Naïve Cronyism and Neutral Competence: Patronage, Performance, and Policy Agreement in Executive Appointments

Abstract

Much of the bureaucratic literature suggests that, when staffing the bureaucracy, executives want agents who are both responsive to their political needs and possess the competence needed to fulfill their directives. However, institutional barriers—such as the requirement for legislative confirmation—exist that may make pursuing a strategy of responsive competence difficult, if not impossible. Here, I examine a model of bureaucratic appointments that allows for informationally imperfect agencies. I show that when legislative assent is required, trade-offs between ideology and either patronage or agency performance—or both—are often required to ensure legislative confirmation. The same dynamics are not present for unilateral appointments. Finally, using a data set that incorporates the ideologies of federal program managers, the performance of federal programs, and whether program managers were patronage appointees, I conduct a series of empirical tests that support the model’s predictions.

INTRODUCTION

In the late 19th century, a quarter century before he was elected President of the United States, Woodrow Wilson (1887) offered his prescription for what has since become known as the “politics-administration dichotomy”—in particular, the idea that “administration lies outside the proper sphere of politics” and that “administrative questions are not political questions” (210). These notions became the foundation for the arguably normatively attractive ideal of “neutral competence,” the core tenet of which Kaufman (1956) described as “the ability to do the work of government expertly, and to do it according to explicit, objective standards rather than to personal or party or other obligations and loyalties” (1060).

Over the years, however, this ideal has become harshly questioned, with some going as far as to deride it as nothing more than a strawman that has since been “firmly rejected as a naïve misunderstanding of the inherently political context and nature of the administrative process” (Moe 1985, 265). Instead, as many of its detractors have argued, presidents practice a form of “responsive” competence that has only grown more pervasive over time (e.g., Moe 1985; Weko 1995). According to this school of thought, presidential administrations are best served by an appointments strategy wherein individuals are appointed “on the basis of loyalty, ideology, [and/or] programmatic support” in order to “enhance responsiveness throughout the administration,” while “professionalism, expertise, and administrative experience” are emphasized in order to “enhance organizational competence” (Moe 1985, 245). However, the inevitable clash between the “demand for political preferment” and the “demand for technical competence,” due to presidents’ overarching desire for loyalty and effective implementation of their policy agendas, oftentimes results in responsiveness at the expense of competence (e.g., Bawn 1995; Epstein and O’Halloran 1999; Heclo 1988, 46; Wood 1990; Wood and Waterman 1991).¹

Importantly, this focus on loyalty—and the corresponding decrease in competence—directly translates into decreased bureaucratic performance and a lessened ability to effectively pursue the policy goals of the sitting president. For example, Lewis (2007) has shown that agencies headed by political appointees receive lower Program Assessment Rating Tool [PART] scores than those headed by careerists.² Similarly, those agencies headed by patronage appointees—those presumed to be the most loyal to the president—receive even lower scores than those headed by appointees in general (Gallo and Lewis 2012). Numerous other works have argued a similar point—that is, emphasizing loyalty to the exclusion of competence results in a bureaucracy unable to effectively implement a president’s agenda, due to the fact that a minimal level of competence is necessary for effective policymaking (e.g., Aberbach and Rockman 2000; Gailmard and Patty 2007; Huber and McCarty 2004; Lewis 2008, 2009; Nathan 1983). These findings, of course, raise the question of why executives would ever want to emphasize loyalty over competence in the first place and why former Reagan aide Lyn Nofziger noted that “the first thing [administrations] do is get loyal people, and competence is a bonus” (Nofziger 2003, 37). Perhaps even more important is the fact that the loyalty-competence trade-off does not provide an answer as to why legislatures are willing to confirm nominees. Indeed, why would a rational legislature confirm a nominee that is both incompetent and loyal to the president, especially given the decline in Senatorial deference in the last 30 years (Ho 2007)?

Thus, the goal of this article is to uncover the factors that lead to arguably incompetent agencies. I structure this article as follows: I first examine the extant literature on this topic and discuss in detail two theories that purport to explain why the loyalty-competence trade-off occurs, with competence sacrificed in the name of responsiveness. I then draw on the existing bureaucratic appointments literature and present a series of formal models wherein I relax the implicit assumption of perfect competence and, therefore, allow—but by no means require—a loyalty-competence trade-off to enter into political calculations.³ I show that legislatures confirm patronage nominees because they are oftentimes closer ideologically to the legislature than are other types of appointments, as presidents are often forced to make trade-offs between patronage and policy in order to ensure their less-competent patronage nominees are confirmed. Other times, presidents are more willing to consider ideologically distant nominees when they are sufficiently competent. Finally, using a data set that incorporates the ideologies of federal program managers, the performance of federal programs, and whether program managers were patronage appointees, I conduct empirical tests of the model’s predictions regarding the trade-offs that need to be made to ensure confirmation. Following these results, I present some avenues for future research.

