Delegation or Unilateral Action?

Transcription

1 Delegation or Unilateral Action? KENNETH LOWANDE March 20th, 2015 Studies of unilateral action typically conceive of presidential directives as movement of an existing status quo policy. Yet, bureaucratic cooperation is often necessary if presidential action is to be truly unilateral. In this vein, I argue that such directives can be more fruitfully studied as instances of delegation. I develop a theory of delegation within the executive branch, modeling the conditions under which the president is likely to delegate and provide discretion to administrative subordinates outside the Executive Office. I test this theory by analyzing executive orders for content containing authority delegated to subordinates. I show that, contrary to the dominant paradigm of presidential studies, the politics of direction action is mitigated by the necessity for bureaucratic cooperation. Ph.D. Candidate, University of Virginia, lowande@virginia.edu Paper prepared for presentation at the 2015 Southern California Law and Social Science Forum. I thank Jeff Jenkins and Craig Volden for helpful comments and suggestions, Matt Dickinson, Will Howell, and Rob McGrath for sharing their data, as well as Thomas Gray and Nick Jacobs for volunteering their time in service of coding validation.

2 On April 8th, 1952, Harry Truman issued an executive order that seized the property of dozens of American firms (among them, the Youngstown Sheet & Tube Company) whose output was indispensable" to the Korean War effort. Among notable cases cited by scholars of presidential power, E.O is second in importance only to 9066, issued by FDR, which authorized Japanese interment during World War II. If one looked to dominant theories of unilateral action for admittedly stylized explanations of this particular case, we would have to conclude that Truman singlehandedly altered the status quo. 1 Yet, a few basic features of the order and the case itself render this explanation unsatisfying. First, E.O did not order agents from within the Executive Office of the President to begin seizing steel mills it delegated authority to the Secretary of Commerce, Charles Sawyer. Second, the order did not prescribe in legal minutiae how the Secretary was to seize and manage the mills, it gave him substantial discretion. The Secretary was permitted to act through or with the aid of such public or private instrumentalities or persons as he may designate, to determine and prescribe terms and conditions of employment under which the plants, facilities, and other properties possession of which is taken pursuant to this order shall be operated, to issue regulations he deemed necessary, and even to return possession of the mills to their owners if he judged it appropriate. This implies E.O should be deemed an instance of delegation to bureaucratic agents, as opposed to a change in policy with the stroke of a pen (Mayer, 1999). But these facts are not sufficient to establish that presidential orders ought to be looked at as instances of delegation. For instance, if it were the case that Truman pre-negotiated the actions of the Secretary, and that perfect compliance was virtually guaranteed by top-down control, the principal-agent relationship may be insignificant or uninteresting. However, that is not the case in either the example provided or the larger universe of presidential actions. According to archival material and interviews after the fact, Sawyer was a reluctant participant, who cooperated on the condition that he would be provided a free hand in carrying out the order (Marcus, 1977). According to Neustadt (1960), even Truman s unambiguous intent did not compel Sawyer to act in a predictably compliant way. He initially refused to put union wage demands in place, and only did so after they were invalidated by the Supreme Court (21-22). These historical accounts suggest that without discretion, the Secretary would have refused to comply and that even then, full compliance did not occur. More broadly, in nearly every instance of presidents acting alone, the president must rely on agents within the bureaucracy who have some degree of institutional independence, such that studies (most notably, Neustadt and Nathan [1985]) frequently cite the importance of executive management and agency 1 Though it would later be restored following the Supreme Court s ruling in Youngstown Sheet & Tube v. Sawyer (1952) 1

3 compliance in discussions of presidential success. In this paper, I analyze presidential directives as delegation. I present a model of delegation within the executive branch, designed to approximate the presidential policymaking and bureaucratic implementation. Placed in the context of the existing paradigm, this theoretical framework suggests that bureaucratic agency is a mechanism by which members of Congress secure better policy outcomes when the President acts alone. This is because presidential directives often delegate to subordinates, who may be influenced by the threat of sanctioning from Congress and relevant committees. I support this perspective with an analysis of important executive orders issued between First, I demonstrate that a substantial majority of presidential actions involve delegation to bureaucratic agents. I find, in-keeping with expectations, that Presidents provide less discretion to bureaucratic agents as the policy disagreement between the President and relevant congressional committees increases. Moreover, I show that agents receive more discretion when the President is faced with a large congressional majority capable of imposing substantial costs on those who implement the president s policies. These findings suggest that, beyond the influence of Congress and the Judiciary, presidential policymaking is contingent on bureaucratic agency. I. Unilateral Action, Agency Problems, and the Executive Branch In his seminal book, Howell (2003) writes that modern presidents often exert power by setting public policy on their own and preventing Congress and the courts and anyone else for that matter from doing much about it (14). This perspective, elaborated in Moe & Howell (1999), has influenced a generation of quantitative research on the presidency, which attempts to uncover the precise political circumstances that enable the President to act alone. Empirically, much of this research has focused on the study of presidential directives: executive orders (Mayer, 1999; Fine and Warber, 2012; Chiou and Rothenberg, 2014), proclamations (Rottinghaus and Maier, 2007; Rottinghaus and Lim, 2009), signing statements (Kelley and Marshall, 2010; Ostrander and Sievert, 2012), and memoranda (Cooper, 2002; Lowande, 2014). More recently, the conceptual focus on what presidents can accomplish alone has informed investigations of the president s role in the distribution of federal spending (Mccarty, 2000; Berry et al., 2010; Hudak, 2014; Reeves and Kriner, 2015). This work has furthered our understanding of the president s influence over policymaking. However, there are several reasons to believe that as a theoretical base, it remains inadequate. First, as the introduction suggests, the stylized depiction of a status quo movement fails to capture the 2

