Overview

The model is a strategic game of presidential policymaking that is played between a President and an Appointee. The President’s formal authority in the game is limited to two specific rights: the right to issue a directive to the Appointee and the right to remove (and replace) the Appointee after observing whether the directive was implemented. This setup allows us to focus on the question of when a president’s formal authority to exercise these two rights will actually yield real authority over the direction of policy change. That is, the model specifies the President’s formal authority exogenously, and allows any real authority to emerge endogenously from the strategies taken by each player. ^{Footnote 6}

In the first period, the President decides whether to issue a directive to the Appointee and, if doing so, the Appointee decides whether to implement the directive (the game ends if the directive is not issued). After observing the Appointee’s decision to implement the directive, the President then decides whether to replace the Appointee at a cost k, which reflects the transaction costs associated with removal and replacement. In the second period, the Appointee again decides whether to implement the directive (either the incumbent Appointee or the replacement). The President’s payoff depends on how many times the directive is implemented. ^{Footnote 7} This setup reflects the implementation of a procedural directive, where appointees must continually make choices about whether to comply with the directive (Mayer Reference Mayer2002). ^{Footnote 8}

I assume that the Appointee is career motivated in that the Appointee values implementing the directive, but receives no added payoff from simply holding office or having a replacement implement the directive. These assumptions broadly reflect a situation where an appointee’s opportunity cost of holding office, and thus forgoing employment elsewhere, only yields a net benefit if the appointee personally accomplishes something, that is, if an appointee implements the president’s directive. ^{Footnote 9}

The directive can vary in terms of: (i) the benefit it supplies to the players when implemented; (ii) how costly it is to implement and (iii) how costly it would be to the President if it were issued but not implemented. These three facets of a directive are assumed to be correlated in that they all depend, to varying degrees, on the “stakes” of the directive, b. What these stakes represent can have different interpretations, such as the salience or scope of a directive, or how politically ambitious it is. ^{Footnote 10} The model simply assumes that as b increases, the three facets listed above also increase. Furthermore, the stakes of the directive are assumed to be exogenous, which captures the idea that a president is somewhat limited by the supply of potential directives. (Naturally, a president would prefer a tradeoff-free directive with high benefits and low costs, but the model requires that increasing the benefit imposes at least some increase on the cost of implementation and the cost of failure.)

The degree to which each player is affected by the stakes of the directive does, however, depend on each player’s ability. For the Appointee, $\phi $ is a policy competence parameter that captures the Appointee’s ability to implement the directive, which could reflect prior work experience, education, and related factors. For the President, $\sigma $ is a political competence parameter that captures, among other factors, a president’s stock of political capital, such as the ability to minimise any political fallout from a directive that is issued but not implemented (Christenson and Kriner Reference Christenson and Kriner2014).

The President’s agency problem is driven by two sources of uncertainty. The first is uncertainty over whether the Appointee is loyal, that is, whether the Appointee shares the President’s policy goals and thus will benefit from implementing the directive (only a loyal Appointee benefits from implementing the directive). Secondly, the President is uncertain about whether the policymaking environment, or state, in which the Appointee works is loyal and, thus, whether the Appointee can rely on the relevant office, bureau, or commission to assist in implementing the President’s directive. I assume that the Appointee does not have to pay to implement the directive when the state is loyal.

I parameterise the President’s uncertainty about the Appointee using $\alpha$ , which is the prior probability that the Appointee is loyal. This probability applies both to the incumbent Appointee and to any replacement. Given that presidents are generally able to select appointees who share the same party affiliation, it seems reasonable to assume that $\alpha$ is large, as I do in the analysis below, but still less than 1 to keep open the possibility of an adverse selection problem.

I parameterise the President’s uncertainty about the state using $\beta$ , which is the prior probability that the state is loyal in any period of the game. If the state is loyal, the Appointee can offload the cost of implementation, which reflects a situation where career bureaucrats assist in implementing the directive. But when the state is disloyal the Appointee must work alone and pay the full cost of implementation. Thus, setting $\beta$ to a higher value reflects a more reliable policymaking environment for the Appointee. Although a president obviously prefers a higher value of $\beta$ , the uncertainty generated by this parameter reflects the fact that a president has only so much control over the bureaucracy, which is largely inherited from the president’s predecessors. ^{Footnote 11} And unlike the Appointee’s loyalty, the President cannot influence the state through replacement. Nonetheless, given the relative professionalisation of the federal bureaucracy, it is reasonable to assume that $\beta$ is large, as I do in the analysis below, although not as large as $\alpha$ given that the bureaucracy is still inherited, not chosen.

