussion on agencies’ optimal degree of specificity in rule writing). Agencies also must comply with the anti-parroting principle by adding some modicum of specificity beyond the original statutory delegation of authority. See Gonzales v. Oregon, 546 U.S. 243, 257 (2006) (establishing the anti-parroting principle); see also supra note 33. Furthermore, the inter­preta­tions that stem from existing regulations must at a minimum be reasonable (even if they need not be the only possible interpretation). 380 Auer v. Robbins, 519 U.S. 452, 461 (1997); see also Decker v. Nw. Envtl. Def. Ctr., 568 U.S. 597, 613 (2013) (“It is well established that an agency’s interpretation need not be the only possible reading of a regulation—or even the best one—to prevail.”).

Auer’s critics might counter that the mere availability of the self-delega­tion strategy is itself a constitutional harm regardless of whether agencies actually use it. This argument proves too much. The U.S. constitu­tional system is rife with opportunities for potential abuse, and the Framers’ attitude toward these potential avenues was not to eliminate them but to design institutions and specific textual bars that would make abuse less likely to occur. 381 See The Federalist No. 51, at 322 (James Madison) (Clinton Rossiter ed., 1961) (“If angels were to govern men, neither external nor internal controls on government would be necessary. In framing a government which is to be administered by men over men, . . . you must first enable the government to control the governed; and in the next place oblige it to control itself.”). The combination of evidence that agencies do not engage in self-delegation and that they apparently have at least some limits on their ability to operate within the bounds of the regula­tions they have written should be enough to quell any concern that exist­ing legal structures are failing in their fundamental task. 382 It is worth noting that Manning’s later writings on the separation of powers are in some tension with the self-delegation critique. Manning has argued that there is no “free­standing principle of separation of powers” and that formalists ought to confine their argu­ments to textually grounded separation of powers limitations. John F. Manning, Separation of Powers as Ordinary Interpretation, 124 Harv. L. Rev. 1939, 1944 (2011) [hereinafter Manning, Separation of Powers] (emphasis omitted); see also William N. Eskridge, Jr., Relationships Between Formalism and Functionalism in Separation of Powers Cases, 22 Harv. J.L. & Pub. Pol’y 21, 29 (1998) (arguing that “we ought not consider functionalism and formalism as inevitably antipodal” because both in practice make conse­quentialist arguments about particular arrangements of institutional power). This insis­tence that separation of powers arguments be grounded in the text of the Constitution, lest they simply become another form of metafunctionalist theory, runs into problems when, as in the case of Auer, there is no constitutional text that is directly relevant to the combi­nation of powers in administrative agencies. To reconcile the self-delegation critique with this more textualist approach to formalism, one would have to be skeptical of much more than just Auer in administrative law, essentially reading the vesting clauses as plain bars on any combination of functions in agencies. See supra note 126 and accompanying text.

In fact, the research findings presented here hold more general implica­tions for debates about the place of administrative agencies in the separation of powers framework. Separation of powers formalists 383 See Manning, Separation of Powers, supra note 382, at 1958 (“Formalist theory presupposes that the constitutional separation of powers establishes readily ascertainable and enforceable rules of separation.”); Thomas W. Merrill, The Constitutional Principle of Separation of Powers, 1991 Sup. Ct. Rev. 225, 226 (describing formalism as “emphasizing that ‘[t]he Constitution sought to divide the delegated powers of the new Federal govern­ment into three defined categories, Legislative, Executive, and Judicial, to assure, as nearly as possible, that each branch of government would confine itself to its assigned respon­sibility’” (alteration in original) (quoting INS v. Chadha, 462 U.S. 919, 951 (1983))). fre­quently assert that the breakdown of the constitutional division of author­ity in the administrative state will (or at least probably will) have far-reaching consequences for limits on government power. 384 See Yoo & Phillips, supra note 16 (“Our constitutional system has the genius of diffusing power among three branches of the national government . . . . [F]orsaking the Founders’ limits on government in the name of administrative ease would bring far more ruin than failing to attend to the popular policy of the day.”) The constitu­tional critique of Auer, grounded as it is in the self-delegation thesis, is a primary example, but there are many other such arguments that institutions or doctrines that purportedly relax the strict separation of powers promote certain undesirable behavior. 385 See supra notes 30–32 and accompanying text. Such arguments tend to catch on, in part because they rely on intuitive models of rational behav­ior geared toward power seeking and hint at a relatively simple solu­tion: restoring some kind of strict separation, or at least greatly diminish­ing the scope of administrative agencies’ authority, in an effort to eliminate some asserted perverse incentives. But the analysis in this Article reveals the limitations of relying on intuitions about the combina­tion of authorities and powers in agencies. Arguments that are unteth­ered from evidence and from an understanding of agencies as institu­tions may be intuitively appealing but utterly mistaken. Any realistic account of structural or doctrinal incentives must therefore grapple with what actually motivates bureaucrats, drawing support from what empiri­cal studies reveal about the plausibility of intuitions derived from theory. 386 See supra Part IV.

This observation carries special importance at a moment of heated debate about the future of the administrative state in the U.S. constitu­tional system. As Professor Gillian Metzger recently noted, the administra­tive state currently appears to be in the midst of crisis, ensnared between attacks from insurgent “anti-administrativists” and defenses from equally impassioned supporters of the administrative state. 387 Metzger, supra note 371, at 8–46. Auer is a primary exhibit for the kinds of attacks being levied by principled anti-administrativists: It is a purported example of the admin­is­trative state being predisposed or programmed by administrative law to “lose[ ] its way.” 388 Aaron L. Nielson, Response, Confessions of an “Anti-Administrativist,” 131 Harv. L. Rev. Forum 1, 1 (2017),
vol131_Nielson.pdf [].
Auer is thought to be low-hanging fruit in this regard 389 To be clear, Nielson himself does not appear to endorse all of the assault on Auer—for reasons unrelated to the validity of the self-delegation thesis. See Nielson, Beyond Seminole Rock, supra note 85. —so low, in fact, that the empirical basis for the self-delegation thesis has been taken for granted, almost as a matter of intuition. But intuition needs verification. Despite looking with some of the most powerful meth­ods available and considering different possible mani­festations, I have been unable to detect any effect on agency rule writing associated with Auer deference. 390 See supra Part III. Moreover, the lack of evidence of any agency self-delegation in the rulemaking process should hardly be surprising, if one only considers what an extensive body of social science research teaches about agency officials’ behavioral tendencies and what administrative law scholarship has to say about the immediate legal concerns confronting agency decisionmakers. 391 See supra Part IV. The administrative state is an enormously com­plex organism, and administrative law is not the only constraint on, or facilitator of, agency discretion. 392 Coglianese, Rhetoric and Reality, supra note 27, at 95 (“[A]dministrative law is embedded within a complex web of politics, institutions, and organizational behavior. Within this web, law is but one factor influencing behavior in government agencies among a variety of institutional, professional, social, financial, and political factors that interact with each other, and . . . adapt and change over time.”). If it is as difficult to demonstrate that agencies have gone off course in other areas as it has been in the seem­ingly easy case of Auer, then the implication is that pat assumptions about administrative law’s behavioral incentives should not be enough on their own to justify calls for reform.