4 essence of presidential directives themselves. Second, as Mayer (2009) points out, taking the unilateral action paradigm to its logical end results in a conclusion which may be normatively unappealing. If the ambiguities of presidential authority mean that the boundaries of presidential power are determined by precedent [...] if presidents have incentives to act first [...] if Congress and the courts face hurdles [...] presidential power becomes uncontrollable and sinister. (443) Even if value judgements are resisted, an important question remains: in an era of congressional polarization, what is to prevent the president from becoming all powerful? If we take into account that acting independent of Congress requires cooperation from bureaucratic agents, then the gradual implications unilateral action become less sinister" and more contingent. Presidential history is replete with cases of agency heads reinterpreting presidential directives, or refusing to comply outright. 2 This is not to say that unilateral action is a complete misnomer, or that the existing paradigm is wholly incorrect. However, it is important to build on a central point: that when Congress passes a law or the President issues and executive order, both actors are relying on bureaucrats for policy change. Absent that recognition, theories risk aggrandizing the capabilities of the president and downplaying the agency of the bureaucrat. The appropriate research question, then, is: how does bureaucratic agency influence policy outcomes when the president attempts to go it alone? Reframing the issuance of presidential orders as instances of delegation sheds light on the content, rather than the frequency of unilateral action and, by implication, provides more information about the degree to which action is unilateral, and the potential for deviation from the president s preferred outcomes. Importantly, benchmark studies of legislative delegation of policymaking authority to bureaucratic agents provide a conceptual way forward. They have highlighted the critical trade-offs for members of Congress looking to delegate power: between policy outcomes and discretion (Epstein and O Halloran, 1994, 1996), between outcomes and executive aggrandizement (Volden, 2002a), and between informed and ideological personnel (Gailmard and Patty, 2007, 2013). Additionally, some theoretical work on delegation and information sharing conceives of the key principal as either a legislature or an executive (Patty, 2009; Ting, 2009). This suggests the basic framework of this study may shed light on a principal beyond the median voter in Congress. Given this, presidential directives offer a rare opportunity for verification of proposed theory. Systematic empirical support for delegation models rests on a few essential studies (Epstein and 2 For instance, Secretary of the Treasury Louis McLane under Andrew Jackson (during the Bank of the US controversy). A more recent example is Melissa Hathaway, former Cyber Security Czar under Barack Obama, who inside sources said resigned after spinning her wheels under the administration after serving under in the Bush White House (Gorman 2009). Many other cases are outlined by Neustadt (1960). 3

5 O Halloran, 1999; Volden, 2002b; Huber and Shipan, 2002). Since then, as Moe (2012) notes, formal work on delegation has largely outpaced empirical testing. 3 I.1 The Presidential Branch Importantly, the theoretical framework in this paper builds on the basic point that agency problems are pervasive in the executive branch. That is, when presidents want a policy changed, they must rely on subordinates with agency. Their wishes and directives are not self-executing. Perfect monitoring is implausible. The multiplicity of policy areas and the limitations of a single office generate an asymmetry of time, energy, and knowledge. This fundamental feature of the president s position has been the basis for the study of presidential management of the bureaucracy and influence over policy outcomes. Among scholarship on politicization, for example, the notion that presidential and bureaucratic preferences often diverge is at the core of most explanations for the politics of appointments (Lewis, 2008; Hollibaugh et al., 2014). Moreover, bureau responsiveness, even after politicization, is not guaranteed (Dickinson and Rudalevige, 2004). Agency problems are a well-known, systemic part of presidential administrations. Consider, for instance, the Nixon Administration s adversarial relationship with the executive bureaucracy (Aberbach and Rockman, 1976). Conceiving of presidential directives as acts of delegation, then, incorporates much of what is already assumed (explicitly or implicitly) in scholarship on the president s relationship with the executive branch. However, this literature also highlights an important challenge. That is, under the broadest definition, nearly every action the president takes can be considered an act of delegation. A delegation-all-the-way down perspective risks returning to an understanding of the President as overwhelmed and ultimately incapable of seriously influencing policy (e.g. Lowi 1985). Thus, it is essential to acknowledge that when delegating, the President faces a range of potential agents, who vary in terms of ideological disposition (Clinton and Lewis, 2008; Chen and Johnson, 2014), institutional independence (Selin, 2015), and ultimately, monitoring cost. While precise specification of the institutional cost to monitoring the President s agents may by worthwhile, I show that analytical and empirical leverage can be gained through a necessary simplification. That is, I conceive of the president s decision to delegate as a dichotmous choice: delegate to external agents those in government corporations, independent agencies, and cabinet departments; or, delegate to actors within what has been called the presidential branch the White House (WHO) and Executive Office 3 For key examples of theoretical development in this area, see: Huber & McCarty (2004), Callander (2008), Wiseman (2009), Fox & Jordan (2011), Callander & Krehbiel (2014). 4

