Executive action that lasts

Abstract

Unilateral presidential action is thought to be limited by the ability of successors to easily reverse past decisions. Yet, most executive actions are never formally revoked. We argue that because of presidents’ unique position as chief executive, some actions create outcomes that make policy reversal more difficult or even infeasible. We develop a novel measure of policies with more immutable consequences and analyze the revocation of executive orders issued between 1937 and 2021. We find the degree of outcome immutability reduces the influence of political conditions on policy revocation. We further examine these dynamics in three cases in which presidents have substantial discretion – diplomacy, non-combatant detention, and police militarization. Scholarship has long highlighted the president’s first-mover status relative to other institutional actors as a key source of their power. Collectively, our argument and evidence demonstrate this applies to their relationship with successors.

Introduction

According to a consensus of pundits and scholars, unilateral actions by chief executives are limited because they can be easily reversed by successors. Presidential directives are revoked, international agreements are broken. This reversibility may enhance a president’s accountability because the public can identify they are responsible, replace them, and then expect swift changes from a successor.

But reversing the outcome of a policy is not always as easy as a signature. President Trump wanted the construction of a barrier along the U.S.-Mexico border. Though the Biden administration issued a proclamation that “paused” construction in 2021, the physical construction of the wall had durable consequences. Miles of barriers built during the Trump administration are still standing. Less obviously, halting construction required canceling contracts and for legal counsel to reverse their positions in numerous eminent domain and environmental review cases brought when the government acquired the land needed for construction. In short, the outcomes this policy produced ensured complete reversal would cost taxpayers – while potentially hindering the president’s ability to prevail in future cases unrelated to the border wall. Knowing that the current president disagreed with the policy of a past president was not enough to explain the durability of this executive action.

We identify a core property of government action that renders it resistant to the whims of future political coalitions: the degree of outcome immutability. We argue public policies whose implementation requires the release, movement, alteration, or designation of government assets for some purpose, are typically more difficult to change. This fact changes the calculations of politicians considering reversal. Scholars have long argued that institutions and policies are dependent on historically contingent paths (e.g., Pierson, 2000; Orren and Skowronek, 2004). Our study affords a unique opportunity to systematically analyze the influence of history on future decision-making (e.g., Liebowitz and Margolis, 1995; Page, 2006). We argue that some policies produce outcomes that are robust to democratic turnover.

We analyze the formal revocation of executive orders issued between 1937 and 2021, which replicates and extends Thrower (2017). We develop and implement a reproducible measurement procedure to identify policies more likely to produce immutable outcomes. Though most executive orders are never formally revoked, we show orders with immediate consequences for the purpose, release, or movement of government assets and personnel are about 9 percentage points less likely to be revoked by future presidents. We also show that turnover in party control of the presidency tends to trigger formal reversal of policies, but primarily for orders that have continued, mutable consequences. Finally, we find that actions with potentially immutable consequences have seen a secular decline since the middle of the 20th century.

To supplement this analysis, we then illustrate how immutable outcomes impacted the policies available to future presidents in three case studies: the detention of non-state combatants, the designation of embassies, and the control of military surplus. These cases appear to be archetypes of executive action. They are publicly salient policies on which presidents enjoy substantial legal discretion. Reversals of unilateral action should be effective. But strikingly, each case demonstrates how subsequent presidents were constrained by the actions of their predecessors.

As Moe (1985) famously writes, “the president is virtually forced to accept the basic institutional framework he inherits from his predecessors” (243). Our study elaborates. Presidents who leverage their executive functions to make policy typically constrain their successors. Paradoxically, this implies that the discretion of today’s president can reduce the discretion of their successor. About 75% of all executive orders have never been revoked, and the number of executive actions on the books has monotonically increased since the middle of the 20th Century. Many more outcomes cannot be undone. Ultimately, we argue, this has important implications for the normative arguments in favor of a unitary executive.

Policy durability

Actions of the state are not permanent, and their durability has dramatic consequences. The longevity of new laws is associated with present and future party coalitions (e.g., Corder, 2004; Maltzman and Shipan, 2008; Berry, Burden and Howell, 2010). Most critically, research demonstrates that absent outright repeal of laws or elimination of agencies, political contests around the implementation of policy can steer laws away from their stated intent (e.g., Patashnik, 2008; Patashnik and Oberlander, 2018). Because these dynamics impact the efficacy of policy, understanding why policy is durable is as important as understanding its initial enactment.

This applies to domestic, as well as international relations, where the durability of state actions is thought to be influenced by similar regime-specific factors. Some scholars posit that democracies make more credible and lasting commitments than autocracies owing to the existence of legislative checks and routine leadership turnover (Gaubatz, 1996). Others hold that these effects are conditional on things like the ideological distance between incoming and outgoing leaders (Gray and Kucik, 2017) or the costs and benefits of agreements for executives (Martin, 2005) and particular domestic constituencies (Dai, 2006). The formality of agreements is also thought to affect whether some state actions are transient or more lasting. According to Lipson (1991), informal agreements are attractive because of their flexibility but are also more easily abandoned. Conversely, states that make formal pledges to abide by certain rules are more likely to comply owing in part to the high reputation costs of abrogation (Downs and Jones, 2002; Keohane, 1984; Poznansky, 2019; Simmons, 2009). The durability of state actions is also determined in part by the extent to which they are self-enforcing. This concept has been used to explain the longevity of the Concert of Europe (Slantchev, 2005) and the liberal international order (Ikenberry, 2001).

