The Limits of Judicial Control and the Nondelegation Doctrine

Abstract

The nondelegation doctrine has been fought over for decades, yet scholars have not examined a foundational question: can judicial doctrine materially shape legislative drafting practices? Even if a strong nondelegation doctrine provides legislators an incentive to draft narrow statutes, they would have many reasons to persist in broad delegations, and it is not clear whether the doctrinal incentives predominate. Here, I examine the relationship between the nondelegation doctrine and lawmaking behavior at the state level using several novel datasets, including a collection of state session laws between 1990 and 2010, and a comprehensive survey of state nondelegation judicial decisions over the last 20 years. Contrary to the common assumption, I find that the robustness of the nondelegation doctrine appears essentially unrelated to legislative drafting practices. This pattern suggests the limited extent to which judicial doctrine can control legislative practices; it also suggests a revived nondelegation doctrine at the federal level is unlikely to effectuate the hopes of proponents or the fears of opponents.

1. Introduction

The nondelegation doctrine is a central figure in ongoing debates over the constitutional powers of the legislature and in turn administrative agencies. Though the Court has not relied on the doctrine as the grounds for striking down a statute since the New Deal, lower courts have done so in recent years,¹ and commentators often call for a revival of a more robust nondelegation doctrine.² According to proponents, the benefits that would flow from a revived doctrine include enhanced political accountability (e.g., Ely 1980; Fiorina 1982),³ and the promotion of rule of law values (Lowi 1969), individual liberty,⁴ and fidelity to constitutional text (e.g., Schoenbrod 2008).

For any of these benefits to confer, however, the legislature must conform to the requirements of a stronger doctrine. The legislature, that is, must delegate less, and when it does delegate, it must do so in more precise terms.⁵ It is commonly assumed that legislative practices would follow the commands of judicial doctrine. Posner and Vermuele, for instance, observe the standard view that “[u]nder the non-delegation doctrine, Congress will enact narrow statutes in order to avoid reversal by the courts” (2002: 1761). Yet this fundamental legislative predicate of the nondelegation doctrine has largely escaped theoretical consideration, and it has entirely escaped empirical scrutiny.⁶ That is, despite the presuppositions of the literature, we have no evidence at all suggesting that legislatures might respond to the nondelegation doctrine by delegating less or drafting more specific statutes.

As a theoretical matter, it is moreover quite plausible that a stronger nondelegation doctrine in fact would have little influence over legislative drafting practices. One way of understanding the standard view is that a stronger nondelegation doctrine provides an overwhelming incentive for legislators to conform by drafting fewer, more specific statutes. But even if the fear of future invalidation does provide the legislature with an anti-delegation incentive, administrative, and political constraints imply that the courts will not police every delegation (Stiglitz 2018).⁷ Further, legislators will of course continue to weigh the many forces that compelled them to delegate in broad terms in the first instance: the limited capacity and plenary time of the legislature (e.g., Cox 2006), the complexity of problems at hand and the expertise of agencies (classically, Landis 1938; Epstein and O’Halloran 1999), various problems of time consistency (e.g., Rogoff 1985; de Figueiredo 2002), an electoral incentive to build trust with the public (Stiglitz 2018), and perhaps even the desire to avoid electoral blame for any ill effects of policy choices (e.g., Ely 1980; Fiorina 1982).⁸ How legislators weigh these factors is unclear a priori—theory provides little guidance—and represents an important empirical question given the consistent calls for a more robust nondelegation doctrine.

The main objective of this article is to question and empirically examine the equation of a nondelegation doctrine with legislative practices. Critically, do the delegating statutes that emerge from political systems with strong nondelegation doctrines appear different from delegating statutes that emerge from political systems with weak nondelegation doctrines? Do changes in the nondelegation doctrine in these political systems induces changes in legislative drafting practices? This article addresses the surprising silence of the literature on these questions by providing the first empirical examination of the relationship between the doctrine and drafting practices.

Following the spirit of Justice Brandeis, I examine the nondelegation doctrine in the American states.⁹ The great virtue of this state-based approach is that it allows us to examine how the nondelegation doctrine plays out in reasonably similar settings: though the constitutional details of the states differ a good deal (that is why they hold interest), the states exist within a federal system that constrains the extent to which they may differ. The guarantee clause, for instance, plausibly places limits the differences among states (Dorf 1998), and the states further share a common history that induces a measure of homogeneity in institutions (e.g., Acemoglu et al. 2002). This intermediate level of variation, sufficient to hold interest, but not so great as to render comparisons incredible, is ideal for examining the constitutional debates that consume us at the federal level yet which cannot be answered by referring only the federal constitution and government.

The findings from these exercises tend to upset the foundations of the case for the nondelegation doctrine. Across a host of drafting metrics relevant to the nondelegation doctrine and validated against congressional statutes from the New Deal era, the presence or absence of a strong nondelegation doctrine appears essentially unrelated to the drafting practices in state legislatures. For example, we do not see legislatures in states with a strong nondelegation doctrine fashioning statutes with less precatory language. Similarly, I find that changes in the nondelegation doctrine over the last 20 years induce virtually no legislative responses that manifest in the most obvious forms of legislative drafting patterns. These results survive a series of standard robustness checks and probes of assumptions. This pattern undercuts much of the debate over the contentious nondelegation doctrine. Toward the end of the article, I speculate on the reason why the nondelegation doctrine, a dominant topic in public law and the subject of heated debate for decades, appears to matter so little in practice.

This article proceeds as follows. First, I discuss the history of the nondelegation doctrine, with particular reference to the states, presenting some simple summary statistics of states with different nondelegation regimes. Second, I conduct the article’s main exercise, examining the relationship between the nondelegation doctrine and the drafting practices in state legislatures. For this exercise, I examine both a preexisting coding of state delegation regimes, as well as a novel dynamic coding scheme that accounts for changes in doctrine over time. My conclusions discuss the results from these exercises and connect the findings with recent literature on Congress-Court feedback loops (Gluck and Bressman 2013; Bressman and Gluck 2014).

2. Nondelegation in the States

2.1 Doctrinal Typology

The nondelegation doctrine, in basic form, maintains that the legislature cannot delegate legislative power to other entities. Constitutionally, the foundation for this position is Article I vesting clause, providing in relevant part that “All legislative Powers herein granted shall be vested in a Congress of the United States.”¹⁰ This textual position, proponents argue, is supported by background common law traditions perhaps extending to Bracton and Coke (Duff and Whiteside 1928; see also Hamburger 2014), and serves functional values associated with the rule of law (e.g., Lowi 1969) and political accountability (e.g., Ely 1980). The doctrine is rarely used at the federal level, and in fact the Court has not invalidated a statute on this basis since 1935, when the Court struck down parts of the New Deal as unconstitutional delegations of legislative power.¹¹

Yet even if the doctrine is essentially silent at the federal level, at least in an overt sense, it continues to thrive in many states. Other scholars have examined the nondelegation doctrine in the states (most notably, Greco 1993; Rossi 1999), adopting the following basic typology of nondelegation doctrines. The first type of nondelegation regime approximates the strict nondelegation doctrine expressed in Schechter, envisioning agencies as merely “filling in of the details” (Jaffe 1947) rather than “lawmaking.” For this set of states, the statutory standards for administrative action set forth by the legislature must be relatively precise and specific. The most notable states that fall in this category include New York, Florida, and Texas, where courts have repeatedly invalidated statutes as excessive delegations. Under the second type of nondelegation regime, it is constitutionally permissible for the legislature to delegate under somewhat looser and less specific statutory standards for administrative action, effectuating a shift toward administrative discretion. A final type of nondelegation doctrine, sometimes referred to as the “new” nondelegation doctrine (Seidenfeld and Rossi 2000), focuses on procedures and safeguards established by the agency rather than on statutory standards set forth by the legislature.¹² This last set of states, which include Oregon and California, for example, has arguably the most permissive approach to the nondelegation doctrine.

