1. Introduction
In June of 2002, during George W. Bush’s first term as president, the Republican-controlled House Subcommittee on Fisheries Conservation, Wildlife and Oceans held an oversight hearing on coral reef conservation. A primary focus of the hearing was the implementation of a Clinton-era executive order establishing a Coral Reef Task Force.Footnote 1 The order was far-reaching, coordinating agency efforts in conservation and restoration of reefs, mapping and monitoring efforts, and engagement with the scientific research community.Footnote 2 As Republican subcommittee chair Wayne Gilchrest stated in his opening remarks, the committee was “seeking to examine what activities have been conducted and, importantly, the level of success that they have achieved.”Footnote 3
The hearing transcript reveals members of both parties deeply engaging with agency witnesses about their efforts to implement the executive order, asking questions about agency reef mapping efforts, international coordination, and funding for research programs. One line of questioning focused on how Congress could help with agency implementation efforts: Democratic Representative Neil Abercrombie of Hawaii pushed a witness from the Department of the Interior on what types of legislative action Congress could take to facilitate their work, and whether the requested budget was sufficient for the undertaking. The general tone of the hearing was substantive and collaborative, and the hearings were followed by a report to Congress detailing Task Force and agency activities.
This example highlights two points about the role of presidential unilateral action in the US separation of powers system. First, “unilateral actions” are not, generally speaking, literally “unilateral.” Rather, they are directives to executive agencies, requiring agency cooperation for their implementation. Clinton’s executive order establishing the Coral Reef Task Force did not itself constitute a national coral reef conservation strategy; rather the work of developing and implementing such a strategy was devolved to the Task Force and the agencies of the executive branch. Second, unilateral action does not always cut Congress out of the policymaking process. Even when Congress is not directly involved in the development of an executive order, the fact of such orders requiring action from agencies creates an opportunity for Congress, who shares oversight responsibility with the president, to shape agency implementation. These two points—that unilateral action is an act of delegation to administrative agencies, and that this delegation creates an opening for congressional influence—are the focus of this paper.
Existing scholarship often considers the exercise of presidential unilateralism as itself a sign of expanding executive power at the expense of Congress (e.g., Moe and Howell, Reference Moe and Howell1999; Howell, Reference Howell2003; Rudalevige, Reference Rudalevige2005). In this light, the use of executive unilateralism raises important questions about the appropriate role of the executive branch in a contemporary separation-of-powers system. On the one hand, the parochialism and collective action problems inherent in Congress may make the presidency a more appropriate agent of policymaking for today’s complex and fast-paced political environment (e.g., Howell and Moe, Reference Howell and Moe2016). On the other hand, the expansion of presidential power raises important questions about how to maintain the balance between the legislative and executive branches that was intended by the constitutional framers.
Existing scholarship exploring these issues typically characterizes unilateral action as the president “going it alone” in pursuit of his policy objectives, focusing on institutional-level conflict between the president and Congress. Much of this literature takes the frame of the president acting strategically to either thwart or incentivize congressional action. From that perspective, scholars have investigated how many executive orders (or “significant” executive orders) are issued under different conditions such as unified or divided government, timing within the president’s administration, periods of congressional gridlock, and the ability of the president to influence his copartisans in Congress (cf. Waterman, Reference Waterman2009; Lowande and Rogowski, Reference Lowande and Rogowski2021). Recent scholarship, however, has complicated the story by calling into question the extent to which unilateral action is truly “unilateral” at all. Agencies are involved in formulating executive orders in the first place (Rudalevige, Reference Rudalevige2021), and implement the directives once they have been issued (Kennedy and Rudalevige, Reference Kennedy and Rudalevige2025). As Lowande and Rogowski explain, “[t]aken literally, ‘unilateral’ power is only unilateral with respect to Congress, the Judiciary or other non-executive actors. Every presidential directive is an order to an administrator. We suspect few researchers would disagree with these points, but how they ought to inform theories of unilateral power is contestable. To date they mostly have not,” (Reference Lowande and Rogowski2021, p. 8).
This paper builds on this emerging area of research by investigating how the administrative politics of unilateral action shape inter-branch political dynamics. The argument begins with the premise that executive orders are directives to executive-branch agents. Because agencies are subject to oversight not only by the president but also by Congress, this opens the possibility that Congress can shape policy outcomes by exercising oversight of the implementation of executive orders. The conventional expectation is that Congress can pass legislation in direct response to executive orders with which they disagree. This paper, on the other hand, focuses on an indirect response: by exercising oversight over implementing agencies, congressional committees can shape policy outcomes at the agency level. This possibility was explored theoretically by Lowande (Reference Lowande2018a), and this paper expands on that work with empirical analysis.
In particular, I test the hypothesis that how presidents delegate in unilateral actions—the amount of authority they grant to agents (“delegation”), and the flexibility they provide in the use of that authority (“discretion”)—informs Congress’s oversight activity. I find that when presidents delegate more authority to agents to implement directives, and when that authority comes with a broader grant of discretion, Congress is more likely to engage in oversight of that implementation. Overall, I demonstrate a larger role for Congress in shaping and overseeing the implementation of unilateral policy than has previously been appreciated.