The durability of executive policymaking

Though the relative durability of policies enacted by legislatures and executives has not received explicit attention, most commentators and politicians agree that executive action is not built to last. Scholarship on executive agreements has long argued “that treaties are more durable than congressional-executive agreements” (e.g., Nyarko, 2019). The New York Times summarized the conventional view: “While presidents are able to act alone when in line with the Constitution and current law, their actions can be easily undone by the next president” (Parlapiano and Andrews, 2015).

For example, in U.S. foreign aid, presidents have traded restrictions on the support of abortions. The “Mexico City” policy prohibits foreign aid dollars from going to foreign organizations that provide abortions. President Reagan enacted the original policy, President Clinton reversed it, President Bush put it back in place, President Obama again reversed it, President Trump expanded it, and President Biden eliminated it again. ¹ With cases like these in mind, the idea that reversal is easy and limits the long-term implications of executive power is featured in both textbook descriptions of unilateral power (e.g., Pika, Maltese and Rudalevige, 2020) and research (Thrower, 2017, 2018). Warber (2005) summarizes this perspective by writing that “the president’s power to veto or change existing orders is absolute” (53).

This perspective has informed theories of policy reversal. In studies of American political institutions, Maltzman and Shipan (2008) and Berry, Burden and Howell (2010) focus on the enactment and subsequent political context to examine how long policies “survive.” The former also argue that complex laws are more likely to be changed because of their uncertain effects at enactment and that divisive laws increase the likelihood that future majorities will be able to revisit legislation. Adopting this framework, Thrower (2017) provides the most comprehensive account of the reversal of executive action in an analysis of the revocation of executive orders. We build on this work by identifying a characteristic of policy that renders it resistant to reversal. While prior theories mention “transaction costs” (e.g., Berry, Burden and Howell, 2010), these obstacles are set aside or assumed to be exogenous to other components of the process.

But there are examples of durable executive action that suggest these transaction costs are worth investigating and incorporating into theory. In the U.S., President Trump attempted, but failed, to end the Obama administration’s Deferred Action for Childhood Arrivals program. Both President Bush and President Trump publicly opposed the work of the Office of Environmental Justice within the Environmental Protection Agency, first created by President Clinton in 1993. But the order that created the office remains in effect, and the work of the Office continues. Moreover, commentators often suggest that the unilateral enactment of trade barriers by President Trump reorganized global supply chains in ways that may outlive the tariffs themselves (Guilford and Kopf, 2019). In summary, there seem to be aspects of executive action that last – even if successors would prefer to reverse them.

Reversing executive action: theory

Presidents inherit a set of “active” executive orders that can be formally revoked, amended, or replaced. Doing so is not frictionless, in that the decision to address any particular order involves dedicating time and personnel and likely involves consultation with officials in implementing agencies (Rudalevige, 2021). Presidents retain, at least on paper, the power to reverse past policies. The question is what induces them to do so.

To elaborate this problem, we first distinguish between policies and outcomes. In general, policies refer to declarations, agreements, and laws that specify what the government should do in the future. In this way, once specified, they require some future enforcement. Outcomes, on the other hand, refer to what actually happened after enactment. For example, leaders may adopt sentencing guidelines for criminal offenses or the construction of new border walls. The corresponding outcomes are criminals sentenced and barriers built. Adopting a particular policy may make the intended outcome more likely, but issues related to capacity, discretionary enforcement, and other delays may create divergence between what was intended and what actually happened.

Presidents value both. They have preferred policies that map onto preferred outcomes. But they also value policies for the symbolic or performative signal policy sends to an audience about the president’s political brand. For example, the president may issue guidelines for public schools on bathroom usage by transgender students. Whether or not those policies effectuate differences in school bathroom usage, the president sees value in sending a signal to an audience about their position. Most importantly, the decision to revoke an executive order has an immediate effect on the alignment of the policy with their own preferences, and by implication always impacts the political benefits of the president.

However, we argue whether disposing of an existing executive order impacts outcomes is, in part, a function of the degree of “outcome immutability.” Highly immutable outcomes of policy typically involve the release, movement, alteration, or designation of government assets for some specified purpose. The basic managerial powers to employ, purchase, and pay have been used by modern presidents to advance their policy agendas (Gitterman, 2017). Though it is analytically convenient to consider policymaking by legislatures and executives as interchangeable, executive power often means exercising direct control over public personnel and property.

When presidents sign policies with these features, they limit the effect of subsequent policies on outcomes, in two distinct ways. First, subsequent reversals may be less able to undo the outcomes already produced by the active policy. Property given away cannot be easily reclaimed. Facilities sold to the private sector cannot be easily reclaimed. Declassified information cannot be made a government secret again. ² Second, subsequent reversals may have limited effects on future outcomes. A ban on the release of certain types of records classified prior to a particular date, for example, might have little future effects on the declassification because there is a finite number of records within that category. In both cases, the outcome produced by the active policy limits the effects of subsequent policies on achieving successors’ preferred outcomes.

This discussion also helps distinguish the degree of outcome immutability from another well-known political dynamic: policy feedback. Policy feedback occurs when a policy produces outcomes that change the set of politically desirable policies for subsequent policymakers. Put differently, policy feedback captures the idea “that policies are not merely the products of politics but also shape the political landscape by altering state capacities” (Patashnik and Zelizer, 2013). Work on policy feedback effects in the social sciences is too vast to review here (Campbell, 2012). The classic case is the tens of millions who have benefited from Social Security, and how those beneficiaries become politically empowered to prevent reversion. In our view, policy feedback and outcome immutability can occur without the other, and both can occur simultaneously.