6 (EOP). The growth of the president s immediate institutional apparatus has been well documented and studied (Dickinson, 1996; Rudalevige, 2002; Dickinson and Lebo, 2007). President s have the capacity to develop policy in a wide variety of issue areas with the resources at their disposal within the WHO and EOP. Thus, this simplification nonetheless preserves an essential point, made first by Moe s (1985, 1993) perspective on presidential politicization and centralization: that within the WHO and EOP, presidents face fewer collective action problems (compared with Congress) and lower risks of policy drift (compared with the rest of the executive bureaucracy). Building on these premises, I present a model of presidential delegation and leverage a data set of discretion in executive orders to verify its central intuition. II. A Theory of Presidential Delegation Given the motivating conceptual shift, a theory of delegation from the Oval Office should answer two questions. First, when (and why) will the president delegate to agents outside the relative arm s reach of the White House and Executive Office? Second, and relatedly, what conditions the level of discretion provided to these agents? I present a straightforward spatial model of delegation within the executive branch with three players: the President (P), a congressional committee (C), and an external agent (A). The congressional committee may be thought of as the committee with oversight jurisdiction over the corresponding agency. 4 In the model, the President attempts to implement a policy outcome ( ˆx R) nearest to their preferred (x P ), while minimizing the resource cost (τ) associated with developing policy in the White House and EOP. To avoid that cost, the President may delegate via directive to an external agent (e {0,1}) and provide them with a level of discretion (D [0,R + ]). 5 As the elected heads of the federal bureaucracy, presidents have acquired institutional resources designed to centralize design-making in the Executive Office of the President (EOP). Nonetheless, complete centralization is impossible. Presidents must set priorities, engaging in a key trade-off when selecting bureaucratic agents. Thus, a core assumption of the argument is that presidents have limited resources to formulate policy within their immediate domain. 6 In the context of the model, this means that 4 This is particularly important, given the broad understanding of sanctioning behavior I put forth. Drafting punitive legislation and holding hearings occur at the committee level, such that assuming C to be the congressional floor median may inappropriately limit the threshold of political support that determines whether Congress engages in sanctioning activity. 5 I define discretion (like Epstein & O Halloran 1999) as delegated authority, together with the severity of procedural and oversight constraints placed on that authority. 6 It may be useful to think of the analogous legislative environment: A make-or-buy framework views Congressional committees as appendages of floor majorities (the make option) and bureaucratic agencies as contractors (the buy" option; Epstein and O Halloran 1999). In the Presidential context, the EOP would be the equivalent in-house producer, whereas agents in Cabinet departments, independent agencies, and commissions would play the role of contractors. 5

7 President s must pay some exogenous cost if they forgo external delegation. The congressional committee looks to obtain a policy closest to its bliss point (x C ) through the lever of agent sanctioning (s {0, 1}). Here, they engage in ex post actions in response to the President. That is, Congress could enact a law that directly punishes the agent, subject to well-known coalition thresholds for lawmaking (filibuster and veto-override pivots). These coalition thresholds are what provide existing theories of unilateral action with their implications. In absence of large Congressional majorities, it is difficult for Congress to marshall the political resources to enact a law intended to supersede a presidential directive. However, in absence of a super-majority, there are still sanctions Congress can impose on the agents. Limitations riders can be inserted into appropriations bills (MacDonald 2010). 7 Bureaucrats can be called to testify on Capitol Hill. They can be held in contempt of Congress. This suggests that Congress can impose targeted sanctions on executive branch agents that do not require the political capital demanded by lawmaking. This secondary action imposes costs directly on the agents implementing the president s program. Another important feature of the model is that the agent, if chosen to make policy, can opt out by refusing to comply (v {0,1}) with the president s directive. If it opts in, it selects a policy conditional on the level of discretion supplied by the President. While outright non-compliance is not a typical feature of delegation models, there is reason to believe that this activity influences policy-making. 8 Deviation and resignation is a regular part of presidential administrations. For example, in 1933, Treasury Under Secretary Dean Acheson resigned rather than implement FDR s Executive Order 6102, which required all newly mined gold bullion to be delivered to the federal government. To pick a more contemporary example, in 2004, then-acting Attorney General James Comey refused to reauthorize a wiretapping program under George W. Bush. More generally, since the Clinton Administration, presidents have used directives to set (or reset) rule making deadlines which are frequently broken or ignored entirely. Moreover, this feature may render the model more generalizable beyond the American context. Bureaucratic noncompliance is a pervasive issue which underlies many comparative studies on corruption. Thus, assuming that bureaucrats have the capacity to disobey provides a more accurate, general depiction of executive politics. Ultimately, the model suggests that in equilibrium, observations of bureaucratic non-compliance should be low (or nonexistent). Thus, as 7 Because members of Congress imbed this instrument within must pass legislation, it arguably requires a lower political threshold to enact. 8 Ting (2015) explores legislative strategies for achieving agency compliance in terms of policy selection a substantively different form of non-compliance than agencies opting out. 6