WHY DOES THE TRADE-OFF OCCUR?

Proponents of responsive competence have long argued that presidents view a neutrally competent bureaucracy as insufficient for their immediate policy needs, and that the best way to ensure they achieve their policy goals vis-á-vis the bureaucracy is to staff it almost entirely with loyalists responsive to their needs (e.g., Moe 1985). Still others have argued that excessive competence in general is viewed as a veritable threat to the effective prosecution of the president’s policy goals, due to the fact that those with previous relevant experience have their own personal agendas and beliefs about how agencies should be run, which may conflict with the goals of the president (e.g., Cohen 1998; Pfiffner 1983).⁴ To this end, competence is desired insofar as it enables presidents to achieve their policy goals, and no more. Perhaps reflecting this preference, political appointees tend to have less relevant experience than do career-service workers (Gallo and Lewis 2012; Lewis 2007).

Interestingly, this seemingly self-defeating behavior on the part of presidents may be completely unnecessary, even when the lack of obvious loyalty among competent bureaucrats is taken into consideration. To this end, it has been argued that career-service workers—who are arguably the most competent and least (overtly) loyal bureaucrats of all—are not necessarily bound by their own personal ideologies and are instead generally willing to adapt to presidents’ changing policy demands, despite fears to the contrary (Edwards 2001; Rourke 1992; Wilson 1989). These empirical observations potentially undermine the optimality of the strategy of avoiding competent bureaucrats out of fear of policy insubordination.

On the other hand, Lewis (2008, 2009) has maintained the trade-off is not so much due to a fear of competent agencies undermining the will of the executive, but rather is a function of simple fact that patronage appointees tend to be less competent than professionals (Gallo and Lewis 2012; Lewis 2007). To this effect, “presidents have to ensure bureaucratic responsiveness . . . . tough the distribution of patronage. They are willing to trade some competence in order to get agencies to do what they want them to do and to build political support” (Lewis 2008, 202). Indeed, patronage appointments “provide a means for presidents to hold supporters in line, signal policy commitments, and accomplish their policy goals” (Lewis 2008, 208). In addition to these immediate political benefits, those in charge of staffing presidential administrations view patronage appointments as having long-term benefits—for example, Chase Untermeyer, President George H. W. Bush’s Director of the Office of Presidential Personnel, referred to low-level patronage appointments as being a sort of “farm system” where appointees garnered “experience and credentials so that during [the] administration or in some future administration they [would] be able to compete [for higher-level positions]” (Untermeyer 1999, 19).

DIRECT EXECUTIVE APPOINTMENT MODEL

To analyze formally the theoretical underpinnings that drive different theories of agency competence and responsiveness, I propose a formal model not unlike others used to analyze other aspects of bureaucratic appointments and delegation (e.g., Bendor and Meirowitz 2004; Bertelli and Feldmann 2007; Gailmard 2002; Hollibaugh forthcoming; Hollibaugh, Horton, and Lewis forthcoming; McCarty 2004; Volden 2002), many of which are based on the seminal agenda-setter model of Romer and Rosenthal (1978). Initially, I consider the rather simple case where the executive can unilaterally make agency appointments without requiring legislative assent. In the United States, roughly two-thirds of all appointments are made in this fashion.^5,6

The direct executive appointment model consists of two players—the Executive and the Agency. Both players are assumed to have quadratic preferences over policy outcomes on a single dimension, represented as $ui(x)=−(x−xi)2$ for all $x∈X⊆ℝ$ and $i∈{E,A}$ . I assume that decisions are delegated to agencies because of their superior information and expertise regarding policy decisions and consequences. Formally, the outcome is $x=p + ω$ , where $p∈ℝ$ is the policy chosen by the agency and $ω∼U[−Ω,Ω]$ , where $Ω∈ℝ++$ , represents factors that are unobserved when statutes are written and agency staffers are chosen, but which are (potentially) observed by the agency before policy implementation. However, in contrast to previous models and to account for the possibility that the executive may try to induce incompetence, I relax the assumption that agencies can discern the true value of ω without error. Rather, agencies observe ω with probability c, and observe no shock whatsoever with probability $1−c$ , thus acting as if $ω=0$ , due to the symmetry of the distribution from which ω is drawn.⁷ I denote this observed value of ω to be $ω^$ .