Suppose, for example, a president would like to open up a federally controlled area to offshore oil drilling. They sign an executive order that leads to leases and resource extraction. Policy feedback effects occur when the outcomes generate vested interests. Energy companies sink resources into the venture, and political reversal would cost them much more than if they had invested elsewhere. Conversely, the construction of the new oil rigs might generate public backlash among beachgoers and beachfront property owners. These are “standard” cases of policy feedback, and the policy has produced outcomes that change the political context encountered by subsequent presidents.

But this is also a policy that produces more immutable outcomes – in this case, the leasing of public land and the extraction of a finite resource. The effects of any subsequent policies on outcomes, including policy reversal, will be different because of this property of the active policy. Most obviously, the oil will not be returned to the earth, and the effects on the surrounding ocean ecosystem will remain. In this way, it would be difficult or even impossible to reverse the outcomes already obtained. Relatedly, the effects of a halt in future leasing will be limited. The difference between the counter-factual number of leases that would have been signed absent a new ban, and the actual number under the ban, is smaller because of the outcomes produced by the predecessor policy.

The question, then, is how does this property of policy change presidents’ incentives for disposing of the active orders they inherit? First and foremost, because executive orders with more immutable outcomes reduce the effect of subsequent policies on outcomes, we expect these executive orders to be more durable. By durable, we mean that they will remain formally “active” – untouched by successive presidents, regardless of partisan affiliation. These new orders are less likely to reverse prior outcomes or affect future outcomes, so they will be less valuable for successor presidents to dispose of.

The disposition question, however, also modifies Thrower’s (2017) and other similar arguments about the effect of partisan turnover on the durability of policy. When presidents inherit “active” orders, some set of those orders will have been signed by presidents of the opposite party. Naturally, these policies, and the outcomes they produced, are the ones most valuable to revoke. So, in expectation, party turnover in the presidency should lead to revocation. But outcome immutability modifies this calculus. A subset of these opposition-party orders will have produced immutable outcomes, which means revoking the policy itself will be less valuable. This does not mean it has no value – as revoking the policy of the opposite party, without effecting the outcome still may have some consumptive, symbolic benefits. But, again, because these two sources of utility are additive and separable, outcome immutability renders these orders lower priority, all else equal. This means the effect of partisan turnover on revocation will be lower for orders with immutable outcomes.

A similar dynamic follows for the conditions under which orders were originally signed. Thrower (2017) hypothesizes that orders signed under divided government tend to be more ideologically moderate. Put differently, there is another subset of “active” orders that were signed under unified party control of Congress and the presidency, and the president in power is likely to get the policy they desire with less compromise. For this reason, Thrower argued that the executive orders signed under unified government would be more likely to be revoked by successors. Here again, outcome immutability mutes this effect. For orders that influence immutable outcomes, the policy itself may have been more ideologically extreme, but the outcomes are durable, and that impacts the relative value of addressing these policies with a revocation. Thus, we expect this property of policy to reduce the magnitude of the negative relationship between issuance under unified government and order longevity.

Finally, our framework suggests that the prevalence of immutable actions should decline in frequency during the postwar period. We can illustrate the reasons with a relatively straightforward example: public land. The disposition of public land today determines what can be done with it tomorrow. If the government sells it, leases it to surface mining firms, or uses it to test nuclear weapons, that limits the available policy options for future politicians. Moreover, the amount of land in public possession has been declining since 2020. Gradually, the decisions of past political actors limit the choices of the present. The “low hanging fruit” – like landscapes with incontestable natural beauty now protected from development – has been picked. We argue a similar dynamic operates for many actions that involve the disposition of government assets. Barring some exogenous shock like the sudden addition of assets (e.g., the acquisition of new land), we should observe a decline in policymaking of this kind as the options shrink.

This raises an important consideration. The fact that some outcomes are harder to reverse than others is certainly known to presidents, and this likely generates incentives for them to change the kinds of executive orders they sign. This, in turn, may impact the distribution of executive orders available for the next president to overturn. More specifically, there are two pathways in which the degree of immutability could be used strategically, which would change the interpretation of our expectations.

The first possibility is that policymaking with “frictionless” or comparatively mutable outcomes is totally dominated by actions with a higher degree of immutability. That is, presidents see that these actions are harder to reverse and therefore never take actions that are easy to undo. The second possibility is that presidents select issues or goals they believe are the most controversial, fearing they will be the first to be reversed, and render those outcomes more immutable by the way they write their orders.

If the first possibility were true, we would observe mostly orders that generate immutable outcomes. Since we do not, it implies there are some limiting factors that prevent presidents from locking in everything. We think there are two worth mentioning. Most obviously, some areas of policy simply have nothing to do with the release, movement, alteration, or designation of government assets for some specific purpose. If the president wants policies that have an effect on these outcomes, they must dedicate resources to continually enforce them. Importantly, we can look at the distribution of immutable orders across issue areas for evidence of this.

The other limiting factor is Congress. For presidents to use (or use up) these assets, Congress often has to create them. That means some Congress (present or past) has to supply the resources that would allow the president to “lock in” outcomes. Even national security (de)classification, which the president has near-exclusive legal authority over, requires substantial resources to actually carry out. It is well known that preference divergence, as measured by divided government, generally reduces the discretion Congress is willing to provide presidents. That will mean presidents’ ability to make outcomes more immutable will be constrained.