8 with other political phenomenon with equilibrium effects, like the presidential veto (?), observed cases may understate their overall impact on the broader process. The utility of the President is given by U P = x P ˆx e (1 e)τ (1) so that they attempt to minimize the disutility associated with distant policy and the cost of developing policy in-house. The committee s utility is governed solely by the policy outcome, such that U C = x C ˆx (2) Finally, the Agent s utility is governed by policy outcomes and (if incurred) the cost associated with a congressional sanction (c [0,R + ]). U A = x A ˆx sc (3) I present the complete game sequence below, but before moving on, it is important to highlight ways in which the model differs from relevant landmark work. First, in contrast to Howell (2003), presidential directives are seen as the beginning of process by which the president resets the status quo. Notably absent are critical pivots in Congress, the Judiciary, or the lawmaking process, more generally. This reflects a key distinction: whereas the unilateral politics model outlines conditions that lead to presidential action, the model below attempts to explain how the agency of bureaucrats influences the final outcomes after the decision to act alone has been made. Second, in contrast to landmark models of delegation (Bawn, 1995; Epstein and O Halloran, 1999; Volden, 2002a; Gailmard and Patty, 2007), the principal s key trade-off is not motivated by gains in expertise. Instead, delegation is attractive because it allows the President to dedicate limited resources to other policy initiatives. I take this to be a more accurate characterization of the principal-agent scenario encountered by the President particularly when compared to the substantive justifications for the information asymmetry between Congress and the bureaucracy. Members of Congress are said to have limited time, staff, and knowledge that renders them less able to match the expertise of agents. But the President is surrounded by a full-time staff of experts, many of whom, are employed at will and possess specialized knowledge in substantive areas. This is not to say that all expertise needed to produce policy could be contained within the 7

9 WHO or EOP quite explicitly, its limitations will be captured by τ. But I argue those limitations should be characterized as a shortage of resources, not a shortage of potential expertise. II.1 Sequence of Play 1. Nature selects the status quo policy, q. 2. The President chooses whether to delegate or develop policy within the White House and EOP (e {0,1}). 3. (a) If e = 0, then the EOP develops the President s preferred policy ( ˆx = x p ) and the President pays a resource cost, τ. (b) If e = 1, then the President selects a level of discretion, D. i. The Committee chooses a sanction rule, σ : x s. ii. The Agent chooses whether or not to comply (v {0,1}). iii. A. If the Agent refuses to comply, then the policy reverts to the status quo ( ˆx = q). B. If the Agent complies, then the Agent selects a policy, x [x P D,x P + D]. (c) The Committee, according to its prior rule (σ) chooses whether to sanction the Agent (s). (d) Play ends and payoffs are distributed. A few simple examples underscore the applicability of this setup to the political process in question. Presidents, in pursuit of their goals, observe some existing policy that they would desire to move. Faced with an initial choice of whether to use resources within their immediate domain, or delegate to external agents. In 1976, for instance, Gerald Ford chose to delegate policymaking functions to the Federal Energy Office within the Executive Office, rather than vest those functions in the newly created Federal Energy Administration (predecessor to the Department of Energy). 9 If a President delegates, they are then faced with the task of formulating the agent s mandate to make policy. Their directive can be limited, or it can provide the agent policy latitude. The President could, for example, require that the agent consult with other agencies or departments, as Harry Truman did when he redelegated wartime employment functions to the Department of Labor in Relatedly, the President could simply specify the new policy in great detail, to limit the range of policies the agent could implement. 11 Next, Congress, observing the President s directive, makes its preferences known to the agent. This kind of Congressional posturing has become particularly salient recently, as members of Congress reacted to Barack Obama s series of immigration-related directives 9 E.O Performance by the Federal Energy Office of Energy Functions of the Federal Energy Administration (July 30, 1976) 10 E.O Transfer of Certain Agencies and Functions to the Department of Labor, (September 19, 1945) 11 Indicators of constraints on discretion are discussed in greater detail in the measurement section of this paper. 8