I reproduce the Greco/Rossi (GR) classification of states in Figure 1.¹³ For the opening parts of this article, I adopt this GR classification, though for simplicity in the analysis to follow I collapse the “loose standards” and “procedural safeguards” categories.¹⁴ The classification, therefore, consists of two groups of states: one set with demanding nondelegation doctrines approaching the version sought after by various scholars; and the other set with nondelegations that they would find wanting for failing to provide sufficiently rigorous standards. As is evident from the figure, a fair number of states maintain a strong nondelegation doctrine—19 states, that is, roughly 40% of states.

Figure 1.

Nondelegation Doctrine in the States.

2.2 Nondelegation Doctrine and State Characteristics

Based on the history of the nondelegation at the federal level, one has the prior belief of the doctrine surviving principally in polities with small economies and perhaps with conservative political cultures. However, even a casual examination of the set of states in each category belies this belief. A somewhat more systematic assessment supports this view. As reported in Table 1, states with a strong nondelegation doctrine do not, on average, in fact appear to differ materially from states with a weaker nondelegation doctrine. There, I report the average values for a number of key social, political and economic indicators, broken down by the states’ nondelegation regime. From the left to right-most column, I report the relevant value for the strong nondelegation states, the weak nondelegation states, the difference between the strong and weak nondelegation states, and the p-value associated with this difference in means.

Table 1.

Nondelegation Doctrine and State Characteristics

	Nondelegation regime
	Strong	Weak	Difference	p-value
Income (log) (2010)	9.75	9.79	−0.04	0.38
GDP (log) (2010)	11.38	11.21	0.17	0.58
Population (log) (2010)	15.30	15.06	0.24	0.42
Unemployment (2010)	8.51	8.87	−0.36	0.56
GDP Growth (1950–2010)	0.03	0.03	0.00	0.54
McCain Vote Share (2008)	0.51	0.47	0.04	0.20
Presidential Rep. Vote Share (1952–2008)	0.54	0.52	0.02	0.16
Legislative Professionalism Index (1979–2010)	0.20	0.20	0.00	0.99
Legislative Professionalism Rank (1979–2010)	25.99	25.18	0.81	0.85
Number of House Seats (2010)	123.56	102.81	20.75	0.31
Number of Senate Seats (2010)	39.05	39.65	−0.60	0.85
Number of Words in State Cons’t	10.09	10.13	−0.04	0.86
Proportion w/ Term Limits	0.32	0.29	0.03	0.85
Year of Statehood	1838.84	1836.90	1.94	0.89

For example, according to the economic indicators, it does not appear that the nondelegation doctrine survives only in states with small economies that do not much require the modern administrative apparatus: neither state income nor state GDP, in 2010, differs by delegation regime; nor do the states differ materially in terms of population by nondelegation regime; or in terms of unemployment rate; further, over the last 60 years, the strong and weak nondelegation regime states do not have significantly different growth rates—on average, the economies of both types of states grew at about 3% annually over the last 60 years. On the political dimension, though strong nondelegation states appear slightly more conservative than weak nondelegation states, the difference is small in magnitude and is not statistically significant. During the 2008 presidential election, McCain received about 51% of the vote in states with strong nondelegation doctrine; by comparison, McCain received about 47% of the vote in states with a weak nondelegation doctrine. Examining the average Republican presidential vote share in the states between 1952 and 2008, the spread between the nondelegation regime tightens to two percentage points (and remains nonsignificant). Finally, institutionally, the two sets of states do not generally appear very different. For instance, on a standard measure of legislative professionalism, which takes into account factors such as legislators’ salaries, the size of the legislative staff and other resources available to members, and the time spent in session, the average rank of the two sets of states is straight down the middle: the average rank of a state with a strong nondelegation doctrine is 25, just as it is for states with a weak nondelegation doctrine.¹⁵ Along these same lines, the size of the legislatures are roughly the same on average in the two sets of states; the constitutions have about the same number of words; the proportion of states with term limits is also about the same; both sets of states received statehood around the same time as well.

Of course, these summary statistics do not represent causal estimates. Still, the fact that across a range of social, political, and economic indicators, we observe no meaningful difference between states with strong and weak nondelegation regimes is striking and highly suggestive. The long-standing debate over the nondelegation doctrine is motivated on the premise that much turns on the presence or absence of the nondelegation doctrine. It is therefore of considerable note that the same nondelegation regime, weak or strong, exists in roughly the same measure in different types of states, rich or poor, liberal or conservative, large or small.

3. Nondelegation and the Legislative Predicate

Now consider the central question of the relationship between the nondelegation doctrine and legislative drafting practices. The functional virtues that flow from the doctrine, such as enhanced political accountability, are predicated on the idea that legislatures respond to the doctrine by drafting more specific statutes. Yet this point is assumed, and we have no evidence on this essential premise of the debate over the doctrine. Here, I examine legislative behavior and drafting practices in the states according to their nondelegation regime.

3.1 Lawmaking Data and Delegation Metrics

I examine lawmaking behavior in the 50 states between 1990 and 2010.¹⁶ Before addressing any differences between strong and weak nondelegation states, consider some basic features of lawmaking in the states. Combined, the state lawmaking data contains some 400,000 session laws, approaching in aggregate one billion words, over the 20 year series considered. Of these, roughly 61,000 session laws, or roughly 17% of the total, appear to delegate lawmaking authority to administrative agencies.¹⁷ However, the variation in lawmaking behavior in the American states is truly astounding. Over the 20 year series, legislative productivity—that is, initially, the total number of session laws produced—ranges from roughly 2000 in Ohio, to approximately 20,000 in California. The length of each session law also varies considerably, with, again on average over the series, South Dakota drafting laws of under 1000 words, and Ohio drafting laws of over 17,000 words, or roughly 70 pages of double space text.

The fact that Ohio drafts a few, long laws, and other states, such as Alabama, draft many, short statutes suggests that much of this variation may be explained by different practices in the use of omnibus legislation.¹⁸ Another metric of drafting practices, which effectively adjusts for omnibus drafting practices, is the total number of legislative words produced in a given year. By this metric, the variation remains substantial, though the portrait of legislative activity changes considerably. The least productive states, by this metric, include Wyoming, South Dakota, and Alaska, each drafting, on average, fewer than 300,000 words of legislation in a given year. By comparison, the most productive states, each drafting, on average, over 1,500,000 words of legislation in a given year, include California, Illinois, Virginia, Ohio, Florida, New York, and Texas.¹⁹

The analysis focuses on two dimensions of delegation: first, the incidence of delegations; and, second, the qualitative features of the delegations.²⁰ Consider first metrics related to the level of delegating activity. One metric is the proportion of statutes that delegate authority to administrative agencies.²¹ A second, closely related measure is the number of delegating statutes occurring in a state-year. A third metric of delegating activity is the number of grants of rulemaking authority. A virtue of this last metric is that we can aggregate over statutes in a state-year, making it less sensitive to differences in practices with respect to omnibus legislation.