These limiting factors help explain why the second story – that presidents reserve policies with a higher degree of immutability for more controversial or contentious issues – will not change the interpretation of our results. Without any limiting factors, we would expect to observe more immutable orders to be more likely to be formally revoked by successors. This would be a straightforward selection story – these were systematically more politically vulnerable to begin with. Instead, we argue the limiting factors result in a set of inherited directives that is sufficiently diverse in topic, political importance, and degree of immutability in outcomes to uncover the relationships we expect.

Implications

1. (1) Policies with immutable consequences will be less likely to be revoked.

2. (2) Partisan turnover in the presidency will increase the likelihood a policy is revoked. This effect will be lower in magnitude for policies with immutable consequences, relative to others.

3. (3) Issuance under unified government will increase the likelihood a policy is revoked. This effect will be lower in magnitude for policies with immutable consequences, relative to others.

4. (4) The frequency of policies with immutable consequences will decline over time.

Research design

A significant upshot of the foregoing is that any attempt to estimate the durability of executive action needs to account for outcome immutability. The longevity of executive orders, for example, is not simply a function of current and future political contexts, but also the type policy outlined by that order. To illustrate these dynamics, we rely on evidence from both systematic patterns in the population of “active” executive orders from 1937 to 2020, as well as case studies. Analyzing the population of executive orders allows us to test broader generalizations about the importance of the immutability of outcomes for predicting the revocation of policies.

In contrast, case studies allow us to demonstrate the existence of the mechanisms we describe while characterizing the impact of past policies on both outcomes and future policy options. For this reason, we selected cases in which this was most likely to occur, but that the standard view predicts would be fleeting or unstable. It would be trivial to show, for example, that immutability was at work in military conflict or covert action. Instead, our cases focus on areas of policy in which chief executives should have high discretion, but that the actions of a predecessor constrained the efficacy of a successor’s policy options as a result of immutability.

Assessing patterns of executive order durability

This section analyzes formal revocation patterns in the nearly 6,500 executive orders issued between 1937 and January 18, 2021. ³ Since our objective is to build on and render our framework comparable to existing theoretical perspectives, we first replicated and extended Thrower (2017). Thrower examines factors associated with the formal longevity of executive orders from 1937 to 2013 and finds strong support for predictions based on spatial policy disagreement. We extended the dataset to the present, updated the original data with the most up-to-date records, and replicated these main findings to ensure that our analysis was comparable (see SI pg. 1–5). We also adopted a different approach to coding the revoking and issuing presidents, which we detail in the SI.

Measuring immutability

Our key measurement task is to identify executive orders that enact policies which lead to more immutable outcomes. We regard the “mutability” of an outcome prescribed by some order as a continuous concept, but adopt a dichotomous classification procedure because we have found, in practice, that it is very difficult to be more precise using only the documents themselves. Immutable outcomes are those where reversal is impractical or the costs of reversal are markedly higher relative to the status quo ante. Specifically, we code an executive action as involving immutable outcomes if any one of the following conditions are met:

1. 1. …release of non-recoverable assets or personnel. This includes, for example, the sale of land, facilities, physical property, or the release of classified information, or the extension of land trusts for Native Americans. It also includes the temporary closing or opening of government offices (e.g., for holidays or during government shutdowns), though we code these as ceremonial, and they are not included in the main analysis.

2. 2. … physical movement or alteration of property. This includes, for example, military construction projects, relocating bureaus, or the disposal of property.

3. 3. …designation of assets or personnel for some specific purpose. This includes, for example, deploying military personnel, or designating public land for various purposes (e.g., coal production, wildlife protection, or military target practice). This also includes the enactment of pay raises or exemptions from retirement.

Note, it is important to distinguish between the “likelihood” of reversal or the size of the costs and the question of immutability. If the function of the order is to do one of the above, we argue it will be difficult to reverse for non-political reasons. Put differently, we code orders based on their function, not some continuous scale of the probability the outcome it produces is immutable.

For example, Executive Order 8578 sets aside land near Nellis AFB in Nevada for target practice. It satisfies our definition above because it designates some land for a specific purpose. The fact that it can (and was eventually) repurposed is immaterial to the judgment about whether it qualifies as a case of an immutable executive action. (It is worth noting that bombing that land made it good for few other uses. Eventually, it was given to the Atomic Energy Agency for testing nuclear weapons, and today is the proposed resting place for the United States’ nuclear waste.)

Many orders do not include a specific outcome – that is, they create some kind of commission or ask for recommendations. These actions may themselves lead to some outcome that is difficult to reverse, but because the order itself is removed from that outcome, and is not the proximate or efficient cause of it, we do not code these actions as immutable. For example, Executive Order 11237 prescribes regulations for land acquisitions by two departments. These regulations themselves might lead to land acquisitions that are difficult to reverse, but because the order itself does not perform the function of acquiring that land, it is not a case of an immutable executive action. Likewise, Executive Order 12803 describes procedures for privatizing federal infrastructure, but does not order specific roads or bridges to be sold.

Overall, this measurement strategy is an indirect one: we cannot actually observe a policy’s effects on outcomes. Instead, we focus on classifying actions that perform the functions described by our definition and thus, assume they were carried out. This approach is subject to error, which, in our view, partly justifies the complementarity of case studies. Though we describe our coding procedure and report additional diagnostic information in the SI, we found this task could be performed consistently across raters (89% inter-rater agreement, κ = 0.76). We were also cognizant of the potential for bias introduced by researcher judgment. Thus, we implemented our coding procedure through a combination of hand-coding and string-matching. This means 63% of orders were coded without being read by a human. The remaining were coded with the researchers blind to the date and president who issued the order.