The President’s uncertainty about the Appointee and the state sets up a classic agency problem. Should the President remove the Appointee after observing a bad outcome, that is, a directive that is not implemented, or should the President give the Appointee the benefit of the doubt and assume that the bad outcome was the result of a disloyal state? If the President is willing to remove the Appointee after a bad outcome, the Appointee will internalise this when deciding whether to implement the directive. But if the President is not willing to remove the Appointee after a bad outcome, even a loyal Appointee might take advantage of the President and not implement the directive when the state is unfavourable. For a disloyal Appointee, the situation is more straightforward: The Appointee does not benefit from being in office and, thus, is unresponsive to the President’s removal threat and never implements the directive.

As in other political agency models, the President’s control over policymaking works through two mechanisms. One is through replacement, where the Appointee can be removed after failing to implement the directive. The other works through coercion, or changing the behaviour of Appointee through the credible threat of removal, whereby the Appointee is willing to implement the directive even if it brings negative utility in the first period because the Appointee wants to remain in office in the second period. ^{Footnote 12} (As an empirical matter, it is noteworthy that if accountability works mostly by changing the behaviour of appointees, we may never observe the removal of appointees.)

Given the President’s agency problem, some directives will be easier to enforce than others. This is most apparent in the case where the Appointee is loyal, whereby remaining in office for a second period is desirable, all else equal, but the state in the first period is disloyal so the Appointee bears the cost of implementing the directive. In this case, the model suggests three types of directives: (i) a directive is self-enforcing when the Appointee will implement it regardless of the removal threat, which echoes Neustadt’s description of a “self-executing” directive; (ii) a directive is unenforceable if the Appointee will never implement it and (iii) a directive is enforceable if the Appointee will only implement the directive when the removal threat is credible.

What is noteworthy about this taxonomy of directives is that only the enforceable directive is suggestive of presidential power, at least coercive power. For the other directives, the President’s influence over policymaking, if any, can work only through the mechanism of replacement.

So far I have described a removal rights model, where the President has the right to remove the Appointee. To establish a benchmark set of predictions, which will be useful in the next section, I also developed a benchmark model, which is identical to the model just described except that the President lacks removal rights and thus cannot remove the Appointee. In terms of the taxonomy of directives, the benchmark model has only two types of directives, namely those that are self-enforcing and those that are unenforceable (the President has no coercive power, so there are no enforceable directives). Because the two models yield a number of different predictions, they can help us to identify the impact that removal rights have on the strategic choices made by each player, including whether the President will issue a directive and whether the Appointee will implement it.

Preview of results

The issuance decision

When will the President issue a directive? In general, the President wants to issue a directive when: (i) the probability that the Appointee is loyal, $\alpha$ , is relatively high; (ii) the probability that the state is loyal, $\beta$ , is relatively high and (iii) the directive is either self-enforcing or, if the President has removal rights, the directive is enforceable and the President’s removal threat is credible.

To illustrate how these factors shape the President’s decision to issue a directive, Figure 1 plots the President’s expected utility from issuing a directive as a function of the stakes of the directive, b. The thick lines show regions where the President’s utility is identical regardless of whether the President has removal rights, and the thin lines show regions where the President’s utility differs. The figure also identifies the regions where the directive is self-enforcing, enforceable or unenforceable, which are divided by the cutpoints ${\underline {b}}$ and $\bar b$ . What is immediately clear is that the region in which a directive falls can have a dramatic impact on the President’s utility. The President’s utility is generally lowest from issuing an unenforceable directive and it is generally highest from issuing a self-enforcing directive. And whether the President gains from issuing an enforceable directive depends on whether the removal threat is credible. ^{Footnote 13}

Figure 1. The President’s issuance decision.