Now consider metrics related to the qualitative nature of the delegations—that is, their level of specificity, or the level of constraint that they impose on agencies with delegated authority. As in the measures related to the quantity of delegations, no standard measurement approach exists, and I therefore adopt a variety of metrics. As a first measure, approximately following Huber et al. (2001) and Huber and Shipan (2002), I consider the number of words per delegation of rulemaking authority as relevant to its specificity.²² The motivation for this metric is that statutory controls manifest themselves, in part, in the length of the statute; though increased statutory controls might not always transmit to increased statutory length, the idea is that, in general, more detailed statutory requirements require more words to express. A second metric focuses, instead, on linguistic indicators of broad delegations. We can examine the prevalence of “precatory” words associated with permissive delegations of authority, under the view that statutes that use such terms tend to have broader delegations of authority.²³ The set of relevant words derives from a collection of key words in federal statutes identified in Judge Tatel’s American Trucking dissent, which in his judgment represent broad delegations of legislative authority, augmented by a set of permissive words identified by Mashaw (1997).²⁴ Again, though such linguistic features may not invariably denote broad authority, the theory of the measure is that they tend to do so and therefore correlate with the broadness of the delegation. A third metric examines the structure of statutes. In particular, I consider the prevalence of definitions in delegating statutes, under the view that more definitions indicate more statutory specificity.²⁵

To summarize and further clarify, I consider the following drafting metrics as relevant to the nondelegation doctrine.

1. The percent of laws passed containing a delegation of authority in a state-year. That is, $\frac{\left|D_{st}\right|}{\left|A_{st}\right|}$⁠, where D_st is the set of delegating laws passed in state s during year t, and A_st is the corresponding set of all laws passed in that state-year.

2. The number of delegating laws passed in a state-year. That is, $\left|D_{st}\right|$⁠.

3. The number of grants of rulemaking authority in a state-year. That is, $\sum _{i\in D_{st}}{r}_{ist}$⁠, where r_its is the number of grants of rulemaking authority contained in statute i in the indexed state-year.

4. The number of words contained in delegating laws passed in a state-year per grant of rulemaking authority. That is, $\frac{\sum _{i\in D_{st}}{w}_{ist}}{1000\sum _{i\in D_{st}}{r}_{ist}}$⁠, where w_ist is the number of words contained in statute i in the indexed state and year.²⁶

5. The density of “permissive” or “precatory” words in delegating laws passed in a state-year. That is, $\frac{1000\sum _{i\in D_{st}}{p}_{ist}}{\sum _{i\in D_{st}}{w}_{ist}}$⁠, where p_its is the number of precatory words in statute i in the indexed state and year.²⁷

6. The density of definitions in delegating laws passed in a state-year. That is, $\frac{1000\sum _{i\in D_{st}}{d}_{ist}}{\sum _{i\in D_{st}}{w}_{ist}}$⁠, where d_ist is the number of definitions contained in statute i in the indexed state and year.²⁸

Notice that the first three of these metrics relate to the level of delegating activity, whereas the second three relate to the qualitative nature of the delegations. The nondelegation doctrine is most plausibly thought of both reducing the quantity of delegations, as well as increasing the specificity of delegations when they do occur. I report summary statistics on the drafting outcomes of interest in Table 2.²⁹

Table 2.

Drafting Metrics: Summary Statistics

	Mean	Std. Dev.
Percent Laws Delegating	0.17	0.09
Number of Delegating Laws	61.69	43.6
Number of Rulemaking Provisions	159.79	145.67
Words (1000s) per Rulemaking Authority	3.19	1.48
Density of Permissive Language	7.75	1.23
Density of Statutory Definitions	1.21	0.76

See main text and Online Appendix for detailed definitions of these legislative drafting variables.

3.2 Validation Exercise: New Deal Legislation

One way to validate these delegation metrics is to deploy them against a period of lawmaking during which it is uncontroversial that a legislature changed its delegating practices. To the extent the metrics are valid, they will detect this historical transition. I conduct this exercise with respect to New Deal legislation.³⁰ “In the now-familiar understanding of the legislative role of the [New Deal] period,” one scholar characterizes the consensus, “Congress often restricted itself to identifying a problem and requesting that an agency develop a solution” (Sunstein 1987). As indicated below, this exercise indicates that most of the metrics reflect this common understanding.³¹

In Figure 2, I plot the proposed delegation metrics as applied to federal legislation between 1929 and 1938. This series, therefore, starts with the Hoover administration, around the time of the 1929 Wall Street Crash, and continues through the end of the period during which Congress passed the major New Deal legislation. A vertical, dashed line highlights this transition in the figures, as we move out of the Hoover administration and into FDR’s New Deal. The critical question is whether the delegation metrics detect the change in delegating behavior so often attributed to the New Deal.

Figure 2.

Federal Delegations and the New Deal.

Consider first the metrics relating to the level of delegating activity. First, as depicted in the top left panel, the percentage of laws that delegate authority increases dramatically as we move into the New Deal. In the years immediately preceding the New Deal, Congress delegated in about 5.5% of laws; during the New Deal, Congress delegated in about 11% of laws, and over 15% in the 1933–34 period.³² The second measure, in the top right panel, detects this same pattern: the number of laws passed by Congress each year containing delegations increased from about 25 in the pre-New Deal period to around 50 in the New Deal period.³³ The third measure of activity levels is essentially congruent: the number of provisions granting rulemaking authority increased from an average of 74 per year in the 1929–1932 period to an average of 139 per year in the 1933–1938 period.³⁴ These measures, therefore, line up with the consensus position that the quantity of delegations increased dramatically during the New Deal; this suggests that these measures detect substantively relevant aspects of lawmaking behavior.

Of the three measures relating to the qualitative nature of the delegations, two detect the expected patterns in New Deal legislation. The first qualitative measure—the length of the delegating laws—suggests that Congress drafted considerably shorter delegating laws during the New Deal period relative to those passed during the Hoover administration. The average delegating law passed during the Hoover administration contained roughly 3000 words per grant of rulemaking authority; by comparison, the average New Deal delegating law contained only 2250 words per authority, a decrease of roughly 25%.³⁵ This pattern fits with the notion that Congress included relatively few qualifications or conditions on its New Deal delegations. The second measure—the density of precatory language—also indicates that delegations became more permissive in the New Deal era. Congress appears to have used precatory language much more commonly in the New Deal era than during the Hoover administration: between 1929 and 1932 Congress used 7.74 precatory words per 1000 words of delegating legislation produced; during the New Deal, Congress used 8.72 precatory words per 1000 words of delegating legislation, an increase of roughly 13%.³⁶ The final qualitative measure of delegations—the density of statutory definitions—in fact runs in the opposite direction as expected.³⁷ That is, the density of definitions appears to increase during the New Deal period. However, the prevalence of definitions may reflect the novelty of the New Deal regulatory programs, and I report results relating to the density of definitions in the analysis below for completeness.