In the next section, I will describe in more detail the theoretical framework for this analysis. In particular, I will discuss how delegation and discretion can be understood in the executive branch context, and provide theoretical expectations for the anticipated relationship between such delegation, on the one hand, and congressional oversight activity on the other. I then describe the data and empirical approach that I employ in the analysis, including the measures of delegation and discretion. Results of the analysis follow, and a discussion of these results and next steps for research concludes the paper.
2. Delegation and oversight in existing scholarship
2.1. Delegation in the executive branch
The related concepts of delegation and discretion are the subject of extensive scholarship. Definitions of the two concepts tend to follow similar lines: delegation refers to delegated authority, or empowering agents to take actions; while discretion refers to the latitude agents have in choosing how to take those actions.
A number of approaches have been taken to studying delegation and discretion empirically in the legislative context, including using document length as a proxy measure for discretion (e.g., Huber and Shipan, Reference Huber and Shipan2002); or ratio of new budgetary authority (delegation) to document length (discretion) (Bolton and Thrower, Reference Bolton and Thrower2019). A frequently-cited approach comes from Epstein and O’Halloran (Reference Epstein and O’Halloran1999), who developed a “delegation ratio,” representing the proportion of provisions in a bill that grant authority to the executive branch, and a “constraint ratio,” representing the proportion of 14 possible types of procedural constraints that are present in a given bill.
In the executive branch context, there is a rich theoretical literature on delegation to administrative agents and the associated discretionary windows (e.g., Bendor and Meirowitz, Reference Bendor and Meirowitz2004; Gailmard and Patty, Reference Gailmard and Patty2012). However, empirical work on the president-as-principal has tended to focus on the president’s use of the tools of the administrative presidency, such as appointments and centralization, to shape bureaucratic incentives more generally (e.g., Rudalevige, Reference Rudalevige2002; Lewis, Reference Lewis2003, Reference Lewis2008). Explicit empirical measures of delegated authority and discretionary windows have thus far been limited to congressional principals.
The underlying concepts of delegation and discretion are analogous in the executive and legislative context. The primary difference is of course that legislative authority rests solely in Congress, and can be delegated to administrative agencies by Congress. In the executive context, delegation can refer to delegations of executive authority (either constitutionally granted or delegated to the president by statute), but also presidential direction or activation of agencies’ statutory authority. The original authority for agency action may stem from Congress but, as a practical matter, presidents can shape how agencies use it. This characterization of the political role of the president in activating and directing agencies’ statutory authorization is consistent with Kagan’s legal argument that “Congress generally should be understood to have left authority in the President to direct executive branch officials in the exercise of their delegated discretion” (Reference Kagan2001 p. 2246).
For this analysis, delegation refers to the scope of the agent’s directed actions; for example issuing permits, making determinations, or promulgating rules that affect whole industries. Importantly, it does not refer to how much policy is expected to move in a spatial sense: agencies can promulgate a rule that codifies the status quo, for example. Discretion refers to how much flexibility agents have in determining which policy ends to pursue through the exercise of that authority, and how to pursue those ends. I discuss how I operationalize these concepts into usable measures below.
2.2. Congressional oversight
The mechanisms by which Congress is able to shape executive branch activity are numerous, and go well beyond direct legislation. Indeed, it is unlikely that legislation is a particularly frequent approach, due to the difficulties associated with its utilization. As Kriner and Schickler explain, “[i]nformation asymmetries, steep transaction costs in coalition building, and the looming threat of a filibuster or presidential veto all suggest that legislative efforts to constrain presidents will often fail, even when a strong majority of members opposes the president’s action” (Reference Kriner and Schickler2016, p. 5). We must cast a wider net in the search for congressional modes of shaping and responding to agency behavior.
The tools, both formal and informal, that Congress can use to oversee agency activity are extremely varied. Indeed, Feinstein (Reference Feinstein2018) notes that almost all congressional activity could be considered “oversight” in the sense of shaping bureaucratic activity. Taking a similarly broad view, Selin and Moore define oversight as “the process of reviewing, monitoring, and supervising the implementation of public policy by the executive branch” (Reference Selin and Moore2023 p. 187), and identify a wide range of overlooked oversight mechanisms including program evaluation, staff review of executive programs and policies, and ad hoc intervention. Other work examines procedural oversight (Lowande and Potter, Reference Lowande and Potter2021), informal contacts between members of Congress and agencies (Lowande, Reference Lowande2018b; Ritchie, Reference Ritchie2023), and appropriations committee reports (Bolton, Reference Bolton2022).
For this analysis, I follow extensive precedent in using congressional hearings as a measure of oversight activity (e.g., Aberbach, Reference Aberbach1990; Kriner and Schickler, Reference Kriner and Schickler2016; MacDonald and McGrath, Reference MacDonald and McGrath2016). One benefit of using committee hearings is that they enable me to clearly link oversight activity to individual executive orders. This linkage, which is critical for my analysis, is unfortunately not feasible with currently available data on other types of oversight such as correspondence logs and committee reports.