10 in November Senator Jeff Sessions (R-AL), for example, reminded the applicable departments that Congress has the power and every right to deny funding for unworthy activities. 12 This sequence appears to reflect another recent presidential initiative, Barack Obama s proposed ban on armor-piercing ammunition. In response, Rep. Jim Sensenbrenner (R-WI) introduced legislation to abolish the agency responsible for the president s initiative, the Bureau of Alcohol, Tobacco, and Firearms (ATF). 13 Though the President pays political costs for overturned or impeded policies, bureaucrats endure punishments which very in severity. Thus, appropriately, the Committee has the ability to sanction them directly, which influences the agent s actions. The final outcome is a function of the agent s decision to opt-in to the President s directive, and the policy the agent implements. II.2 Solutions Given the number of possible policy orientations and status quo, the inclusion of multiple principals increases the number of scenarios exponentially, compared with other benchmark models of delegation. Since the basic intuition of the model can be gleaned without specifying every possible scenario, I focus my analysis on the situation depicted in Figure 2. Without loss of generality, I assume the ideal points of the President and the Committee are 0 and 1 respectively, and that 0 < x A < 1. Though this is a useful simplification, it also approximates a typical scenario in policymaking among separated powers. Agency missions, appointees, and policy tasks are often the explicit result of inter-branch bargaining. Consequently, I argue it is reasonable to assume that their ideal outcome lies somewhere between their elected principals. Figure 1 x P 0 x A x C 1 Working backwards, consider the committee s choice of sanction. Recall, this is a deterministic function of the policy selected by x however, there are an infinite number of arbitrary decision rules (σ) that the committee could specify, some of which, will be more useful than others. For simplicity, consider two possible sanction rules: Request (σ 1, s = 1 iff x C x > x C q ) and Demand (σ 2, s = 0 iff x = x C ). The 12 November 20, Quoted in Shabad (2014). 13 Marcos, Cristina Republican proposes abolishing the ATF amid bullet ban controversy, The Hill March 5th. The ATF has since withdrawn the proposal preserving the status quo. 9

11 intuition behind each is straightforward. The committee demands when it sanctions the agent unless it secures its ideal point, and it requests when it only sanctions if the new policy makes it worse off. Given this set of sanction rules, we can specify the agent s optimal policy choice (x ) for every possible q. For Demanding, the optimal policy choice can be expressed as: D if D < x A xσ 2 = x A if x A < D < x C if x A x C c and x C < D (4) x C if x A x C < c and x C < D Here, demanding only induces a behavior change when the President has provided a sufficient level of discretion and the sanction cost is sufficiently high. When the committee adopts a sanction rule of Request (σ 1 ), the number of optimal strategies expands considerably, given that x becomes a function of the status quo, the sanction rule, its associated cost, and how much discretion it has been provided. The agent s optimal policy selection for any given status quo, level of discretion, and cost of sanction conditional on σ 1 can be written as: D if D < x A q if D x C q x C, x A D < c and x A < D < x C, x C < q x A if D > x A and x A > q if x A < D < q and x A < q < x C xσ 1 = if x A q c and x A < q < x C, D > q if x A > q x C and x C < q, D > q if x A < q x C, x A (q x C ) c and x C < q, D > q if D < x C q x C and x A < D < x C, x C < q if D x C q x C, x A D > c and x A < D < x C, x C < q (5) 2x C q q if x A < q x C, x A (q x C ) < c and x C < q, D > q if x A q < c and x A < q < x C, D > q Given this characterization of x, we can illustrate the effect of non-compliance (v = 0) on the universe of possible policy selections. Because Congress can only apply negative pressure (sanctioning imposes a cost), 10

12 the movement of policy becomes the chief determinant of whether the agent will opt-in. This rule is presented in Lemma 1. According to this condition, for the agent to comply with the president s directive, the new policy must at a minimum make the agent indifferent between x and the status quo. Put simply, the President cannot force the agent to implement an x that makes the agent worse off in terms of policy. In Figure 2 and 3, which depict x given q, this means that the region between x P and x A represents a kind of bureaucratic gridlock interval, wherein delegation always results in the status quo (Brady & Volden 2006, Krehbiel 1998). 14 Lemma 1. Given x P < x A < x C, v = 1 only if x A q x A x q. Proof: By definition, s = 0 when v = 0, so, given U A, v = 1 when x A x s c x A q. Since c R +, s c 0, and the sanction becomes irrelevant for satisfying the inequality. Next, consider the task of satisfying this condition in the context of the level of discretion provided by the President. Since D defines the set of policies the agent may select, it figures prominently into whether the agent will comply with the directive. Lemma 2 presents this relationship. Importantly, for the agent to comply the President must provide a level of discretion greater than or equal to the policy that would render the agent indifferent between it and the status quo: Lemma 2. Given x P < x A < x C, v = 1 only if D x A q x A q. Proof: Implicit, given Lemma 1 and x [ D,D]. Lemma 1 and 2 specify minimum conditions for the agent to opt-in, however, this choice is further constrained by the committee s ability to sanction. Even when both conditions are satisfied, the committee can prevent the agent from opting in if the sanction is costly enough. For example, when q < x C (so that the committee prefers to maintain the status quo) and c > x A q, no level of discretion can induce the agent to comply with the president s request. When this cost is less than the potential policy gains on the part of the agent, the President must provide a level of discretion which compensates for enduring a sanction (x A D c). Figure 2 illustrates this basic point. Larger values of c necessarily expand the gridlock region, and require the President to provide more discretion than would otherwise be necessary without the threat of sanction. 14 Note that, though Lemma 1 and 2 are qualified in terms of the spatial orientation x P < x A < x C, they hold for any configuration of the players. Since sanctioning only applies negative pressure after a policy has been selected, the President and the committee can never induce the agency to implement a policy that makes it worse off. 11