Altogether, this brief tour of New Deal drafting practices suggests that the metrics identified above bear meaningfully on the nondelegation doctrine. If the New Deal is a prime example of a period characterized by a transition in delegating behavior, the metrics reflect this fact: we see marked changes in the percent of laws delegating authority, the number of laws delegating authority, and in the number of provisions granting rulemaking authority, all suggesting the validity of these activity measures; we also see changes in the length of delegating laws and in the density of precatory language, both suggesting that the measures relating to the qualitative nature of delegations have substantive relevance.

3.3 Doctrine and Lawmaking: Initial Assessments

Several impediments complicate an understanding of the relationship between the nondelegation doctrine and legislative behavior. For instance, it may be that some difficult-to-observe aspect of the structure of the state economy, or the presence of some other state institution,³⁸ or perhaps a dimension of state political culture, drives both drafting practices and the state courts’ nondelegation doctrines. Similarly, we must guard against the possibility that the doctrine arises endogenously to state institutional or political forces.³⁹

As an initial pass of the data, however, it is useful to consider the simple relationship between nondelegation regime and drafting practices, without first attempting to account for such complications. I examine this relationship in a regression framework,

$$L_{st}=\alpha +\gamma D_{st}+{\mathrm{\epsilon }}_{st},$$

(1)

where L_st is some measure of drafting practices in state s at time t, and D_st is an indicator taking a “1” if state s in year t has a strong nondelegation doctrine, as coded by GR, and a “0” otherwise. Later, I add covariates to this regression in an attempt to control for possible omitted variables that plausibly drive legislative drafting practices and might also be correlated with nondelegation regimes. Our interest is in the γ coefficient, which informs us whether the indicator for a strong nondelegation doctrine is correlated with with drafting practices—that is, it tells us if states with a strong nondelegation regime tend to have fewer delegating laws and laws written with more specificity.

This simple exercise produces six models, one for each legislative drafting outcome, reported in columns 1, 4, 7, 10, 13, and 16 of Table 3 (columns labeled by model number, i.e., “M1,” “M4,” and so on). The coefficients on the strong nondelegation indicator reveal that the doctrinal regime is generally associated with a slightly higher proportion of delegating statutes (M1), but fewer total delegating laws (M4), more rulemaking provisions (M7), longer session laws per grant of authority (M10), fewer permissive words (M13), and session laws with a greater number of definitions (M16). However, none of these coefficients attain statistical significance, and moreover we have yet to control for various possible confounding political and economic covariates. Also notable is the fact that the fit of these models tends to be extremely poor; the nondelegation doctrine appears to explain only a small fraction of variation in drafting behavior in the states.

Table 3.

Nondelegation Doctrine and Drafting Practices: An Initial Assessment

	Percent delegating			Number delegating			Rulemaking provisions
	M1	M2	M3	M4	M5	M6	M7	M8	M9
Nondelegation	0.04	0.03	0.03	−1.78	−6.33	−6.67	35.22	18.16	16.96
	(0.02)	(0.02)	(0.02)	(9.58)	(9.73)	(9.7)	(36.54)	(34.33)	(34.66)
Covariates	N	Y	Y	N	Y	Y	N	Y	Y
Year FE	N	N	Y	N	N	Y	N	N	Y
N	1005	984	984	996	975	975	996	975	975
R2	0.04	0.13	0.15	0	0.07	0.11	0.01	0.13	0.15
	Length of Laws			Permissiveness			Definitions
	M10	M11	M12	M13	M14	M15	M16	M17	M18
Nondelegation	0.09	0.04	−0.01	−0.21	−0.18	−0.17	0.03	0.04	0.03
	(0.31)	(0.31)	(0.31)	(0.23)	(0.25)	(0.25)	(0.21)	(0.22)	(0.22)
Covariates	N	Y	Y	N	Y	Y	N	Y	Y
Year FE	N	N	Y	N	N	Y	N	N	Y
N	996	975	975	996	975	975	996	975	975
R2	0	0.03	0.07	0.01	0.05	0.06	0	0.05	0.06

Notes: Standard errors clustered by state and reported in parentheses. Dependent variable in models 1–3 is the proportion of laws in the state-year that delegate rulemaking authority; in models 4–6 the number of laws that delegate authority in the state-year; in models 7–9, the number of total rulemaking provisions contained in laws passed in the state-year, among laws delegating authority; in models 10–12 the mean number of words in laws containing delegating language in the state year per grant of rulemaking authority; in models 13–15 the number of “permissive” words per one thousand words in delegating statutes in the state-year; in models 16–18 the number of definitions per one thousand words in delegating statutes in the state-year. See text for precise definitions of these drafting outcomes. Covariates include: log personal income per capita, one year change in income, log population, one year change in population, log GDP per capita, one year change in GDP, unemployment, one year change in unemployment, an index of legislative professionalism, and an indicator for divided government.

Including covariates to account for the possibility that nondelegation regime is correlated with economic or political conditions that drive legislative behavior tends to have little effect on the sign of the relevant coefficients, as reported in columns 2, 5, 8, 11, 14, and 17 (again, denoted in the table with “M2,” “M5,” and so on), though they tend to attenuate somewhat in magnitude.⁴⁰ In all cases, the coefficient on the nondelegation doctrine remains statistically insignificant after introducing these covariates. Notice, however, that the fit of the models tends to improve markedly: for example, including political and economic covariates increases the r-squared on the regression for the number of grants of rulemaking authority to 0.13 (M8) from 0.01 (M7).

A final related exercise involves adding year fixed effects to these models, effectively accounting for unobservable political or economic trends common to all states in a given year.⁴¹ These year fixed effects do not alter the coefficients on the nondelegation doctrine indicator in any noteworthy way, as reported in models 3, 6, 9, 12, 15, and 18. The sign, rough magnitude, and significance of the coefficients remain for the most part unperturbed from those discussed immediately above.

As a whole, though, these initial results thus provide, at best, tepid support for the fundamental premise in the debate over the nondelegation doctrine. Although the sign of many of the coefficients runs in the expected direction—with the doctrine associated with fewer delegating laws, less permissive language, for example—the relationship is weak, and the sign of other coefficients run against doctrinal expectations.⁴²

3.4 Doctrine and Lawmaking: State Dynamics

The initial analysis has an obvious limitation: the measure of the nondelegation doctrine remains fixed throughout the series. This prevents us from examining the consequences of doctrinal changes within states, and from probing assumptions necessary to the credibility of the estimates reported above. Here, I address this limitation by comprehensively canvasing nondelegation judicial decisions in the states. These judicial decisions allow us to develop a measure of doctrinal “tightening” within states, facilitating a more rigorous examination of the relationship between the nondelegation doctrine and drafting practices.

3.4.1 State Invalidation Data

I define the domain of judicial decisions in the following manner. I first identify a universe of constitutional nondelegation cases by searching for state supreme court cases with Westlaw keynotes relating either to the statutory “standards for guidance” or the “delegation of powers: in general,” thus casting a fairly wide net. Between 1990 and 2010, the period under consideration, this returns a total of 163 cases. I then read each case to determine whether the court invalidated a statutory scheme as an unconstitutional delegation of legislative power. Over the 20 year period I consider for this series, I identify a total of 22 judicial invalidations of state statutes due to the nondelegation doctrine. In Figure 3, I show the number of invalidations among states with strong and weak nondelegation regimes.

Figure 3.

Nondelegation Invalidations in the States: 1990–2010.