Executive orders that perform these functions implicate a wide variety of governance tasks. They create executive positions, offices, and departments. They release government reserves of food and oil. They open public land for leasing and declassify government secrets. They authorize cabinet secretaries to confiscate private property. They authorize the sale of military surplus and deploy military personnel to control domestic civil unrest. As we show in Table C.2, these orders are concentrated in the areas of transportation, public lands, civil rights, government operations, and international affairs. Nonetheless, most topics have a substantively significant number of orders coded as immutable, and there are only two topics – education and social welfare – with none.

Importantly, this variation suggests, consistent with our previous argument, there are some limiting factors on the strategic use of immutability in outcomes. It is no accident that the highest rate is in Transportation and Public Lands, whereas the lowest are in Education and Social Welfare. Public lands can be disposed of and labor disputes can be resolved via executive order. Over most of the time series in question, presidents have almost no property or assets that can be manipulated to impact education or welfare policy. It would be categorically infeasible to write policies that produce outcomes that are more immutable.

Overall, we found that about 41% of executive orders described a policy that fell into one of those categories. We take this as initial descriptive evidence that the properties of policymaking we consider in this study are an important category of the total policymaking output of presidents. ⁴ We also find that orders appear less likely (-19 pp.) to be coded as producing immutable outcomes if issued under divided government. This is additional evidence of the limiting factors on these kinds of orders, in that presidents under divided government may be given fewer concrete assets to dispose of by opposing Congresses.

Another way to contextualize its importance is to compare the latent significance of orders categorized as producing immutable outcomes to those that do not. Chiou and Rothenberg (2014) developed a measure of latent significance for orders issued between 1947 and 2002. For the 3,512 orders rated by Chiou and Rothenberg in our dataset, immutable orders tended to be more significant (Figure C.1). Though it is possible the association has changed in the previous two decades, this might allay the concern that durable orders are simply routine, ceremonial, or administrative orders that pose no political controversy.

Findings

The data are consistent with the idea that outcome immutability impacts the formal reversal of policy. Specifically, we find that executive orders that directly implicate the immutability mechanism are substantially more durable, that their revocations are poorly explained by political context, and that they have seen a long-term, secular decline since the middle of the 20th century.

We first turn to the straightforward comparison of revocation across types of orders. Following others, we limit our main analyses to “non-ceremonial” orders, which excludes policies like the creation of honorifics (i.e., service awards), the closing of offices on holidays, and the renaming of federal facilities. ⁵ The vast majority (74%) of executive actions are never revoked. In fact, setting aside a mass revocation of orders from 1986, the number of orders active (on paper) has monotonically increased since the beginning of the time series. As expected, however, orders that do and do not implicate immutable outcomes have significantly different revocation rates, with 30% of mutable actions revoked, while only 21% of immutable actions are revoked. This gap is statistically distinguishable from zero by convention (p < 0.001) and consistent across both ceremonial and non-ceremonial orders. As we go on to show in our revocation analysis, immutability is associated with lower revocation rates even after controlling for political context and other time-invariant features of executive orders, such as whether they appeared in the news, were based on statutory authority, their word count, or their policy topic.

We next turn to analyze the relative importance of political context for revocation. Though our results are robust to alternative approaches, our preferred strategy is to model the probability that an executive order is revoked in the following presidential term. This simplifies the analysis considerably by reducing the dimensionality of the data. That is, instead of an unbalanced panel dataset and survival model with complex inter-dependencies, we examine the single cross-section of non-ceremonial executive orders issued between 1937 and 2020. We describe additional details in the SI, but this allows us to examine the impact of the key explanatory variables of interest. However, it narrows our substantive focus to what happens to orders in the short term (or about 1–7 years, depending on what point in the initial term the order was issued). This also helps reduce the influence of “housekeeping” orders that occur decades latter, include mass revocations, and do not represent the kind of politically-motivated reversals of interest. In addition, as we show in the SI, approaches that leverage survival models are highly sensitive to model specification. In particular, the inclusion or exclusion of intercept shifts leads to different estimates and conclusions, which we believe may be due to dependence between observations (e.g., orders revoke other orders).

Consistent with expectations, the marginal effects plotted in Figure 1 suggest that orders with immutable consequences are associated with political context in ways that differ from other orders. We first estimate a logistic regression predicting order revocation in the following full presidential term. As regressors, we include a multiplicative interaction (and its associated constituent terms) between the four dichotomous states of interest: immutability, issued under divided government, next president from opposing party, and next period under divided government. Substantively, this means the model estimates an intercept shift for each unique combination of the relevant political conditions and order type, and marginal effects average over the relevant states. The main results also include controls considered important by past work: foreign policy order, issued during wartime, inflation, election year, administration change, end of term, and a yearly time trend (see Table C.1). ⁶ We report simulated marginal changes in probability of being revoked in the following full presidential term.

Figure 1. Policy revocation and outcome immutability. Reports marginal change in the probability of revocation and 95% confidence intervals, for non-ceremonial executive orders issued between 1937 and 2020. Estimates are simulated from logistic regressions using an observed case approach (Hanmer and Ozan Kalkan, 2013). Linear models include a multiplicative interaction between immutability, opposing president next, divided government next, and issued under divided government, along with a vector of controls: foreign policy, war, inflation, election year, administration change, end of term, and time trend. Complete regression results are reported in Table C.1.