13 Figure 2: Discretion Region under Variable Sanction Cost (τ = 0) x* xp xa xc c =.3 c > x A q c = 0 x P x A x C 2x C q Figure 3: Discretion Region under Variable Sanction Cost (τ =.5) x* xp xa xc c =.3 c = 1 x P x A x C 2x C q 12

14 Turning to the committee s choice of sanction rule, recall that, intuitively, only conditional strategies have the potential to induce a positive behavior change (from the committee s perspective), so it is left to specify the conditions in which Request and Demand (and all those in between) are optimal. Given v and x, we can express the optimal sanction rule in the following way: σ = σ 1 σ D s = 1 iff x D if q < x C if x C < q, 2x C q D x C (6) σ 2 if x C < q, x C < D When the status quo pits the committee and the President s ideal policies directly at odds, it pronounces that it will sanction if the policy is shifted further away from their ideal point. For distant status quos, in which all players benefit, their optimal strategy is to demand the closest available policy, conditional on the level of discretion provided by the President. Next, before presenting the President s optimal choice of D, it is useful to introduce some additional notation. Let c denote the sanction cost which satisfies the condition x A (q x C ) < c (7) and conversely, let c denote the cost that does not. That is, the sanction cost is high when it is greater than the policy gain between the agent s ideal policy and the point which would allow them to avoid sanctioning altogether. In other words, this is the value of c which induces the agent to condition its choice on the committee s preferred outcome. With this in hand, we can characterize the equilibrium selection of D: Proposition 1. Given x P < x A < x C, the President s optimal choice of discretion can be written as D = 0 if 2x C q < x A 2x A q + c if x A < q < 2x C and c = c (8) 2x C q if x A < q < 2x C and c = c Proof: In turn, P provides no discretion when q < x A, given Lemma 1 and 2, which imply that for these status quos, A will either opt out, or opt in and select xp. P provides discretion commensurate with A s policy gains and c, given c. When c, P must provide discretion to satisfy C s demand. P provides no discretion when the status quo is sufficiently distant from all, since all benefit from the move to x P. 13

15 Though the implications of the proposition will be discussed in following section, a few points are worth noting. For instance, holding fixed the policy positions of the actors, discretion is increasing in c, for x A < q < 2x C. Intuitively, the more trouble the committee is capable of giving the agency, the more discretion (and by extension, policy concessions) the president will have to permit. Another important point is that if the cost reaches the threshold of c, discretion contracts and expands with the ideological distance between the President and the Committee. For a subset of q, when the potential cost is sufficiently high, the optimal level of discretion is a direct result of the agent s desire to avoid congressional ire. Finally, we can characterize the circumstances in which the President will delegate to external agents (e = 1). Given U P, this becomes a straightforward comparison of the resource cost paid to develop policy in-house and the policy loss the president would incur by delegating. In this case, that cost is equivalent to the level of discretion the president must provide in equilibrium. This basic trade-off is presented in Equation 9. e 0 if D > τ = 1 if D τ (9) Moreover, the knowledge of τ also changes the distribution of x, as the committee s optimal sanction rule under some q changes to Demand. In Figure 3, the increase in τ produces a region in which the committee can obtain it s ideal point given knowledge of the costliness of developing policy in house and the conditions for bureaucratic compliance. II.3 Implications The model depicts clear trade-offs for a president with the intent to go it alone. The first, and most critical component of the model is that when the president signs a directive, a new status quo is not final. The policymaking process continues as the agency charged with carrying out the directive and the appropriate congressional committee charged with monitoring the agency attempt to influence the outcome. These additional considerations lead to three hypotheses about the nature of unilateral presidential action, which all suggest that such action is contingent on bureaucratic agency even in the absence of uncertainty. The first and second hypotheses illustrate the trade-off between White House centralization and policy. Mathematically, Hypothesis 1 emerges from Equation 9, but it supports much of what has been written about questions of centralization. Rudalevige (2002), for example, argues that presidents centralize legislative policy formulation 14