Caution is warranted when extrapolating from hearings to congressional oversight more generally, because patterns of oversight vary across modalities (Aberbach, Reference Aberbach1990; Selin and Moore, Reference Selin and Moore2023). However, in this case I argue that hearing activity should be taken as a “hard test” of my theory. Hearings are extremely costly to participants—both to agencies and to members. Given the array of other modes of oversight available to members, it is thus likely that hearings represent only a small fraction of the oversight activity that Congress engages in for a given policy issue. Evidence to support this idea can be found in the hearing transcripts themselves: members frequently reference reports they requested, letters they sent, or conversations they had with agency personnel when giving remarks during a hearing. For example, in 1973 a House subcommittee held a hearing about two executive orders issued by President Nixon that allowed the Department of Agriculture to inspect farmers’ tax returns for statistical purposes.Footnote 4 Members were concerned that the IRS regulations implementing the order did not put sufficient limits on the amount or type of information that would be released. Representative Jerry Litton testified that, because of this concern, he “began a personal investigation” which included “several off-the-record conversations with Department of Agriculture officials,” and provided quotations from letters between agency personnel and committee members. While future research should expand the analysis to other forms of oversight, hearings are a reasonable proxy for oversight activity for the purposes of the current study.
2.3. Theoretical expectations
Two important theoretical premises emerge from the foregoing discussion of existing scholarship. First is the premise that unilateral action, generally speaking, relies on implementation from agencies. The president is not a unitary actor in a literal sense, but rather relies on agency cooperation to pursue unilateral policy change. When a president issues a unilateral directive, he is directing agencies to take a particular action. How much agencies are being asked to do, and how much latitude they have in doing it, can vary from directive to directive.
The second premise is that agencies must be responsive not only to presidents, but to Congress as well. The tools that Congress has to shape agency policymaking are varied, and Congress, both collectively and as individual members, deploy those tools to try to maintain some control over agency policy outcomes.
Combining these two premises results in the main argument of this paper: when a president issues a unilateral directive, that directive shifts policymaking from the White House into the agencies. To the extent that policymaking is shifted to the agency, then, Congress can exercise its usual control mechanisms to oversee agency implementation. Overall we would expect to see that as more decision making and policy implementation are passed to an agency, the more opening there is for congressional intervention.
This expectation is formalized in the Delegation hypothesis, below. In particular, I expect that hearing activity should be increasing in administrative delegation. The logic behind this expectation is that delegation to agencies enables Congress to try to shape the policy outcome. If an executive order makes a determination, for example, without any need for additional agency action, then there is no opening for Congress to shape implementation. If, however, the directive requests new regulations, Congress has a chance to influence the ensuing rulemaking process. Similarly, if the directive requires agency implementation of a policy program, Congress can try to influence how that implementation takes shape.
Delegation hypothesis: Oversight activity is increasing in the amount of delegated authority in an Executive Order.
A related hypothesis concerns discretion. Recall that while delegation describes the extent of agency action, discretion describes the flexibility agencies have in choosing how to take that action. As with delegation, more discretion is an indicator that decision-making has moved into the agencies, and thus is more susceptible to congressional oversight. This expectation is formalized in the Discretion hypothesis.
Discretion hypothesis: Oversight activity is increasing in the amount of discretion in an Executive Order.
To summarize, both delegation and discretion indicate that policy implementation and/or design is located in agencies. Because Congress can exercise oversight to shape agency activities, we should thus expect the extent of congressional oversight to be increasing in both of these characteristics. But how should they interact? In particular, does the extent of discretion in a directive condition the relationship between delegation and oversight? My expectation is that they have a mutually-reinforcing effect. A directive that includes more delegation is asking the agency to do more, and a directive with more discretion is asking the agency to decide more about how to do it. Either of these dimensions alone would plausibly invite congressional oversight. But together, they indicate that the agency has both broad responsibility and broad flexibility, which should mean that there is ample opportunity for a congressional intervention to shape outcomes. Thus I expect that, while the association between delegation and discretion and oversight are separately positive, they should have a mutually-reinforcing impact. This expectation is formalized in the Discretion interaction hypothesis:
Discretion interaction hypothesis: The relationship between oversight activity and Delegation depends on the level of Discretion. In particular, the association between Delegation and oversight should be larger when Discretion is also high.
The next section includes details about the data used to examine these hypotheses, including how Delegation and Discretion are measured and operationalized.