First, Figure 1 confirms the simple difference described earlier: orders with immutable consequences are about 6.6 pp. less likely to be revoked. Second, the full sample of orders exhibits the substantive patterns consistent with existing theory: revocation is more likely when the next president is opposing (about 4.5 pp.) and when the order was issued under unified government (about 3 pp.). The baseline probability of being revoked in the following term is 11%, so these effects are large in scale. But Figure 1 suggests they are largely driven by orders with more mutable consequences. More specifically, for orders with immutable outcomes, the effects of changes in partisan control of the presidency or issuance under divided government cannot be signed. It is important to note, however, that these differences are only conventionally significant for issuance under divided government – but not party turnover in the presidency. For the latter, the difference across immutable and mutable policies is more uncertain, even if the coefficients are in the expected direction. This is consistent with the idea that the symbolic benefits of revoking policy may be stronger in the case of party turnover. These results are robust to alternative model specifications, the inclusion of ceremonial orders, and the exclusion of orders issued by FDR, which we report on SI-12-13.

Finally, we examine secular trends in immutability by examining the proportion of executive orders in a given year with such consequences. As Figure 2 illustrates, the functional consequences of executive action have fundamentally changed since the late 1930s. Until the early 1950s, a majority of executive orders had direct and immediate consequences for government assets and personnel. In the contemporary period since the Reagan administration, the figure is about one in five. Executive orders in the contemporary period function in ways that are similar to the stylized account provided by Thrower (2017). They tend to initiate policymaking processes or change rules associated with the management of the executive branch, they can be (and are often) easily revoked.

Figure 2. Immutability over time. Reports the proportion of actions coded as immutable, along with a LOESS line with 95% confidence intervals, and includes both ceremonial and non-ceremonial orders.

In summary, we find strong evidence consistent with three of our implications (1, 3, and 4), and weaker evidence for a fourth (2). However, this approach only examined formal revocation of policies, which was a necessarily indirect way of understanding the relationship between policy durability and outcomes. Thus, we next leverage three case studies to illustrate the interplay between policies, outcomes, and revocation.

Case studies

The statistical analysis of formal revocation cannot offer evidence of how immutability works in practice or demonstrate the impact of these policies for outcomes. Below are several case studies that illustrate these lasting effects. For each case, we focus on the president’s chief executive functions that effectively raised the cost of reversal for successors. This does not mean mutability was the only mechanism thwarting reversal. Notably, each case also presents the opportunity to understand this mechanism alongside policy feedback. Due to space constraints, we present this additional case study material in the SI that describes this mechanism (SI–16-21). These case studies also have an important feature. Specifically, in each instance the president in question had different preferences from their predecessor and actively tried, but failed, to reverse an executive order and its associated outcomes.

Military surplus

To illustrate the dynamics of outcome immutability, we first turn to an area over which the state – and by extension, the president – maintains near monopoly power: the possession of military weaponry. This issue has generated significant political controversy, along with corresponding incentives for new presidents to reverse the actions of their predecessors. Since 1997, the federal government has distributed surplus military weapons, equipment, and supplies to local law enforcement agencies through the 1033 program (10 U.S.C. § 2576a). To date, the total value of transferred property has exceeded $7.4 billion, and secular increases in military spending led to the availability of more controversial equipment, such as armored vehicles, assault rifles, and night vision optics (Coyne and Hall, 2018). By 2014, the program attracted significant media attention after police in Ferguson, MO were photographed with this equipment. Controversy over the program surfaced again after the Summer protests subsequent to the murder of George Floyd, when news outlets documented police using armored vehicles for crowd control (Lehren et al., 2020). The most controversial property is still owned by the federal government, and the program itself is left to the discretion of the Secretary of Defense, who is appointed by and serves at the pleasure of the sitting president.

In 2015, President Obama signed Executive Order 13688, which recalled tracked armored vehicles, certain kinds of camouflage clothing, bayonets, and grenade launchers – while also prohibiting the distribution of .50 caliber weapons (Working Group, 2015). This recall effort, though small in scope, was largely effective – all agencies who had the impacted equipment had removed them from their inventory within 6 months (Lowande, 2021). When the Trump administration took over the program in 2017, they revoked E.O. 13688 in its entirety (Trump, 2017). At first glance, this case seems to be explained by the standard spatial framework for policy reversal. President Obama acted unilaterally in 2015, shifting the status quo left. President Trump revoked that order, restoring a more conservative policy. Executive action was as durable as the political coalition that enabled it.

But while the Trump administration revoked E.0. 13688 and all its associated policies, their reversal had no practical impact on the distribution of military surplus. ⁷ There was no distribution of .50 caliber weapons, which were now technically available. There was no general increase in the weapons and equipment in circulation. In summary, while Obama’s policy is technically no longer the law, it had a lasting (and in some cases, irreversible) impact on outcomes. This meant that the Trump administration’s policy choice was meaningless in terms of its effect.

Immutability helped make the Obama action durable. Put simply, because of the nature of the policy change ordered, some outcomes could not be reversed. Participating agencies were required to return the recalled equipment to the Defense Department. The department then had to decide what to do with property that could no longer be used by local law enforcement, but was also unusable by the military. Storage is costly and carries some risks, so in most instances, they chose “demilitarization,” which requires the mutilation of weapons and equipment to the point of scrap metal. In the state of Michigan, for example, the aged, tracked armored vehicles were taken to a national guard facility and blown up during target practice (Beggin, 2020). There is also a finite amount of military surplus available for distribution, and the destruction of that equipment meant additional property distributed would differ in kind.