Several observations fall out of these data and this figure. First, the solid majority of states—33—did not invalidate any state statutes in the series under consideration. Second, eleven of the seventeen states (65%) with at least one invalidation classify as having a strong nondelegation doctrine according to Greco (1993) and Rossi (1999), suggesting that the nondelegation doctrine promotes the invalidation of state laws. Consistent with this view, 25 of the 33 states (76%) with no invalidations classify as weak nondelegation states. Indeed, a strong nondelegation doctrine classification correlates with the number of judicial invalidations at roughly 0.45, substantively important and statistically significant at any conventional level. Third, a few states appear to be particularly active on the nondelegation front, with at least one invalidation per decade: Florida, Virginia, Texas, and Oklahoma. All of these states have strong nondelegation doctrines. Thus, in terms of the frequency of judicial invalidations, the nondelegation doctrine appears to matter—it should not be easily dismissed as mere aspirational formalism or empty judicial posturing.

Even if so, the question remains of whether the doctrine matters for legislative behavior, as we typically assume. As another way of examining the question, I first create a variable that starts at “0” for all states, and then increases by “1” for each invalidation in the state. For example, a state starts at “0” in 1990, and continues at that level until a judicial invalidation; if the state supreme court invalidates a statute in, say, 1995, I add “1” to the counter in 1995, with that level continuing until the next invalidation; and so on.⁴³ The idea behind this metric is to capture the possibility of a “tightening” of doctrine over time, under the theory that each invalidation represents some innovation, however incremental, over earlier doctrine.⁴⁴

3.4.2 Doctrine and Lawmaking: Assessing Doctrinal Dynamics

The question is whether legislative drafting practices adjust to doctrinal tightening. To investigate this question, I estimate the following,

$$L_{st}={\alpha }_s+{\psi }_t+\Gamma Z_{st}+\gamma D_{st}+{\mathrm{\epsilon }}_{st},$$

(2)

where L_st represents some metric relating to drafting practices, α_s is an (optional) state fixed effect, ψ_t is an (optional) year fixed effect, Z_st is an (optional) vector of political and economic covariates for state s and year t, and D_st is now the incremental indicator for the tightness of the nondelegation regime, based on cumulative invalidations, in state s at year t.⁴⁵ Notice that we now have state fixed effects, accounting for time invariant unobservable state characteristics, such as state culture or the presence of other state institutions that remain fixed over the period. Our primary interest is in the coefficient γ, which is informative about whether drafting practices respond to the tightening of the nondelegation doctrine.

I follow the approach of the previous section and examine a full spectrum of drafting practices, progressively adding political and economic covariates and state and year fixed effects, as reported in Table 4. Consider first the results without any controls. For these models, reported in columns 1, 5, 9, 13, 17, and 21 we see that when courts invalidate state statutes on nondelegation grounds, most indicators suggest that legislatures do not react in predicted ways. Indeed, if anything, results run contrary to expectations, with coefficients suggesting that doctrinal tightening results in more delegating laws (M5) and rulemaking provisions (M9), though no coefficient reaches statistical significance at the conventional level.

Table 4.

Changes in Nondelegation Doctrine and Changes in Drafting Practices

	Percent delegating				Number delegating
	M1	M2	M3	M4	M5	M6	M7	M8
Invalidations	0.01	0	−0.01	0	13.89	11.72	12.66	−2.7
	(0.01)	(0.01)	(0.01)	(0.01)	(8.03)	(8.13)	(8.35)	(4.26)
Lead (2-yr)	0.02	0	0.01	0.01	11.47	10.21	9.41	−3.27
	(0.02)	(0.02)	(0.02)	(0.01)	(10.7)	(10.21)	(9.5)	(3.63)
Covariates	N	Y	Y	Y	N	Y	Y	Y
Year FE	N	N	Y	Y	N	N	Y	Y
State FE	N	N	N	Y	N	N	N	Y
N	1005	984	984	984	996	975	975	975
R2	0.01	0.1	0.12	0.75	0.03	0.09	0.12	0.67
	Rulemaking provisions				Length of laws
	M9	M10	M11	M12	M13	M14	M15	M16
Invalidations	56.81	45.1	46.96	−16.58	0.18	0.13	0.01	0.24
	(30.94)	(27.04)	(28.9)	(12.61)	(0.21)	(0.23)	(0.22)	(0.26)
Lead (2-yr)	64.36	55.39	52.61	5.98	0.2	0.08	0.05	0.27
	(49.28)	(43.94)	(41.52)	(11.64)	(0.34)	(0.34)	(0.33)	(0.33)
Covariates	N	Y	Y	Y	N	Y	Y	Y
Year FE	N	N	Y	Y	N	N	Y	Y
State FE	N	N	N	Y	N	N	N	Y
N	996	975	975	975	996	975	975	975
R2	0.05	0.15	0.18	0.71	0.01	0.03	0.07	0.56
	Permissiveness				Definitions
	M17	M18	M19	M20	M21	M22	M23	M24
Invalidations	−0.19	−0.1	−0.07	−0.07	−0.18	−0.24	−0.28	−0.11
	(0.13)	(0.16)	(0.16)	(0.17)	(0.1)	(0.11)	(0.11)	(0.06)
Lead (2-yr)	−0.22	−0.16	−0.19	−0.15	−0.22	−0.28	−0.27	−0.18
	(0.2)	(0.21)	(0.22)	(0.22)	(0.13)	(0.13)	(0.13)	(0.13)
Covariates	N	Y	Y	Y	N	Y	Y	Y
Year FE	N	N	Y	Y	N	N	Y	Y
State FE	N	N	N	Y	N	N	N	Y
N	996	975	975	975	996	975	975	975
R2	0.01	0.05	0.06	0.45	0.02	0.08	0.1	0.73

Note: Standard errors clustered by state and reported in parentheses. Dependent variables and covariates as in Table 3.

Once we account for political and economic covariates, and state and year fixed effects, we see that the coefficients on invalidations tend to diminish in magnitude (M4, M8, M12, M16, M20, M24). In these specifications, judicial invalidations appear to have virtually no effect on any of the outcome metrics related to the nondelegation doctrine; no coefficient is statistically significant.⁴⁶ Further note that the models reported in Table 4 have reasonably precisely estimated zero coefficients, with point estimates near zero and standard errors small relative to standard deviations in the outcome variables (Table 2). The magnitude of the coefficients is also small relative to what we observe in the New Deal exercise. For instance, the percent of delegating laws increased by about five percentage points in the New Deal era, an increase that is far outside of the confidence interval implied by Table 4; much the same is true of coefficients related to other outcomes.

Notice also that these specifications include two-year lead indicators, which take a “1” in the two years preceding the invalidation of a state statute on nondelegation grounds. These leads act a type of “placebo” invalidation and should, under the identifying common trends assumption, have nonsignificant, small coefficients. They also allow us to investigate, for example, whether invalidations follow dramatic expansions in delegating behavior, as happen in 1935 at the federal level, which might raise concerns about the endogeneity of the doctrine to legislative practices. Similarly, emboldened by a weak or divided legislature, the invalidations might follow periods in which the legislature cannot pass significant pieces of legislation, again raising concerns about the endogeneity of the doctrine to legislative practices. However, as is evident from the leads in Table 4, they reassuringly return with small and nonsignificant coefficients, militating against the notion that doctrine is endogenous to legislative behavior (at least in this domain and over this time period).⁴⁷

The results from this exercise, therefore, largely corroborate those from the analysis above: the nondelegation doctrine does not appear to much matter for legislative drafting practices. A tightening of the nondelegation doctrine, as indicated by cumulative judicial invalidations over the series, does not influence drafting practices in the most obvious places we might examine. This is consistent with the earlier sets of results, which indicate, first, that states with and without a strong nondelegation regime do not differ in terms of political and economic characteristics, and, second, that drafting practices do not differ in states with and without a strong nondelegation regime.