16 when the institutional resources within the White House and Executive Office are available. In the context of unilateral presidential policymaking, the model suggests that dwindling resources within the President s immediate control make external agents and the less-than-ideal policies they would select more attractive options. Hypothesis 1. All else equal, as the cost of developing policy within the Executive Office increases, the president delegates to external subordinates more often. Next, consider the influence of the congressional committee s power to sanction. Here, the risk of bureaucratic non-compliance importantly constrains the President s ability to enact x P. The president must buy-off the agent with increased discretion (up to and including the ability to select its own ideal point) as the cost of incurring a congressional sanction increases. Together with Hypothesis 1, this suggests that delegation necessarily involves trading away policy gains a relationship which is compounded when congressional committees can impose greater costs on executive agencies. Hypothesis 2. All else equal, as the cost of congressional sanction increases, the president provides more discretion to external subordinates. Finally, though the precise interpretation is more contingent, the model sheds light on the relationship between the latitude afforded bureaucratic agents, and policy disagreement between the President and Congress. As in Howell (2003), since the model begins with the generation of some exogenous q, the precise relationship depends on the distribution of status quo policies. However, a few stylized facts illustrate what the model suggests we would observe empirically. Recall that, according to σ, the committee sanctions for all status quo policies moved within [x P,x C ]. Thus, as this region expands, the threat of sanction necessarily expands. This directly impacts the level of discretion the president is willing to provide. As the sanction region expands, the President must provide more discretion to buy-off agents closer to the committee (either to avoid sanction or select their preferred policy). This implies the following hypothesis: Hypothesis 3. All else equal, as the policy disagreement between the President and Congress increases, the president will provide less discretion to external subordinates. When inter-branch conflict is high, the president s best option will often be to delegate to an external agent with similar preferences. This agent is provided less discretion, not because the president disfavors the agency or worries about deviation but because the agency itself does not require much discretion to implement a policy which would render it better off. Agencies closer to their committee, on the other hand, will necessarily 15

17 require more discretion (both to satisfy Lemma 1 and to compensate for any sanction incurred), and as a result, paying the presidential resource cost will become more attractive. III. Reanalyzing Executive Orders, Since the proposed model aims to explain variation in the character of executive unilateralism, it is appropriate to reanalyze data which has been the basis for the unilateral politics model s empirical support. Specifically, I model delegation and discretion within significant executive orders from , collected by Howell (2005). The effort to distill the policy significance of legal documents has been a substantial component of work on lawmaking (Mayhew 1991, Clinton & Lapinski 2006) and unilateral action (Mayer & Price 2002, Chiou & Rothenberg 2014), so it is important to underscore the appropriateness of this case selection. Like legislation, it is essential to separate the important orders from the unimportant as a matter of interest. Presidency scholars want to develop theories which explain variation in important rather than mundane actions. For studies of unilateral policymaking, however, there is an added threshold that presidential directives satisfy the basic stylized picture of presidential policymaking. For instance, can exempting an official from mandatory retirement be considered a movement from q to x P? Do these housekeeping orders represent policy selections from a left-right continuum of options? Executive orders, in particular, perform a variety of functions, so that only subset will be of theoretical interest. The 290 non-ceremonial executive orders culled by Howell were selected on the basis of being mentioned in the New York Times. 15 In this case, the profit-motive of the Times usefully helps to satisfy both conditions it serves as a consistent, contemporaneous rater that shies away from reporting the mundane, and thus, produces a set of observations that (for the most part) are worth explaining. As a final step, I omitted 20 orders which establish short term boards of inquiry meant to adjudicate labor-management disputes. Many of these boards were created under the authority of the Labor Management Relations Act of 1947, prescribe no policy movement, and thus, fall outside the scope conditions of either the unilateral politics or presidential delegation models. 15 More recent work by Chiou & Rothenberg (2014) uses multiple raters to model significance as a latent variable. I have limited the analysis presented in this paper for an important reason. As a social scientific enterprise, the perspective I advance speaks directly to the unilateral politics model, so reanalyzing the data that provided its original support is appropriate. 16

18 III.1 Measuring Delegation and Discretion in Executive Orders The presidential delegation model and its implications suggest the empirical analysis must include measures of several key parameters: delegation to external subordinates (e), discretion (D), the president s resource cost (τ), the cost of congressional sanction (c), and the policy disagreement between the president and the congressional committee ( x P x C ). The endogenous parameters (e, D) were produced by hand-coding each order in the sample. The complete coding procedure (reproduced in the Appendix) is adapted from Epstein & O Halloran s (1999) coding of summaries of landmark legislation. It follows a basic logic that presidential directives can be parsed into substantively distinct sections, and that the content of those sections can be used discern the degree of latitude the agent has in implementing policy. First, the number of policy sections in the order is recorded which generally followed the document s section and subsection formatting. Sections which delegate authority are then logged, along with the actors to whom the authority was delegated. Examples of delegated authority vary in degree and cover a wide range of policy areas. They include: orders to carry out a general policy without constraining details, 16 the creation of new agencies, 17 or a directive to develop programs and policies in pursuit of some broader goal. 18 Next, the delegated authority s institutional location is recorded the specific agent(s) referenced by the order. This coding procedure provides two potential measures of External (e). The first is a simple dichotomous indicator variable, coded 1 if the majority of the delegated sections in the order were assigned to external agents. This provides a useful approximation of the decision as outlined in the model presented earlier. However, since some executive orders delegate authority to both the WHO/EOP apparatus and external agents, I also provide a continuous specification the ratio of authority provided to external agents. For the purposes of measuring discretion tracking delegated authority is not sufficient. Recall that the key feature of E.O (Truman s steel mill seizure order) is not simply that it gave the Secretary of Commerce authority, but that authority came with strikingly few explicit restrictions. In other words, it placed few constraints on delegated powers. To account for this, the presence or absence of several constraint types were recorded. Specifically, constraints on delegated authority included: requiring the actor to consult other 16 For example, E.O , issued by John F. Kennedy, which directed the Secretary of Defense to take all necessary steps to intervene for the integration of Alabama s schools. 17 See E.O , which established the Administrative Conference of the United States. 18 See, for example, Richard Nixon s E.O , which provided the Secretary of Commerce with promoting equal opportunity in business ownership. 17