In this way, the Obama action fundamentally changed the policy options available to the next president. It was either impossible or otherwise impractical to revert to the status quo prior to his action. The Trump administration’s revocation of the Obama order in 2017, of course, totally ignores this while retaining the consumptive political value that reversal demonstrated. The formal policy was reversed by President Trump for its symbolic importance.

Detention of non-state combatants

Soon after the September 11, 2001, attacks, the Bush administration was confronted with a decision about what to do with actual and suspected terrorists apprehended in the course of military operations in Afghanistan and elsewhere in the region. They soon settled on Guantanamo Bay, the U.S. military base that has been under American control since the end of the Spanish American War. The decision to house detainees from the war on terror there was unilateral (Cutler, 2010, 65). One senior administration official purportedly likened it to the “legal equivalent to outer space” (Greenberg, 2009, 7). According to the Guantanamo Review Task Force’s Final Report, “[b]y the end of 2002, 632 detainees had been brought to Guantanamo,” a number which grew to roughly 779 over the next few years. ⁸

When Barack Obama took office in January 2009, roughly 240 detainees remained in the Guantanamo Bay Naval Base (Gitmo) (Bruck, 2016). On the second day of his presidency, he issued Executive Order 13492 which called for the closure of the prison “as soon as practicable, and no later than 1 year from the date of this order.” ⁹ Obama’s decision to order the closure of the prison was predicated on his belief that it was immoral and damaging to American foreign policy. Nonetheless, he was never able to close it. By 2013, the special envoy tasked with facilitating closure, Daniel Fried, was reassigned to a different position, and his former position was left vacant (Savage, 2013). As one senior official explained: “The president can’t just wave a magic wand to say that Gitmo will be closed” (Savage, 2010). In this case, there were substantial policy feedback effects that hampered the effectiveness of the president’s order, which we detail in the SI. However, our framework helps explain what an account based solely on policy feedback misses.

Put simply, the closure of Guantanamo was hampered by the immutability of outcomes. The Bush administration’s rationale for using Gitmo was the belief that it held a special legal status. Bruck (2016) has written that, because “there was little expectation that [prisoners] would be criminally prosecuted, […] scant attention was given to the kind of evidence-gathering that would be required in court.” This had several downstream effects. First, the administration made little effort early on to determine which detainees were genuinely innocent and which were not, a practice which is usually done in conflict situations per Article V of the Geneva Conventions (Bruck, 2016). The continued imprisonment of individuals who may have been innocent also likely accounts for the subsequent radicalization of some of the detainees, an issue which Congress and even some senior Obama officials later pointed to as an impediment to their transfer. ¹⁰

The use of enhanced interrogation techniques at Guantanamo also created its own set of issues when it later came to determining the status of prisoners. As Garcia et al. (2013, 33) put it, “The use of any such evidence, or evidence derived from it, in the criminal trial of a detainee would likely be subject to legal challenge under the Fifth Amendment on the ground that the statement was gained through undue coercion.” Despite issuing an Executive Order in 2011 calling for a review of all remaining prisoners designated for indefinite detention, no hearings transpired in the ensuing two years: “According to two government officials, the delay was due to concerns at the C.I.A. that the review process would prohibit evidence elicited by torture – thus conceding that torture had taken place” (Bruck, 2016).

The issue of coercive interrogation under the Bush administration interfered with the Obama administration’s ability later on to transfer detainees to other countries. Obama’s Task Force noted that transferring detainees to places other than the U.S. was plagued by “political and other issues in potential resettlement countries (e.g., public perceptions of current and past U.S. detention policies), third-country views (and sometimes pressure) with respect to detainee resettlement, and public views of the Guantanamo detention facility generally.” ¹¹ The consequences of these immutable outcomes are best summarized by President Obama’s initial speech on the topic in 2009: “we don’t have the luxury of starting from scratch. We’re cleaning up something that is, quite simply, a mess… not caused by my decision to close the facility; the problem exists because of the decision to open Guantanamo in the first place.” ¹²

In summary, it was not simply political disagreements or policy feedback that limited the effectiveness of President Obama’s initiative. The Bush administration’s Gitmo policies produced outcomes that could not be reversed and limited the effects of the subsequent closure order on future outcomes. This rendered ordering the “closure” of Gitmo mostly symbolically beneficial. Likewise, when President Trump reversed the closure with a subsequent executive order, more detainees were never sent to the prison, and his “expansion” was only beneficial in terms of signaling his commitment to a policy direction.

Location of embassies

On December 6, 2017, President Donald Trump officially recognized Jerusalem as the capital of Israel. In May of the following year, the U.S. embassy moved from Tel Aviv to Jerusalem, coinciding with the seventieth anniversary of Israel’s founding. Prime Minister Benjamin Netanyahu lauded the move while Palestinians and many Arab states in the region criticized it heavily. ¹³