4. Mechanisms

The nondelegation doctrine may not influence legislative practices for at least three reasons. First, the doctrine may be regarded as little more than an aspirational formalism: courts develop intimidating-sounding doctrinal formulations, which fold when confronted with the facts of a nondelegation case, as courts submit to political constraints inherent to judicial decision making (e.g., Hamilton's Federalist No. 78; Rosenberg 2008). Second, courts may channel their constitutional concerns about nondelegation into nonconstitutional doctrines, as is widely viewed to have happened at the federal level (e.g., Bressman 2000; Manning 2000). Finally, and most alarming from the perspective of judicial boosters, it is possible that both the constitutional and non-constitutional variants of the doctrine, even if important to judicial behavior, simply do not much matter to legislative behavior given the other factors weighed by legislators.

On the basis of the evidence presented thus far, we can rule out at least the strongest variant of the the first possibility. As depicted in Figure 3, states in fact invalidate statutes as unconstitutional delegations of legislative power. Over the relatively short 20 year series, state courts have invalidated over 20 statutes on this basis. Critically, most of these invalidations took place in states with strong forms of the nondelegation doctrine. This indicates that the doctrine, far from empty formalism or mere judicial posturing, reflects a constitutional commitment that courts enforce. This pattern is itself of interest, substantiating a connection between judicial doctrine and judicial action, yet it amplifies the question of why the nondelegation doctrine fails to systematically influence legislative behavior.⁴⁸

A second possible mechanism roots in the fact that courts express their concerns over delegation through both constitutional and non-constitutional doctrines. Rather than invalidating a statute as an unconstitutional delegation, courts instead often adopt narrowing constructions of the statute that address the same constitutional concerns.⁴⁹ Under this approach, what matters most is the constitutional commitment to legislative specificity underlying the nondelegation doctrine.⁵⁰ The doctrinal channel through which this commitment operates is of less importance—whether directly through constitutional doctrine, or indirectly through the methods courts use to interpret statute, the basic outcome is the same. This at least is one plausible interpretation of the results above, which in the main indicate that the constitutional nondelegation doctrine has a quite marginal effect on legislative drafting practices.

One implication of this doctrinal substitution argument is a higher incidence of controversies over statutory interpretation in legal systems with a weaker nondelegation doctrine.⁵¹ That is, consistent with the common view of the federal system, courts channel their constitutional nondelegation concerns into disputes over statutory construction; by comparison, where the constitutional concerns over nondelegation might be addressed directly, there is correspondingly less pressure on the statutory arguments. Thus, all else equal, under this substitution mechanism, we might expect to see more questioning of agencies’ statutory authority in legal systems with a weak nondelegation doctrine.

To probe this possibility, I gather all cases decided states’ highest courts that involve in central part a statutory question of whether the state agency had the power to issue the rule or regulation at issue.⁵² Contrary to the substitution thesis, disputes over agencies’ statutory authority to make regulations appear more, not less, pronounced in states with a strong nondelegation doctrine, suggesting that the rigor in the statutory doctrinal channel is a complement rather than a substitute for rigor in the constitutional doctrinal channel. Figure 4 plots the distributions of statutory cases in strong and weak nondelegation states, and there one sees clearly that mass is comparatively shifted to the right for the strong nondelegation states; strong nondelegation states tended to have more controversies involving the scope of statutory authority than weak nondelegatin states. This is, of course, only suggestive evidence, in a number of senses; for example, the difference between the mean number of cases in the two sets of states is not statistically significant.⁵³ Nevertheless, this evidence pulls our beliefs away from the view that the doctrinal substation mechanism explains why the nondelegation doctrine appears not to influence lawmaking practices.

Figure 4.

Statutory Challenges in the States: 1990–2010.

This leaves us with another, less sanguine possibility from the perspective those advocating the virtues of the nondelegation doctrine. It may be that the legislature views the occasional invalidation of a delegation as merely the cost of doing business in a modern democracy. Even in the relatively judicially active states, the courts invalidate a small percentage of all delegating statutes passed by the legislature. One live possibility therefore is that the frequency of invalidations is simply insufficient to affect legislative behavior. Moreover, as suggested by the experience of the doctrine at the federal level, the restraint of the judiciary is probably better understood as representing constraints inherent to courts’ role in a democratic polity than as some sort of correctable failing of the judiciary.

5. Discussion

For all the debate over the nondelegation doctrine, it is surprising to find that the doctrine appears not much to influence lawmaking behavior. In this sense, these results point to a foundational misconception in the debates over the nondelegation doctrine. The patterns of this article suggest it is unlikely that shifts in doctrine will much perturb the legislative incentives generated by the fundamental forces driving legislative behavior. These forces most plausibly a consist of a combination of the electoral imperatives to deal with problems confronting the polity subject to legislative institutional constraints, notably plenary time (Cox 2006), and expertise (e.g., Landis 1938); the need to avoid certain curses of dynamic inconsistency in preferences (e.g., Rogoff 1985; de Figueiredo 2002); incentives to build trust with the electorate (Stiglitz 2018); and most concerning to nondelegation theorists, perhaps the desire to avoid blame for divisive policy actions (e.g., Ely 1980; Fiorina 1982).

This pattern of findings therefore suggests that judicial doctrine exerts sharply limited control over legislative drafting practices, at least when legislatures face important countervailing incentives. This conclusion complements recent research on Congress-Court feedback loops. In a pioneering pair of articles, Gluck and Bressman (2013) and Bressman and Gluck (2014) survey congressional drafters and show that they often have little knowledge of doctrines of statutory interpretation, and that even when they do have knowledge of doctrines, they often chose to ignore them for political reasons. Congressional drafters, therefore, often color outside of the lines drawn by the courts, breaking a commonly presupposed feedback loop in which Congress listens to doctrinal drafting guidelines. In a similar spirit, this article indicates that the nondelegation doctrine does a poor job of shaping the behavior legislative drafters.

Future refinements to this analysis might examine additional measures of statutory specificity. Throughout this article, I have attempted to be as broad as reasonably possible in considering different possible measures for the simple reason that no standard measure of statutory specificity exists. In particular, I examine the length of the delegating statute per rulemaking authority, the prevalence of linguistic indicators of permissiveness, as well as structural aspects of the statute (i.e., the prevalence of definitions), and others. Across measures, the finding is against any legislative response to the nondelegation doctrine. Yet one might conjure other measures, and future efforts might consider other plausible candidates. A second avenue of refinement focuses on heterogeneity in the content of statutes. That is, it may be that the nondelegation doctrine binds in some policy areas and not others. Or perhaps it binds on “important” legislation but not on “routine” legislation. Or perhaps on “unpopular” but not “popular” legislative delegations. Or perhaps the nondelegation doctrine influences drafting practices, but expresses primarily in legislation subsequent to the delegations that condition or constrain those earlier actions. A final area of refinement focuses on heterogeneity in state judiciaries, with elected and appointed systems as a prime candidate. Such empirical exercises may then inform a more robust theory of legislative delegations and the possible role of the judiciary in shaping this aspect of legislative activity.