3. Data and empirical approach
3.1. Data on unilateral actions
The empirical analysis draws on two novel datasets: a set of unilateral actions, and a set of congressional hearings. The unilateral actions dataset includes 1421 executive orders issued between 1970 and 2021. These actions were sampled randomly from the 2518 executive orders issued over that date range.Footnote 5 Each document was reviewed by a trained coder, who determined the degree of delegation and discretion granted in each case, using measures developed by Benn (Reference Benn2023).Footnote 6 In the final dataset, each order is scored as either “High” or “Low” on each of Delegation and Discretion.Footnote 7
The Delegation score is assigned based on how much authority to act the agent is granted by the directive. “High” scores are assigned to orders that request agencies implement programs, or authorize general policymaking authority. “Low” scores are associated with orders that grant agencies only indirect policymaking authority, such as those requesting recommendations, or orders that make presidential determinations without explicitly authorizing agency action. The difference between an order scoring “High” or “Low” is essentially whether the order gives the agency direct policymaking authority, or not.Footnote 8
Discretion scores are assigned similarly, but based on the decision-making flexibility granted in the order rather than on the extent of agency empowerment to act. Orders with narrowly-defined policy objectives receive “Low” scores, as there is minimal space for agencies to exercise policy discretion. Orders receiving “High” scores request agencies to develop policies and plans, and can even leave the objective of policy action open-ended. An order that is “Low” on discretion may list particular criteria for making determinations, or provide details about what an agency regulation should accomplish and how. An order scoring “High” on discretion is more open-ended, stating a goal or policy area, but leaving the details up to the agency.
Table 1 provides general descriptions of the types of orders that receive each combination of scores. Documents that score Low on both dimensions are typically revisions to definitions, minor amendments to previous orders (for example changing a date), or presidential determinations without separate requests for action. Documents that score High on both dimensions typically request agencies to both design and then implement large-scale policy programs. Documents that score High on Discretion but Low on Delegation provide broad flexibility to agents to design policy approaches, but the agent is not granted authority to implement any programs. Instead, these orders are most often associated with the creation of advisory commissions, whose function is to make policy recommendations. Finally, orders scoring High on Delegation but Low on Discretion request agency action, typically rulemaking, but the order narrowly defines the policy objectives and details of the program to be implemented, thus severely limiting implementation flexibility. Table 2 presents frequencies of unilateral actions in the sample by Delegation and Discretion scores.
Table 1. Typical directives by score combination
3.2. Data on congressional hearings
The executive orders described in the previous section are the unit of analysis for this project, and Delegation and Discretion the primary explanatory variables of interest. The second dataset, comprising congressional hearings, provides information about the dependent variable of interest: oversight activity. The hearings dataset includes all hearings pertaining to oversight of executive orders between 1970 and 2021 (
$n = 522$). To identify these hearings, I searched the ProQuest Congressional database for the term “executive order” anywhere but the full text of the hearing.Footnote 9 This includes the title of the hearing, the hearing summary/abstract, the titles of included evidentiary exhibits, and the summaries of testimony content. Each hearing transcript was then manually reviewed to confirm that an executive order was discussed during the hearing, and to identify which order(s). In total, I identified 362 hearings, representing 432 executive orders, as some hearings cover more than one order.Footnote 10
The nature and substance of the hearings were quite varied. Some were focused on bipartisan information gathering, while others were more contentious. There were some hearings that focused explicitly on implementation of an executive order, while other hearings focused on a broader policy area and included implementation of a particular order as part of that discussion. A hearing was only included in the analysis if there was a substantive discussion about an executive order at some point during the hearing—passing reference to an executive order was not sufficient for inclusion.
As an example of the type of discussion I was looking for, in 1986 President Reagan issued an executive order pursuant to the Omnibus Drug Law of 1986 mandating drug testing of federal employees.Footnote 11 The Office of Personnel Management (OPM) issued regulations implementing the order later that year, and the Department of Health and Human Services (HHS) issued “scientific and technical guidelines” the following February. Members of Congress were concerned that the regulations promulgated by OPM were too punitive and, in a hearing later that year, members of the House Subcommittee on Human Resources expressed doubts about the constitutionality of the program, its cost, and the accuracy and efficacy of testing.Footnote 12 Later that same year, Congress used a supplemental appropriations act to prohibit the use of appropriations for any drug testing pursuant to the executive order until HHS had issued new guidelines, consistent with the specifications provided by Congress in the Act.Footnote 13
Table 2. Sample distributrion of delegation and discretion scores
A similar example of Congress taking a close look at executive order implementation comes from the Obama administration. In 2014, President Obama issued an executive order setting the minimum wage for federal contractors at $10.10.Footnote 14 During the notice and comment period for the regulations implementing the order, the Department of Labor (DOL) received a request that seasonal guides and outfitters operating on federal lands should be excluded. The final regulations, however, did not make this exception. The Republican-controlled House Subcommittee on the Interior held a hearing in 2015 requesting that DOL make an exception for these types of entities, arguing that they are not “federal contractors” in the traditional sense.Footnote 15 Witness testimony from business owners argued that the new requirement would have a severe impact on their ability to operate, and in many cases would cause such enterprises to shut down or resort to charging exorbitant fees. Despite strong questioning by committee members, DOL did not change the rule. However, at the beginning of his first administration, President Trump issued an executive order making exactly the exemption that Republican lawmakers had been pushing for during the hearings.Footnote 16
To be sure, there are also hearings that focus on accusations of presidential overreach rather than agency implementation. A prominent example is the large number of hearings about President Reagan’s Executive Order 12291Footnote 17 institutionalizing centralized regulatory review in the Office of Management and Budget (OMB). Some questioning during the hearings focused on OMB’s regulations and decision-making in implementing the executive order, but the general focus of hearing activity was on the constitutionality, and advisability, of this major change to how government regulations would be developed and overseen. Clinton’s Executive Order 12866Footnote 18 and Obama’s Executive Order 13563,Footnote 19 making further amendments to the regulatory review process, received similar attention.