Trump’s decision fulfilled a longstanding Congressional wish dating back to 1995 when the Jerusalem Embassy Act was passed. The text of the legislation mandated that the president must move the U.S. embassy to Jerusalem and threatened to severely curtail funds to the State Department’s “Acquisition and Maintenance of Buildings Abroad” if they failed to do so. There was, however, a significant caveat. Presidents could issue a waiver postponing any such decision for six months if they deemed “that such suspension is necessary to protect the national security interests of the United States.” ¹⁴ This provision effectively gave the president unilateral discretion over the issue. ¹⁵ Until 2018, all U.S. presidents from Clinton to Bush and Obama had consistently issued the national security waiver every six months (Anderson and Schwartz, 2017). Both Clinton and Bush promised to move the embassy during their presidential campaigns but later backtracked once in office (Baker, 2016) – desiring to retain America’s status as a neutral arbiter in the negotiations between Israel and Palestine (Lowry, 2018). Unlike the previous two cases, both of which involved a given president unsuccessfully attempting to reverse a unilateral policy of their predecessor, the available evidence suggests that Donald Trump’s successors will not attempt to move the embassy back. Then, candidate Joe Biden, the presumptive nominee, put it bluntly: “I wouldn’t reverse it. I wouldn’t have done it in the first place” (Bard, 2019). The reasons are consistent with our framework. This episode also has useful research design properties in that it provides evidence of our mechanism in a case of a proverbial dog that will (likely) not bark. In other words, while there is unlikely to be an attempt at reversal, the realities of outcome immutability help explain why this is.

First, as U.S. Ambassador to Israel David Friedman put it, no future American president was likely to move the embassy back in part because doing so would require “conclud[ing] that Jerusalem is not the capital of Israel and Tel Aviv is. I think,” he said, “that would be a far more controversial thing to do than what the president did.” For Trump’s successor to reverse the decision, he or she would now have to essentially embrace the former position – that Tel Aviv is the actual capital of Israel. Put differently, the outcome the Trump policy achieved changed the set of feasible policies and their impacts on subsequent outcomes. Second, by placing the U.S. embassy in Jerusalem, Trump eliminated a key bargaining chip that would have otherwise been available during negotiations between Israel and Palestine related to a potential two-state solution. Writing about the prospects for a peace plan, Brian Katulis and Daniel Benaim write that “unilaterally moving the U.S. Embassy in Jerusalem […] without asking for any constructive steps from Israel in response” has reduced the likelihood of such a settlement (Katulis and Benaim, 2018). Thus, even if they thought the move unwise, which evidence suggests they did, there is little they can do. Even if moving it back temporarily would please the Palestinians, it could further alienate the Israelis and, as one senior Democrat put it, would therefore be counterproductive (Wilson and Lillis, 2019).

Finally, moving the embassy involved the sale of government assets like the residency of the U.S. ambassador in Tel Aviv, which would make reversal more difficult (Krauss, 2020). In this instance, the residence was sold to Sheldon Adelson, the largest single donor to President Trump’s re-election campaign. As we illustrate in the SI, like Gitmo, this case also exhibits evidence of strong policy feedback effects. The movement of the embassy mobilized political support that would generate significant political opposition to reversal. However, the immutability of outcomes – namely, the physical movement of the embassy and the asymmetric nature of the policy options available after the fact – helped tie subsequent presidents’ hands.

Discussion

Typically, scholarship investigates the power of chief executives relative to the power of other political actors. In the case of presidential unilateralism, the comparative weakness of other institutions has been highlighted as a source of aggrandizement. An important caveat to this idea is the ability of new presidents to overturn the actions of their predecessors. If voters change their minds about partisan control of the presidency, the new president should be capable of reflecting the change in citizen preferences. Our study highlights the important ways in which presidents’ first-mover advantage extends to their relationship with successors. By leveraging power inherent to the chief executive, they are often able to make future changes in policy limited in their effects on outcomes.

More broadly, this suggests the public accountability relationship implied by some versions of presidential representation is more problematic than originally thought. Policymaking actions can have consequences for outcomes that build in future resistance to change or otherwise cannot be undone. This is a point that is universal to any policymaking, not just the president. However, unilateral action itself offers a useful case to examine the basic consequences of this idea. The functions performed by executive orders and other directives vary considerably. Some have direct and immediate consequences for market prices, government assets, or the employment of individuals. Others are more akin to policy planning documents, or change the management of the executive branch in ways that require continual effort to be effectual. We have shown that these factors matter for the durability of policy by marshaling evidence both qualitative and quantitative. Specifically, we have shown that the immutability of some outcomes can limit the possibility of reversal, and help improve models of executive order longevity. We have also shown how these dynamics effectively constrain future decisions in three cases.

These points have implications for the core of normative arguments in favor of a unitary executive. Kagan (2001) famously argued that “presidential administration” enhanced the accountability of the administrative state by tying the actions of the government to elections. Other practitioners and legal theorists like Calabresi (1995) and Yoo and Gaziano (2018) agree a unitary executive is more accountable because of this property. Our empirical evidence suggests this core tenet of presidential representation overlooks the fact that there are significant, non-political barriers to the reversal of policy over time.

Our study leaves open several promising areas for development. First, while our analysis centered on the American presidency, immutability of outcomes clearly implicates policymaking in other institutional settings. Our description of the implications of certain kinds of policymaking for outcomes is general to policymaking by governments, not just the American presidency. All of the functions performed by U.S. presidents can plausibly be performed by chief executives in general, as well as by legislatures or judicial systems. In addition, our classification procedure necessarily coarsened what is surely a continuous latent concept, leaving room for advances in measurement. Another promising future direction, in our view, lay in investigating the degree to which these features of policy are a function of strategic choices. The basic question that we have not considered here is: if political actors care about their policy legacies, or want to lock in some outcome, when will the political environment permit this? Our study has provided a theoretical framework for describing how they would do so and demonstrating that it is consistent with available evidence. What we have not considered is what circumstances incentivize presidents to look for more or less durable policy.