Finally, I allow for nonpolicy patronage benefits. While patronage appointments “provide a means for presidents to hold supporters in line…. a accomplish their policy goals” (Lewis 2008, 208), agencies run by appointees with connections to the president’s party or campaign (i.e., those most likely to have been patronage appointees) tend to perform worse than agencies run by careerists and nonpolitical appointees, suggesting patronage is one possible method by which bureaucratic incompetence arises (Gallo and Lewis 2012; Lewis 2007). In order to analyze the conditions that might prompt an executive to prefer a patronage appointee over a careerist or other professional for a particular position, I assume that executives face the choice of which type $τ∈{L,H}$ of proposal to make; executives can propose either a high type—a professional $(τ=H)$ —or a low type—a patronage appointee $(τ=L)$ . Both types are able to induce agency preferences $xA$ over the same policy space. However, for any given agency, the competence pool for patronage appointees is assumed to be less deep than the pool for professional appointees. Lower patronage competence can result from many sources, including the fact that patronage appointees tend to have less experience in the agencies to which they are appointed, less subject area experience, and less public management experience in general (Cohen 1998; Heclo 1975, 1977; Lewis 2007). To account for these differences in competence, I assume that if the executive chooses to propose a professional for an agency position, she can find one such that she can set the ex post agency ideal point to some $X⊆ℝ$ and the ex post agency competence to some $c∈[c_,1]$ , where $c_∈(0,1)$ .⁸ However, if she chooses to propose a patronage appointment, then she is constrained and can only set the agency’s competence to some $c∈(0,cˉL]$ , where $cˉL∈[c_,1)$ .^9,10 For notational purposes, define $cˉτ$ such that $cˉτ=1$ if $τ=H$ , and $cˉτ=cˉL$ if $τ=L$ . The (potential) incentive for the executive to propose a patronage appointment is that she derives some nonpolicy benefit $ρ∈ℝ+$ from doing so.^11,12 Thus, the executive’s utility function is as follows:

Similarly, the agency’s utility function is defined as

The game proceeds according to figure 1. After Nature draws ω, the executive can choose whether or not to make an agency appointment. She will decline to make an appointment if her optimal appointment provides her with strictly worse utility than the status quo. Appointees induce an ex post agency ideal point $xA$ and ex post level of agency competence c. If no appointment is made, then the status quo agency $(xQ,cQ)$ , where $xQ∈X$ and $cQ∈(0,1]$ , stays in effect. Following executive action (or inaction), the agency observes $ω^$ and chooses a policy p.

As the informed player moves last, I employ the sequential equilibrium solution concept and solve the game via backwards induction (Kreps and Wilson 1982). Equilibrium solutions and comparative statics are presented in Appendix B.

Given that all players possess quadratic utility functions, with utility decreasing in distance from one’s own ideal point, intuition suggests that the executive’s utility is maximized when she can appoint someone such that the ex post agency preferences align with her own, and the agency is competent enough to translate these preferences in policy outcomes. This rather uncontroversial result is summarized in Proposition 1.

Proposition 1. When the executive is able to make unilateral appointments to agencies, only ideologically convergent and conditionally competent agencies (where $c*=cˉτ$ ) occur in equilibrium.

Whether or not a patronage appointment is made will depend solely on whether the nonpolicy patronage benefits are sufficiently high. Thus, when patronage utility is nonexistent (or the executive is solely motivated by policy), patronage appointments never occur in equilibrium.

Proposition 2. When patronage benefits are nonexistent (or merely sufficiently low), only perfectly competent agencies occur in equilibrium.

LEGISLATIVE CONFIRMATION MODEL

The direct appointment model suggests responsive and conditionally competent agencies always occur when executives can choose exactly who they want to fill agency posts. In every case, the executive’s optimal action is to always induce an ex post agency ideal point such that the agency’s preferences align with her own, and the agency is as competent as possible, conditional on the type of appointment made.

However, as previously mentioned, approximately one-third of federal appointees in the United States require Senate confirmation (Lewis 2008). Moreover, these positions tend to be those most relevant to agency performance and policy. Because of this, I extend the direct executive appointment model to incorporate legislative confirmation. Given this additional restriction, it seems plausible that circumstances exist where an executive will compromise on loyalty and instead seek neutral instead of responsive competence, if doing so will ensure a successful confirmation.