But in general, cases of macro-political battles about the appropriate separation of legislative and executive functions are the exception rather than the rule. Instead, congressional hearings about executive orders tend to serve as an opportunity for legislators to try to influence agency decision-making. Such hearings focus primarily on understanding orders’ design, implementation, and impact; and on steps that Congress can take to shape outcomes to its preferences. Moreover, as I will demonstrate in the empirical sections that follow, I argue that it is specifically the fact that executive orders are themselves acts of delegation to agencies that facilitates this participatory role for Congress.
3.3. Empirical approach
To analyze the roles of delegation and discretion in shaping congressional oversight activity, I combine the executive order and hearings datasets. For each executive order in my sample, I indicate whether a hearing about it appears in the hearings dataset. This approach maintains the order as the unit of analysis, and avoids selecting on the dependent variable, hearings. In assigning these outcome values, I am implicitly assuming that if an order received a hearing, that hearing appears in my dataset. Conversely, if there is no hearing for an order in the hearings dataset, I assume it is because no hearing occurred.Footnote 20 One implication of this approach is that not every hearing that has been reviewed will appear in the dataset: a hearing’s inclusion depends on whether the associated executive order is in the sample. The final dataset includes 1421 executive orders. After matching hearings to executive orders, 101 (about 7%) are coded as having had a hearing. Because the sample does not contain the full universe of orders issued during the time period, not every order that was the subject of a hearing appears in my sample. Of the 432 executive orders identified as being the subject of a hearing, 195 (about 45%) have been included in this sample.
To operationalize hearing activity, I use two variables: Any Hearing is a binary variable that captures whether at least one hearing was held about a given executive order during any Congress. Total Hearings represents the total number of observations (entries in the Proquest search results) in the hearings dataset that were coded as pertaining to that executive order. Delegation is the main explanatory variable of interest. In all models, I interact Delegation with Discretion. This choice reflects that Delegation operates in combination with Discretion rather than independently: presidents choose levels of both variables simultaneously.Footnote 21
In the results that follow, I conduct two main analyses. The baseline analyses are at the level of the executive order, and use Delegation and Discretion to predict hearing activity. In addition to these main explanatory variables, I estimate separate models including fixed effects by:
• President—the president who issued the order.
• Year—the year the order was signed.
• Policy—the policy area of the order, as measured by “major topic” codes from the Policy Agendas Project.
One commonly used control in executive order scholarship is a measure of order “significance,” such as the measure developed by Chiou and Rothenberg (Reference Chiou and Rothenberg2017). Significance measures are included to control for the policy salience or policy impact of a given order. It would be reasonable to expect that Congress would be more interested in conducting hearings on “significant” unilateral actions. However, in this case significance is a post-treatment variable. In fact, whether reference to the order appears in the Congressional Record is one of the raters used in the IRT model that Chiou and Rothenberg used to construct their significance scores. Including significance would thus be using a measure that itself reflects hearing activity to predict hearing activity. Accordingly, throughout this analysis, significance is omitted from the specifications. As a means to control for policy salience, I include the policy-area fixed effects described above. This is not exactly the same as an order-level significance score, but should account for variation in generally high- or low-salience policy areas.Footnote 22
Controls are limited in the baseline models, because many of the covariates that predict hearing activity are measured at the Congress-level, not the order-level. To address this issue, I also include models estimated at the order-Congress-chamber level. That is, I reshape the data such that each observation, rather than being an individual executive order, is an executive order in a given Congress, in a given chamber. The outcome variables are then whether a hearing was conducted or the number of hearings conducted about a particular order in a given chamber during that Congress. The motivation for this change is to leverage differences between the party of the president who issued the executive order, and the party with control of a given chamber at a given time. The analysis thus identifies any hearing activity, not only hearing activity during the Congress in which the order was issued.
In the full specifications using this reshaped data, I add several controls:
• Divided Government indicates whether the party in control of a given chamber at a given time is different from the party of the president who issued the order.
• Senate is an indicator for whether the observation reflects the Senate or House.
An additional concern with the expanded data is that the likelihood of an order receiving a hearing in a given period is not independent of the likelihood of that same order receiving a hearing in any other period. It may be that some orders are more likely to receive hearings overall, or that once an order has received a hearing it is not likely to receive another one. At any rate, we cannot assume temporal independence of observations.Footnote 23 To address this issue, I include two additional controls that account for this cross-temporal relationship:
• Executive Order Age is constructed by counting the number of Congresses that have passed since the order was issued. This variable is included to account for the possibility that Congress may be more or less interested in an order’s implementation at different points in the order’s lifespan.
• Previous Hearings represents the total number of hearings the order received in all previous periods.
The results of these analyses are presented in the next section.