The figure identifies a removal threshold, $\hat b} $ , which indicates the point at which the President’s removal threat becomes credible (when $b \gt \hat b} $ ). When the removal threshold is surpassed, the President’s right to remove the Appointee transforms into actual coercive power over the Appointee, although this power persists only up to $\bar b$ , at which point even a loyal Appointee will not implement the President’s directive (for $b \gt \bar b$ the directive is unenforceable). In the figure, the removal threshold sits midway through the enforceable region, which is a reminder that the President’s removal right is a blunt instrument. For the threat of removal to be credible, the stakes of a directive must be sufficiently high.

The removal threshold also marks the point at which the President’s utility from issuing a directive starts to depend on whether the President has removal rights. As the stakes of the directive increase and the removal threshold is surpassed, a President with removal rights will continue to see utility gains from issuing higher-stakes directives. Yet, for a President that lacks removal rights, the same higher-stakes directives will bring about a decrease in utility. I state the implications of these patterns as a hypothesis.

Hypothesis 1 (H1) A president will only issue a relatively high-stakes directive when the president has removal rights, although the stakes of the directive must still be low enough such that the directive is enforceable and, thus, can be implemented by the president’s appointees if necessary.

The implementation decision

If the President issues a directive, under what conditions will the Appointee implement it? To characterise the Appointee’s implementation decision, I focus on the compliance rate, which reflects the probability of implementation given the uncertainty surrounding both the loyalty of the Appointee and the loyalty of the state.

To compare the predictions of the two models, Figure 2 plots a directive’s expected compliance rate, with the benchmark model situated on the left panel and the removal rights model situated on the right panel. In each case, the compliance rate is given as a function $\beta$ , the probability that the state is loyal (this is labeled as “Agency Loyalty” in the figure).

Figure 2. Compliance rates.

When the President lacks removal rights, as the President does in the benchmark model, the compliance rate is only a function of the uncertainty parameters, $\alpha$ and $\beta$ . This is reflected in the left panel where the compliance rate increases linearly in $\beta$ for a fixed value of $\alpha$ . In contrast, when the President has removal rights and the removal threshold has been exceeded, the compliance rate jumps to $\alpha$ (and stays there). When this happens, compliance depends entirely on whether the Appointee is loyal.

The location of the removal threshold in the figure is re-expressed in terms of the state variable as ${\hat \beta} $ , and to highlight how this threshold varies with the stakes of a directive, I include two hypothetical directives in the figure, ${b_L}$ and ${b_H}$ , where ${b_L}$ has relatively lower stakes and ${b_H}$ has relatively higher stakes. In general, the higher the stakes of a directive, the quicker the removal threshold is triggered so that ${\hat \beta} ({b_H}) < {\hat \beta} ({b_L})$ .

Figure 2 clearly illustrates that the Appointee’s compliance rate depends on both the stakes of a directive and whether the President has removal rights. We can also use the figure to make predictions about what compliance rates should look like on average for different directives, that is, when we average across the range of $\beta$ values (the averages are given by the area under each compliance rate curve). I state the relevant predictions in the following hypothesis.

Hypothesis 2 (H2) Compliance with a directive depends on the directive’s stakes and whether the president has removal rights:

1. (a) When the president has removal rights, the average compliance rate will increase in the stakes of a directive and

2. (b) When the president lacks removal rights, the average compliance rate will be lower than when the president has removal rights.

The model also makes predictions about how the compliance rate should vary with $\beta$ . Thus far, I have referred to $\beta$ in broad terms as the state variable, although we can also think of it as how loyal an agency is to the president. Under this interpretation, it is intuitive that compliance should increase in agency loyalty. ^{Footnote 14} Yet, the model highlights that this only happens when the President’s removal threat is not credible. When the threat is credible – that is, when the removal threshold has been surpassed – the compliance rate plateaus and its correlation with agency loyalty falls to zero. I state these predictions as a hypothesis.

Hypothesis 3 (H3) The relationship between how loyal an agency is to the president and the agency’s compliance with a presidential directive is conditional:

1. (a) If the president’s removal threat is credible, there is no relationship between agency loyalty and the compliance rate; but

2. (b) If the removal threat is not credible, or if the president lacks removal rights, the compliance rate will increase with agency loyalty.