To at least a first approximation, however, is it remarkable that in the nondelegation doctrine, subject of so much debate over so many decades, does not fundamentally structure lawmaking behavior. As a corollary, this also suggests that, to the extent we wish to pursue the functional virtues that supposedly flow from the doctrine—such as political accountability—a revived nondelegation doctrine is not the most promising approach. More generally, it sounds a note of caution for doctrinal reforms that seek to accomplish fundamental behavioral changes in coordinate branches of government.

Supplementary Material

Supplementary material is available at Journal of Law, Economics, & Organization online.

I thank Matt Adler, Tom Clark, Mike Dorf, Sandy Gordon, Dan Ho, Michael Livermore, Jennifer Nou, Eduardo Penalver, Chad Pollard, Aziz Rana, Jeff Rachlinksi, Barak Richman, Chris Schroeder, Doug Spencer, Jeff Staton, Judge Stephen Williams and participants at workshops at the Conference for Empirical Legal Studies, Cornell Law School, Cornell University Government Department, Duke Law School, and Emory University for helpful comments or discussions. I am grateful to Doug Spencer for sharing data on state institutional features. I also thank Caitlin Lucey and Ryan Madden for excellent research assistance.

Footnotes

1. See Association of American Railroads v. DOT, 721 F.3d 666 (D.C. Cir. 2013), vacated by DOT v. Ass’n Amer. Railroads, 135 S. Ct. 1225 (2015).

2. As discussed below, the doctrine also gives animating force to a number of cognate doctrines as well. For recent examples of scholars advocating a stronger nondelegation doctrine, see, e.g., Lawson (2010), Hamburger (2014). For accounts more sympathetic to delegated authority, see, e.g., Mashaw (1985, 1997), Schuck (1998).

3. John Hart Ely, for his part, committed strongly to the political accountability rationale: “[t]hat legislators often find it convenient to escape accountability is precisely the reason for a non-delegation doctrine” (1980: 133).

4. The idea is that legislative specificity reduces the likelihood of the executive trampling individual liberty using broadly delegated authority. For an account of this dynamic, see, e.g., Sunstein (2000: 320).

5. In this way, the supposed consequences of the nondelegation doctrine can be decomposed into a quantitative question and a qualitative question. Quantitatively, do the number of delegating statutes decrease under the doctrine? Qualitatively, conditional on delegation, do those delegations provide more meaningful standards? This two-part decomposition resembles that of Davis (1969), who refers to the two objectives of the doctrine of “prevent[ing] delegation” and “assuring that delegated power will be guided by meaningful standards.” Though the most natural reading of the doctrine’s putative effects follows Davis and posits a decrease in quantity and an increase in specificity of standards, one might imagine other scenarios. For instance, the doctrine might bind on the qualitative feature and not the quantitative feature. Under this scenario, it is plausible that the doctrine would increase both the number of delegating laws as well as their specificity. I recognize such scenarios, but focus on the more conventional understanding when discussing doctrinal expectations. That said, this is ultimately an empirical question, and the design of this study will reveal such patterns if they exist.

6. Notice that even if the predicate is validated some functional arguments would still require work. For example, with respect to political accountability, there remains a question of whether voters respond to changes in legislative drafting practices, blaming legislators for policy decisions when they would not have in contexts of more broadly delegated authority. Some recent experimental work suggests, behaviorally, that delegating authority may be an effective means of shifting blame (e.g., Bartling and Fischbacher 2011; Hill 2015), notable given that theoretically voters might blame the legislature for underlying delegation, even if not for the blame-worth action of the delegate (e.g., Stephenson 2003; Vermeule 2007). However, the experimental evidence tends to derive from scenarios involving direct legislation versus delegation; we have little guidance on the question of whether delegation occurring merely under “more precise” legislative standards, the most plausible outcome from a revived nondelegation doctrine, would change voters/respondents views.

7. Administratively, for instance, even with today’s unproductive Congress, the number of laws produced by the body in a year exceeds the number of opinions produced by the Supreme Court by more than a factor of two. A large literature discusses the political constraints on excessive interventions by the court; see, e.g., Stiglitz (2018) and cites therein.

8. For a review of this literature, see de Figueiredo and Stiglitz (2017).

9. New State Ice Co. v. Liebmann, 285 U.S. 262, 311 (1932) (Brandeis, J., dissenting). For an examination of this notion of states as laboratories, see Gardner (1996).

10. US Const. Art. I, 8, cl. 3.

11. Note, though, as I discuss below the doctrine is widely viewed to continue to influence a range of other doctrines indirectly.

12. It is of interest that federal courts briefly flirted with this agency-centered spin on the doctrine, American Trucking Ass’ns, Inc. v. EPA, 175 F.3d 1027 (D.C. Cir. 1999), though the Supreme Court quickly squashed this possibility.

13. Not represented are Alaska and Hawaii, both of which have weak nondelegation regimes.

14. So categorized, the Greco and Rossi codings are almost identical. To reconcile any differences between Greco and Rossi codings, I adopted a rule whereby a state would only qualify as having a “strong” nondelegation doctrine if both scholars deemed it as such; this rule ends up affecting only one state, Illinois, which Greco codes as having a weak doctrine and Rossi as a strong doctrine. Requiring both coders to classify a state as having a strong nondelegation doctrine weeds out the marginal states and provides the best opportunity for us to observe an effect of the doctrine.

15. Here, I refer to the widely used Squire index of legislative professionalism. Squire (2007).

16. Laws laboriously searched and pieced together from various Lexis databases containing session laws from each state. I would prefer to have a series starting before 1990, but state session laws do not reliably appear in electronic form until that date.

17. I identify delegating laws by searching for laws that both mention an agency or administrator (“agency,” “bureau,” “board,” “commission,” “department,” “director,” “secretary,” “administrator,” “division”) and also contain language delegating rule making authority. I detect the latter using the following regular expression: “(authorized—empowered—shall—may).0, 50(make—made—adopt—prescribe—promulgate).0, 50(standards—rule—regulation)” and similar variants described in detail in the Online Appendix. The idea is to capture a three-part relation involving (a) authority to (b) make (c) rules. Note that prior to conducting these searches, I remove all delegation-related boilerplate language from state session laws, as described in detail in the Online Appendix. I also account for negations of authority, as in “may not”.

18. The use of omnibus legislation is itself of interest as an outcome (see,.e.g., Garrett 2002; Eskridge 2012), though I do not focus on this in the present application. Note that the Alabama legislature drafts the third least number of laws, yet also drafts the fifth most verbose laws on average.

19. As suggested by this brief review, these total legislative words metric tends to correspond with the size of the state’s economy—states with larger, more complicated economies tend to produce more legislation. The correlation between the total number of delegating words in a state-year and the (log) state GDP is 0.52.

20. This decomposition of the supposed effects of the doctrine approximately follows Davis (1969).

21. See note 17 for an explanation of how I detect such delegations; see also the Online Appendix.

22. I say approximately because I normalize word counts by the number of grants of rulemaking authority; doing so is a response to the fact of heterogeneous legislative practices with respect to omnibus drafting. The spirit of the measure, however, is as in Huber and Shipan (2002).