4. Results
As an initial illustration of the relationship between Delegation and Discretion and oversight activity, Table 3 shows the proportion of orders with each score combination receiving any hearing activity. Approximately 7% of all orders in the sample received a hearing, but the proportion of orders receiving a hearing varies considerably depending on the Delegation and Discretion scores. In particular, while only 2.5% of orders scoring Low on both dimensions received a hearing, 12% of orders that scored High on both dimensions received hearings. Scoring High on only one dimension was also associated with increased hearing activity. The table suggests a strong positive relationship between hearing activity and higher Delegation and Discretion scores, offering preliminary support for both the Delegation and Discretion hypotheses. The Interaction hypothesis, however, does not have clear support. While the category with the most hearings is indeed orders scoring High on both dimensions, it is not clear that the cumulative association is larger than the component parts.
Table 3. Proportion receiving hearing, by score category
To explore these relationships more systematically, I estimate a series of baseline models with Delegation interacted with Discretion as the explanatory variables, and the two measures of hearing activity (Any Hearing and Total Hearings) as the dependent variable. In Table 4, Specifications (1) through (4) estimate a Logit model predicting the binary outcome Any Hearing. Specifications (5) through (8) estimate the total number of hearings, using a negative binomial model to account for the right-skew and overdispersion of the underlying count data. In addition to the baseline model, all models are estimated with fixed effects by President, by Year, or by Policy Area, as discussed in theprevious section.Footnote 24
Table 4. Baseline specifications
+p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.
In all models, there is a statistically significant positive association between Delegation and the outcome variable. That is, in all models, an order that authorizes agency action is more likely to receive hearing activity than an order that has a more limited role for the agency. This provides preliminary support for the Delegation hypothesis: it is exactly when the agency has been given the authority to act that Congress has the opportunity to perform its oversight role. The Discretion hypothesis also receives support, with the coefficient on Discretion positive and statistically significant in all specifications.
The interaction with Discretion is significant for the Logit models, but not for the negative binomial models. This could reflect additional noise in the measurement of Total Hearings, which is only a rough proxy for extent of hearing activity.Footnote 25 However, the sign on the interaction term is negative in all models, suggesting that the association between Delegation and hearing activity is weaker when Discretion is high than when Discretion is low. However, while the individual component associations between Delegation/Discretion and hearing activity are weakened when they are both High, the cumulative association is still positive. That is, hearing activity is most likely when both Delegation and Discretion are high.
To facilitate interpretation of the interaction term, Figure 1 presents the results of models (1) and (5) from Table 4. The left panel shows the estimates for Any Hearing, and the right panel shows the estimates for Total Hearings. The x-axis indicates whether Delegation was low (0) or high (1), and the y-axis shows the predicted value of the dependent variable (either Any Hearing or Total Hearings). The blue triangle shows the estimate when Discretion is high, and the red circle when Discretion is low.
Figure 1. Predicted hearing activity by delegation and discretion (baseline).
The figure indicates that, irrespective of the level of Discretion, High Delegation is associated with an increase in hearing activity. That is, executive orders scoring High on Delegation are more likely to receive a hearing than executive orders scoring Low on Delegation. This is true for both High and Low Discretion. However, the association is dampened when Discretion is also high—the association between Delegation and hearing activity is still positive but has a smaller magnitude when Discretion is high than when Discretion is low. However, because both Delegation and Discretion are both positively associated with oversight activity, the cumulative association is still strongly positive. That is, hearing activity is most likely when both Delegation and Discretion are high.
One possible explanation for this counterintuitive finding is that either Delegation or Discretion provides an opening for Congress to try to shape agency outcomes. The mechanism for each dimension to facilitate congressional oversight is similar: high delegation and discretion both represent policy implementation or decisionmaking being put in the hands of the agency. It is that location in the agency that facilitates Congress’s response—their role as overseers of agency action means that they can shape the activities of the agency. It may be that one of these dimensions is sufficient, and that the marginal benefit of high scores on both dimensions is minimal.
4.1. Delegation, discretion, and divided government
While the baseline models provide support for the Delegation and Discretion hypotheses, a more robust investigation requires consideration of partisanship and other Congress-level factors. To facilitate these analyses, I reshape the data to the order-Congress-chamber level, as described previously. In the following analyses, each observation represents a given executive order in a particular Congress in a particular chamber. So, for example, Divided Government indicates whether the party in control of a particular chamber at a particular time is different from the party of the president who originally issued the order.
After expanding observations for each executive order to the order-Congress-chamber level, the number of observations grows to 51,202. Table 5 shows the results of the expanded analysis.Footnote 26 As previously, specifications (1) through (4) estimate Any Hearing using Logit models, and specifications (5) through (8) estimate Total Hearings, using negative binomial models. As in the baseline models above, each specification is re-estimated with President, Year, or Policy fixed effects.
Table 5. Full specifications, any EO age
+p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001
As expected, the relationship between Delegation and hearing activity remains positive and statistically significant in all specifications, providing further support for the Delegation hypothesis. The Discretion hypothesis is similarly supported, with statistically significant positive associations in all specifications. In other words, an executive order scoring high on Delegation (Discretion) is more likely to receive a hearing than an executive order scoring low on that dimension.