The issuance of 12866

Since 12866 is just one executive order, it precludes testing H1 by analysing the president’s issuance decision across many directives. Nonetheless, it is notable that 12866 does not stand completely alone. Rather, every president since Clinton has reaffirmed 12866 in successive orders, including E.O. 13422 (George W. Bush administration) and E.O. 13563 (Obama administration). And 12866 also has predecessors, namely the two review directives issued by the Reagan administration and maintained by the George H.W. Bush administration, E.O.’s 12291 and 12498. Like 12866, all of these orders also limited the issuance of the review directive to the agencies where the president has removal rights, and they did this despite early assurances from Reagan’s Office of Legal Counsel and the American Bar Association that extending the review directive to the independent agencies was within the president’s constitutional authority (Strauss and Sunstein Reference Strauss and Sunstein1986, p.206). ^{Footnote 16}

To gain insight into why the review directive was issued narrowly to agencies where the president has removal rights, we can use the model to construct a counterfactual scenario: What would have happened if the directive were instead issued broadly to all agencies? The main contours of the counterfactual can be seen in Figure 1 above, where ${b_L}$ and ${b_H}$ are two hypothetical directives, one low-stakes and one high-stakes, that we can imagine reflecting the reporting directive and the review directive, respectively. In the figure, the two directives straddle $\tilde b$ , the point at which the President is indifferent about issuing a directive when the President lacks removal rights. If the stakes of a directive are greater than this indifference point, as they are for ${b_H}$ , the President will only issue the directive if the President has removal rights. Doing otherwise would give the President a negative payoff, as reflected by the dashed line that descends below zero. These payoffs capture the idea that a directive with higher stakes is more difficult for appointees to implement alone (although not impossible given that ${b_L}$ and ${b_H}$ fall within the enforceable region), and also that a directive with higher stakes imposes more severe consequences on a president when it is not implemented.

It is easy to see how this counterfactual scenario is relevant to the issuance of 12866. If a president were to issue a review directive to agencies where the president lacks removal rights, a policy disagreement would inevitably arise between the president and one of these agencies over whether to promulgate a regulation, just as it happens with the agencies where the president has removal rights (Heinzerling Reference Heinzerling2014). When the president lacks removal rights, however, a disagreement is harder to resolve because the president’s leverage over the situation is limited. Typically, when the president has removal rights, an appointee will rescind a proposal that, upon review, is found to be problematic for the president. ^{Footnote 17} Yet, when the president lacks removal rights a stubborn appointee does not have to retreat in this way, and can rebuff the president and promulgate the regulation anyway. What the model highlights is that presidents anticipate showdowns like these – and they avoid them.

Compliance rates

In light of H2, the compliance rates associated with the two 12866 directives should depend on the stakes of each directive and whether the president has removal rights. The specific conditions can be re-expressed in a simple two-by-two table, shown here in Table 1, where the compliance rate is: (i) highest for the high-stakes directive when the president has removal rights; (ii) lowest for the low-stakes directive when the president lacks removal rights and (iii) moderate for the low-stakes directive when the president has removal rights. No compliance rate exists for the high-stakes directive when the president lacks removal rights because such a directive is not issued in equilibrium.

Table 1. Compliance rates by agency and directive type

In the analysis that follows, I measure compliance differently for each directive. Although this is unavoidable since the directives require agencies to take different actions, the measures nonetheless broadly capture whether the president received what the president asked for when issuing each directive. And since nearly every regulatory proposal developed by an agency presents a compliance opportunity, we can construct compliance rates for each directive, both across agencies and over time.

Empirical models and results

I estimate compliance rates for each directive using a multi-level model, which has the advantage of accounting for the multi-level structure of the data, where regulatory proposals are nested within agencies and presidential administrations. Each model I estimate includes “varying intercepts” (or random effects) for agencies, years and agency-administration pairs, which helps both to account for correlation within these units and, critically, allows for the inclusion of agency-level variables like RemovalRights_a .

The results from estimating the models are shown in Table 2. In columns 1 through 3, the dependent variable is ComplianceReport_i and in Column 4 the dependent variable is ComplianceReview_i . Overall, the results are largely consistent with the hypotheses above. Removal rights clearly increase compliance, as predicted by H2. So does agency loyalty, although this only happens when the president lacks removal rights, as predicted by H3.