19 actors prior to implementation, 19 requiring the policy be implemented within a time limit, 20 providing explicit details of policy, 21 requiring the actor to submit a report or otherwise notify the President or Congress of its actions, 22 and including a section of definitions within the order. 23 Though extensive coding procedures were outlined, generation of the above variable is inherently subjective. For validation purposes, two researchers with no knowledge of the theory or hypotheses coded a random sample of 40 orders ( 15 percent of the total sample). Despite the complexity of the procedure, inter-coder reliability rates were high, with an ICC of 0.87 for the number of sections and 0.80 for discretion, the dependent variable used in the analysis. A complete description of this validation is included in the Appendix. It may be useful to work through a typical example. Executive Order 12129, which was issued by Jimmy Carter in 1979, contains 10 substantive sections describing the creation of the Critical Energy Facility Program a unilateral" predecessor to energy policy reform in Congress which would come in Four of those sections explicitly delegate authority to the Director of the Office of Management and Budget to institute the program which meant selecting energy producing facilities to receive special permits for construction. However, the order also requires the OMB to consult with other agencies, notify the President and the agencies of their decision, and adhere to procedures and guiding policy principles produced by a previous interagency council. So, though the order clearly delegates power to the Director, it provides boundaries wherein the OMB may operate. In other words, the order delegates, but it does not provide substantial discretion. Combining the above information into a useful index necessarily involves imposing some functional form on the latent outcome of discretion. Epstein and O Halloran (1999) developed a discretion index by counting delegated provisions and major provisions of legislation. Recall that the level of discretion (D) allows the agent to select a policy, ˆx, which is, by definition any point between [x P D,x P + D]. In order to normalize the index, suppose further, that Discretion [0, 1]. That is, discretion is a bounded ratio where 1 denotes complete discretion over some delegated power, and 0 denotes no delegated authority. To produce values of 19 Though E.O gave applicable departments the authority to develop their own pollution-prevention standards for public facilities, it required they consult with the Secretary of Health, Education and Welfare before putting the policies in place. 20 For example, E.O required the National Commission on the Civil Disorders set an explicit deadline for the Commission to submit final recommendations and disband. 21 Though Lyndon Johnson gave the Civil Service Commission to promulgate regulations (E.O ) relating to ethical conduction of public personnel, he also included an itemized list of what constituted unethical conduct, limiting the authority of the Commission to deviate from those standards. 22 For example, Kennedy s E.O required the ACUS to submit reports to the President on a recurring basis. 23 E.O , which prescribed policies for coral reef protection, defined coral reef ecosystems at the outset, which constrained the application of the order by the Secretaries of Interior and Commerce. 24 Carter s plan was rejected by Congress from by the Senate. 18

20 discretion within those bounds, I impose the function in Equation 1. Here, the first term ( d p ) is the number of delegated sections (d) over the total number of sections (p), and represents the amount of delegated authority. The second term is a constraint penalty: the number of constraints in the order (c) divided by the total number of constraint types, multiplied by the amount of delegated authority. To use the previous example Carter s Executive Order the delegation term would be 0.4, and the constraint penalty would be 0.24, for a discretion index of D = d p [ d p c t ] (10) In this way, this discretion index is the proportion of subsections which contain delegated authority, weighted downward by the number of constraints placed on executive agents. The index itself is somewhat arbitrary, in that it weighs each constraint equally. Moreover, each delegated section is also given equal weight despite variance in substantive significance. However, alternative measurement strategies necessitate costly trade-offs. The ratio can be deconstructed and understood in terms of its constitutive parts, such that the point estimates presented later retain some degree of interpretability. Though adopting a hierarchical measurement model (à la Johnson & Lambert 1999, Clinton & Lapinski 2006, Chiou & Rothenberg 2014) may be a useful validation of the dependent variable, the estimates generated would be less meaningful. On the other hand, requiring some rating procedure to weigh the significance of individual provisions would introduce more researcher bias, and render the results far less replicable. The measurement strategy employed here, then, provides an appropriate middle ground. The validity of this measurement strategy must be judged, in part, by a qualitative assessment of the values it produces. To aid in that validation, I have reproduced the tails of the discretion distribution in Appendix C. These tables show the ten orders with the highest and lowest (non-zero) levels of discretion, as determined by the measurement strategy above. Overall, the examples seem to be accurate representations of the key construct. The low discretion orders are all near examples of direct policymaking. They prescribe changes to government policies explicitly, and when they delegate, they place considerable constraints on the actors to whom authority was delegated. Notable examples include changes to regulations governing the secrecy of documents and information. The high discretion tail, on the other hand, shows Presidents vesting considerable, significant authority in a single agency without much detail or explicit limitation. Several of these orders came as Presidents gave the Secretary of Defense extraordinary authority to use military force 19