The interaction between Delegation and Discretion is similar to the baseline specifications: the coefficient is negative in all specifications, and statistically significant at the .05 level in only 4 of the 8 specifications. As before, this negative coefficient does not undercut the positive associations between the individual terms: orders scoring High on both Delegation and Discretion are more likely to receive a hearing than orders scoring High on only one dimension. However the individual association between Delegation (Discretion) and hearing activity is weaker when both dimensions are High than only one dimension.
Surprisingly, the control for Divided Government does not reach statistical significance in any of the specifications. That is, an order is no more likely to receive a hearing during a session if the party in control of a given chamber at the time is different than the party of the president who originally issued the order. If hearing activity were motivated primarily by partisan conflict with the issuing president, we might expect to see a stronger role for Divided Government. Instead, it appears that Congress is motivated to try to shape outcomes in the agency.
Overall, the results of these analyses offer strong, consistent support for the Delegation and Discretion hypotheses. When an order directs an agency to implement policy and make decisions, Congress is more likely to work to shape how the agency responds to that request. This relationship does not appear to be dependent on partisanship, as measured by Divided Government, or on issue salience, as measured with the policy area fixed effect.
The interpretation of the Discretion Interaction hypothesis is more nuanced. While the association between Delegation and hearing activity is not strengthened by High Discretion, the cumulative effect of High Discretion and High Delegation together on hearing activity is strongly positive. In other words, orders scoring High on both dimensions are more likely to receive a hearing than orders scoring High on only one dimension.
5. Discussion
The preceding analyses demonstrate that congressional hearing activity is associated with delegation made by the president to agencies. Executive orders that contain larger grants of authority and flexibility to agencies are more likely to trigger congressional hearings than orders that do not provide agencies with a similar degree of authority and latitude. I argue that this act of delegation creates an opportunity for Congress to respond to unilateralism by shaping implementation. It is exactly the fact that unilateral directives require agency implementation that makes them susceptible to congressional influence.
While this paper provided evidence of an association between delegation and hearing activity, an important question that this finding raises is whether those hearings are effective at changing agency behavior. Explicit empirical evidence of the effects of congressional oversight on executive-branch policy is difficult to identify for several reasons. First, the long time horizons of federal policymaking mean that a congressional intervention might have an observable impact only years later. Additionally, congressional influence may be felt in ways that are observable in agency operations, but not documented in formal changes to rules. Finally it is difficult, if not impossible, to disentangle the influence of congressional oversight activity from the effects of other political actors: oversight activity is most likely for issues that are high salience to at least some constituency, and that constituency is likely to be pressuring agencies both through and external to Congress (McCubbins, Noll and Weingast, Reference McCubbins, Noll and Weingast1987; Boehmke, Gailmard and Patty, Reference Boehmke, Gailmard and Patty2013).
Although the exact effects of hearing activity on agency activity are difficult to determine, some preliminary evidence can guide further research. First, Congress sometimes does take explicit action after a hearing, for example in the form of limitations on how agencies can use appropriated funds. An example of this was presented previously, when Congress forbade agencies from spending appropriations implementing President Reagan’s mandatory federal employee drug testing program. A similar example is provided by President Clinton’s 1994 Executive Order requiring the EPA to consider environmental justice in developing its programs and policies.Footnote 27 In 2007, the Democrat-controlled Senate Subcommittee on Superfund and Environmental Health held a hearing to express disapproval of the EPA’s implementation of this order under the George W. Bush administration. The main focus of the hearing was the EPA’s relaxation of the Toxic Release Inventory reporting requirements in 2006. After Bush left office, the 2009 omnibus appropriations bill included a section invalidating the particular rule that had been the subject of the hearing. While the Clinton Executive Order was itself not at issue, this example highlights the ways that Congress tries to shape implementation over time and across administrations.
Congress can also play a role in preventing executive orders from being issued in the first place. In a high-profile example, in 2011 the Obama administration circulated a draft executive order that would require additional disclosures of employee political contributions from potential federal contractors. The Republican-controlled House Committees on Oversight and Small Business held a joint hearing entitled “Politicizing Procurement,” raising strong objections to the proposed order. While the Democrats in Congress were largely in favor of the order, pointing out that contractors were already subject to disclosure requirements, the partisan acrimony in the hearing reflected broader opposition to the order among affected groups. The news media followed the progress of the order, with Reuters reporting after the hearing that the order was “feared stalled.”Footnote 28 Years later, in 2016, the New York Times reported that Obama was “seriously considering” issuing the executive order.Footnote 29 The order, however, was never issued. As in the previous example, the role of congressional hearings cannot be easily disentangled from pressure faced by the Obama administration from other sources. However it seems likely that the hearings and other modes of congressional oversight served as part of the constellation of political factors that shaped this outcome.
An additional question raised by these results is whether the president can select levels of delegation of discretion strategically to avoid triggering oversight. This question is ripe for further research and theoretical development, but I argue that it is not of particular concern for the results presented here. Because of the way delegation and discretion are measured, they reflect exactly what a president has asked an agency to do. Thus the uppermost consideration must be what the president hopes to accomplish. If the president seeks a change to regulations, he cannot issue a directive requesting recommendations. Such a request might effectively evade congressional attention, but it also fails to achieve the policy objective.