Table 2. Compliance rates (multi-level logit models with varying intercepts)

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

To better appreciate the quantity of interest – the compliance rates – Figure 3 plots a predicted compliance rate for each agency against the measure of agency loyalty. The rates for the reporting directive are on the left panel, which are estimated using the Column 2 model, and the rates for the review directive are on the right panel, which are estimated using the Column 4 model. (Note that an agency shows up twice in each panel because the measure of agency loyalty, AgencyLoyalty_at , varies across administrations.)

Figure 3. Agency compliance rates. Notes: the line segments associated with each circle represent a one standard error deviation from the estimated compliance rate.

Consistent with H2, the figure shows that compliance with the reporting directive is about 30% points higher in the agencies where the president has removal rights (open circles) compared to the agencies where the president does not (closed circles). Compliance with the review directive is even higher, at a rate of around 0.9, which is also consistent with H2 (though note my caveat about comparing compliance rates across directives). And consistent with H3, the figure shows no apparent relationship between compliance and agency loyalty except in the agencies where the president lacks removal rights. Overall, the patterns in Figure 3 bear a resemblance to the theoretical patterns shown in Figure 2 above.

The empirical results also allow us to consider how the stakes associated with each regulatory proposal affect compliance. For the review directive, it is notable that increasing the stakes of a proposal has no impact on compliance. This suggests that the stakes associated with implementing the review directive are already substantial enough to induce the removal threat. Thus, any added variation in the stakes due to the proposals themselves has no discernible impact on the compliance rates. ^{Footnote 20}

For the reporting directive, however, Column 3 suggests that the stakes associated with individual proposals can affect compliance. When the president has removal rights the compliance rate increases by about 13% when moving RegulationStakes_i from its lowest value to its highest value. This pattern suggests that the credibility of the removal threat does, in fact, hinge on the stakes associated with individual proposals, whereby the higher the stakes, the more credible the removal threat. Yet, when the president lacks removal rights, the compliance rate decreases by about 30% for the same shift in the index. This suggests that proposals with higher stakes make it more costly for an agency to comply with the reporting directive. Notably, we can only observe this phenomenon in agencies where compliance is not induced by a removal threat. ^{Footnote 21}

The results in Table 2 are robust to different estimation approaches, including: (i) pre-processing the data using a genetic matching algorithm to create “balance” between the proposals in agencies where the president has removal rights and those in agencies where the president does not (Ho et al. Reference Ho2007); and (ii) controlling for additional factors that may affect the transaction costs associated with replacing appointees, including presidential approval rates, the number of senators in the president’s party, and formal limitations on replacing an appointee, as captured by a popular measure of agency independence (Selin Reference Selin2015). Details on these robustness checks are available in Appendix C.

Removal of appointees

I close my analysis of 12866 by returning to the question of whether presidents remove appointees over noncompliance. In light of H4, this should be a rare occurrence since presidents have an incentive to only issue directives that are either self-enforcing or enforceable. Testing this prediction, however, is not straightforward. When appointees part ways with an administration, they usually do so discretely, without revealing whether their departure was preceded by a policy dispute with the president. Nonetheless, the model offers a framework for making inferences about removal decisions from the data we do observe.

In the case of the review directive, one indication that appointee removals have been rare is that the compliance rates associated with the review directive have been so high (around 0.9 on average). In light of the model, this suggests that the president’s removal threat has, for the most part, been credible, and that agencies have responded to this implicit threat by generally supplying the types of regulatory proposals that fit within presidential priorities (of course, many agencies and appointees are loyal to the president – both β and α are likely close to 1 – so they will comply with or without a credible removal threat). Another indication that appointee removals have been rare is that compliance with the review directive does not appear to depend on the loyalty of the relevant agency or the stakes of individual proposals. As discussed, the lack of variation here is consistent with the existence of a credible removal threat.

In the case of the reporting directive, compliance has generally been lower, at around 0.7 when the president has removal rights. Does this indicate that appointees have been removed for noncompliance? Probably not (although it does indicate that the directive has not been self-enforcing). More likely, the lower compliance rates are driven by variation in the stakes of individual regulatory proposals. Some proposals may push the stakes associated with implementing the directive above the removal threshold, whereas other proposals may not reach this bar. This is consistent with the evidence in Column 3 of Table 2, which shows how compliance with the reporting directive increases with the stakes of individual proposals when the president has removal rights. ^{Footnote 22}