23. Elsewhere, I engage in a similar approach to coding statutes (Stiglitz 2014 b), with theoretical implications examined in Stiglitz (2017); see also Stiglitz (2014a), which measures qualitative features of administrative rules through text contained in the rule preambles, and Livermore et al. (forthcoming) for important advances.

24. Am. Trucking Ass’ns, Inc. v. EPA, 175 F.3d 1027, 1057, 62 (D.C. Cir. 1999) (Tatel, J., dissenting): “reasonabl,” “fair,” “public,” “may.” Notice that I distinguish between “may” and “may not.” I further include precatory words identified in Mashaw (1997, 135): “feasible,” “practicable,” “appropriate.”

25. I detect definitions using the following regular expression: (“—word— term).{0, 100}mean. Thus, I attempt to locate instances in which the legislature first identifies a term, then shortly afterwards discusses its “meaning.”

26. Notice again that we have denominated by 1000 words of statutory language for ease of interpretation. Note that I examine the residuals in the preferred specifications and do not find skewness that would recommend log or other transformations.

27. Notice , again, that we have denominated by 1000 words of statutory language for ease of interpretation.

28. Notice, again, that we have denominated by 1000 words of statutory language for ease of interpretation.

29. Notice that in the analysis below the unit of observation is the state-year. The summary statistics follow this approach; for example, the first statistic reveals that in the average state-year, 17% of laws delegate rule-making authority to an agency.

30. Of course, legislation from this era prompted the two extant delegation-based invalidations at the federal level. Both Panama Ref. Co. v. Ryan, 293 U.S. 388 (1935) and Schechter Poultry Corp. v. United States, 295 U.S. 495 (1935) invalidated provisions of the National Industrial Recovery Act of 1933.

31. The data for this exercise derive from various volumes of the US Statutes at Large. I collect pdfs of the volumes, extract the text from the pdfs, algorithmically break the text into public laws, and then sweep the collection of public laws for obvious errors. These public laws text serve as a the data for a given year, though following the definitions above I often combine the public laws by year.

32. At the year level, the average proportion in the pre-New Deal era year was 0.057, and afterwards 0.111, a difference that is significant at any conventional level (p ¡ 0.01). I provide these statistics suggestively, and note that aggregating at the year level presents power issues.

33. At the year-level, the difference in means is significant at any level (on average, 25.8 laws before the New Deal; 50.2 after, p¡0.01).

34. At the year level, the average in the pre-New Deal era is 73.75, and 139 in the post-New Deal era; this difference is significant at the conventional level (p¡0.05).

35. This difference is significant at the 10% level (p = 0.08).

36. Here, the p-value on the difference is 0.11.

37. At the year level, the average number of definitions increases from 0.16 in the pre-New Deal era to 0.52 in the post-New Deal era; this difference is significant at conventional levels (p = 0.01).

38. For example, state laws and practices with respect to the single subject rule potentially bear on drafting practices. See, e.g., Gilbert (2005).

39. The most natural reading of this problem roughly follows the federal experience: the legislature delegates in response to some pressing public crisis; the courts then relax the nondelegation doctrine in fear various forms of political reprisal. Examining variation across states, this dynamic would tend to produce an “effect” of the nondelegation doctrine, such that it appears the doctrine is constraining drafting practices, when in fact legislative imperatives always fundamentally motivated drafting practices.

40. The Online Appendix contains tables that report coefficients for the covariates; I omit them from the main article to conserve space. The political and economic covariates include: log population, log state GDP, log per capita personal income, the unemployment rate, the Squire (2007) index of legislative professionalism, the year-over-year percentage change in income, the year-over-year change in unemployment, the year-over-year percentage change in state GDP, and an indicator for divided government (which requires dropping Nebraska and its non-partisan legislature from the analysis; however, the results with Nebraska and without the divided government indicator are similar). The economic covariates account for the possibility that the nondelegation regime may be correlated in subtle ways with economic conditions. For example, poorer states may have stronger nondelegation doctrines, and also less delegation, but the lower levels of delegation owe to economic conditions rather than the nondelegation doctrine. And much the same for the political covariates. The exercises from Table 1 suggest that this type of concern is not acute, but it is not costly to include specifications with covariates and I do so.

41. Notice that we cannot introduce state fixed effects in this framework due to the fact that the operative metric of states’ nondelegation doctrine remains fixed throughout the period. I employ another approach in the following subsection that permits a closer inspection of changes in drafting practices within states.

42. Moreover, as suggested above, the most plausible concerns regarding endogeneity run in favor of finding a relationship. See note 37.

43. Notice that for the preferred specification below we include state fixed effects, so we’re identifying off changes within states, and the initial value for this metric is irrelevant.

44. In the Online Appendix, I also develop a coding scheme that runs in the opposite direction, i.e., with the doctrine loosening after courts uphold state laws against some nondelegation challenges. The challenge to including decisions that uphold state laws is that litigants often include a nondelegation challenge as an implausible Hail Mary alternative theory, reducing the informativeness of judicial decisions to uphold state statutes. The Online Appendix develops a technique of separating the informative from the uninformative decisions, but that exercise remains tentative so I place it there. Including doctrinal loosening does not qualitatively change the results reported in the main tables.

45. I have also run these regressions excluding any state with more than two invalidations over the series; the results from this exercise do not differ qualitatively from those reported below, and I make them available on request.

46. The coefficient on definitions is closest to significance but misses the conventional threshold.

47. Note that I also experiment with a leads of three and four years, to the same outcome.

48. In this sense, the relationship between legal doctrine and legal action should be viewed to militate against at least the strongest variants of the position that “legal” or “doctrinal” factors play no role in judicial behavior. For a recent entry in the long-standing debate in political science over whether law “matters,” generally concluding in the affirmative, see Bailey and Maltzman (2008).

49. For a recent and vivid example of such a case—though not in the nondelegation context—consider Bond v. United States, 134 S. Ct. 2077 (U.S. 2014).

50. This position is similar to that of Farina (2010), who observes that the doctrine serves as a vehicle for indirectly expressing underlying concerns about the administrative state.

51. In principle, the two modes of review might be complements rather than substitutes, and in fact that is what the data below suggest. However, only the substitutes argument—that constitutional concerns find their home in methods of statutory interpretation rather than in doctrines explicitly grounded in the constitution—would explain why a judicial commitment to the constitutional norm of nondelegation does not necessarily result in a relationship between a system’s nondelegation doctrine and legislative drafting practices; in the substitution argument, that relationship does not necessarily exist because the concern is finding expression in other, statutory channels.

52. To do so, I rely on Westlaw’s hand coding system, in two ways. First, I rely on Westlaw to identify cases that touch on this question. Second, I rely on Westlaw to only include a keynote when this statutory issue was an important part of resolving the dispute at hand. I collect all cases under the Keynote “15AK385,” which identifies cases that involve the agencies “power to make” rules and regulations. Note also that I read through these cases to see whether they also discuss the nondelegation doctrine. Of the 192 statutory cases between 1990 and 2010, I found two that discussed the doctrine in any detail: McLean v. Hyland Enterprises, Inc., 34 P.3d 1262, 2001 W.Y. 111 (Wyo. 2001); Blank v. Department of Corrections, 611 N.W.2d 530, 462 Mich. 103 (2000).

53. Based on the Westlaw coding, the average strong nondelegation state had 4.94 statutory cases between 1990 and 2010; the average weak nondelegation state had 3.16 statutory cases over this period. The p-value on this difference in means is 0.17.