If the president cannot adjust delegation and discretion to avoid oversight without undermining agencies’ ability to execute his preferences, the alternative is that presidents may choose not to issue directives that they are concerned would be changed by Congress. In this case the concern for the interpretation of the empirical results is that perhaps a large number of directives went un-issued because they would have triggered congressional oversight. Such a dynamic however, bolsters rather than undermines my argument. If the threat of changes to implementation can have a chilling effect on the issuance of executive orders, then that is an additional (if unobservable) channel of congressional influence on unilateral policymaking.
What we are left with is a picture of presidential unilateral policymaking that bears striking similarities to the politics of non-unilateral policymaking. Congress chooses to exercise oversight over agency policymaking when doing so is electorally beneficial. Presidents issue directives in light of both their policy priorities and Congress’s likelihood to constrain such action’s implementation. The strategic environment facing presidents in issuing executive orders is not as insulated from congressional preferences and behavior as previous literature assumes.
6. Conclusion
This paper argues that the administrative politics of unilateral action inform congressional responses to executive orders. When a president issues an executive order, it can be construed as a request to a bureaucratic agency. Because those agencies are subject to oversight not only by the president but also by Congress, the degree of delegation or discretion granted in the order can affect the likelihood of congressional oversight of the implementing agency. The oversight of executive order implementation represents an alternative and under-investigated pathway for congressional influence of bureaucratic policy outcomes. Moreover, the evidence presented here challenges the view that presidents “go it alone” when issuing unilateral directives: not only are such directives best understood as acts of delegation, but their implementation can include a role for congressional input.
One opportunity for further research is to consider additional types of oversight activity, as well as different modes of presidential unilateralism. Agency correspondence logs are a particularly promising source of data on informal congressional oversight (e.g., Lowande, Reference Lowande2018b; Ritchie, Reference Ritchie2023). The challenge with using such logs is linking individual congressional communications to particular executive orders. While existing data sources make this linkage infeasible, targeted FOIA requests could be used to investigate informal oversight in a sample of cases.
Moreover, while executive orders are the most studied mode of unilateral action, recent scholarship has expanded the focus to include presidential memoranda and proclamations (e.g., Rottinghaus and Maier, Reference Rottinghaus and Maier2007; Lowande, Reference Lowande2014, Reference Lowande2024; Lowande and Rogowski, Reference Lowande and Rogowski2021; Kaufman and Rogowski, Reference Kaufman and Rogowski2023; McLain, Reference McLain2024). One emerging explanation for the president’s choice of directive type is that executive orders are relatively more visible, and so he might prefer less visible action when he is more likely to face opposition (Kaufman and Rogowski, Reference Kaufman and Rogowski2023; McLain, Reference McLain2024). However, while the political motivations for the choice of directive type may be strategic, existing scholarship suggests that there is very little functional difference between what different types of directive can accomplish (e.g., Warber, Reference Warber2006; Lowande, Reference Lowande2014). It is functional differences between directive types that concerns us here: it is possible for the president to delegate via memo or proclamation as readily as by executive order. Thus the Delegation and Discretion measures used in this paper are readily applicable to these other types of directive. Indeed, they were developed with a broad variety of directive types in mind (Benn, Reference Benn2023).
In principle, the theory presented here does not depend on the public nature of either hearings or executive orders, and should be relevant to these other types of directive. One plausible mechanism by which delegation can translate to Congressional oversight is through the mechanism of “fire alarms” (McCubbins and Schwartz, Reference McCubbins and Schwartz1984-02). Under this theory, it would not be the fact of delegation itself that stimulated Congressional response, but rather the effect of the directive on agency action. To the extent that unilateral action, of whatever type, influences agency action, once that agency action affects and interest group, Congress will be alerted, irrespective of the original mode of delegation. My expectation would thus be that, even if the extent of delegation varies systematically across directive types, there is no reason for the relationship between delegation and oversight to vary across types, because the mechanism of the relationship is consistent. Future research should test this theory, incorporating different types of unilateralism.
Finally, further work is needed to fully contextualize this role for Congress in today’s politics. How should this broader perspective on responses to unilateral action shape our views of the health of our separation of powers system? One perspective is that, although Congress has an opportunity to shape policymaking outcomes, it is not the full deliberative body that fills this role. Rather, the process articulated here is driven by the particular concerns of individual members’ and committees’ constituencies and interests. This type of particularistic congressional response to executive unilateralism may be under-powered as a tool of asserting legislative prerogative in the face of an expanding executive. However, based on the evidence presented here, this limitation of Congress may be no more pronounced in the context of executive orders than in other policymaking contexts.
This paper shows that the relationship between Congress and presidential unilateralism is more complex than existing literature suggests. Far from “going it alone,” when issuing executive orders the president is relying on cooperation from agency actors. Moreover, those agency actors are subject to influence and pressure from their overseers in Congress. In short, the same political factors that shape policy implementation generally apply to the implementation of executive orders, as well. The politics of presidential unilateralism are the politics of, to borrow from Neustadt, “separate institutions sharing powers” (Neustadt, Reference Neustadt1960 (1990)).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.10060. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/QVX6WJ
Open access