The setup of the model is similar to that of the previous one, as illustrated in figure 2. Like the executive and the agency, the legislature possesses quadratic preferences over policy outcomes. In equilibrium, the legislature will reject an executive’s proposal (assuming one is put forth) if and only if acceptance would make the legislature strictly worse off versus the status quo. Relatedly, I assume that the executive will not attempt a proposal if the legislature is only willing to confirm those that make the executive strictly worse off. Thus, the only proposals that occur in equilibrium are those which are Pareto optimal.¹³

In this model, the agency’s equilibrium policy choice is unchanged from that in the previous one. Similarly, the legislature’s policy preferences in the current model are indistinguishable from the executive’s in the previous model. As before, I employ the sequential equilibrium solution concept and solve the game via backwards induction (Kreps and Wilson 1982), with equilibrium solutions and comparative statics presented in Appendix B. Key results from the legislative confirmation model are presented below.

Proposition 3. In equilibrium, patronage appointments will never result in agencies whose ideal points are strictly closer to the executive’s than professional appointments.

This seemingly counterintuitive result stems from the asymmetry in patronage utility (i.e., the executive receives it, whereas the legislature does not) coupled with the assumption that patronage appointments induce agency competence that is no higher, and potentially lower, than that induced by similarly situated professional appointees. In these situations, if the legislature is sufficiently comfortable with the status quo, it may require the executive compromise somewhat on ideology in order to ensure her less-competent appointee will be confirmed. Regardless of the outcome, it must be noted that Proposition 2 holds in the model with legislative confirmation. That is, sufficiently high patronage utility is a necessary—but not sufficient—condition for imperfect agency competence, due to the fact that executives can counterbalance any loss of policy utility with patronage utility.

Figure 3 provides examples of the best possible appointments in graphical form. Importantly, this is an example where there exist some status quo ideal points so desirable to the legislature that no patronage appointments are acceptable. Additionally, this example also includes regions where the status quo is sufficiently divergent from the legislature such that patronage appointments are theoretically possible, but are situated such that the executive will be forced to make ideological trade-offs to ensure confirmation; here, patronage appointments induce agency ideal point strictly closer to the legislature than similarly situated professional appointments would.¹⁴

As the pool of potential patronage appointees becomes more competent, the difference between patronage appointments and professional appointments grows smaller in terms of the expected outcome, making it more likely that patronage appointments will arise in equilibrium, due to the fact that the amount of patronage utility required will be comparatively small, and the executive will have to “give up” less in terms of agency ideology. In the American context, this is most likely to occur for agencies in areas that have traditionally been areas of expertise for the executive’s party. This is largely because agencies for which potential patronage appointees tend to be most competent (though not as competent as careerists) are the ones that are most ideologically aligned with the president (Lewis 2009). For example, these might include agencies in the Department of Defense for the Republican Party and agencies in the Department of Labor for the Democratic Party.¹⁵

However, it is also true that when the pool of potential patronage appointees becomes more competent, the executive will be more willing to cede ground on the ideological dimension in order to ensure confirmation of a patronage appointee; moreover, the legislature will be more willing to confirm patronage nominees even in light of a favorable status quo. Coupled with the fact that the executive can always find a professional nominee for which the legislature is willing to give up a favorable status quo (though whether the executive will want to make such nominations in the first place is a different matter altogether), it can be shown that ex post agency ideal points disproportionately favorable to the legislature are often the result of professional or high-competence patronage appointments. In these cases, low-competence patronage nominees will not be confirmable.

Proposition 4. Ceteris paribus, the executive will be more willing to induce ideologically divergent nominees if the induced agency competence will be sufficiently high.

EMPIRICAL ANALYSIS

I now examine some of the implications of the models. To do so, I combine PART scores and other agency- and program-level data collected by Gallo and Lewis (2012) with ideal point estimates created by Bonica, Chen, and Johnson (2012), hereafter referred to as BCJ. Together, these sources contain a wealth of information on appointees from a host of federal agencies and the agencies themselves. The BCJ data contain ideal point estimates for all individuals who (a) have made a political contribution of sufficient size since the Federal Election Commission began collecting contribution data in 1979 and (b) identified themselves as working for a federal agency at the time of the contribution. Importantly for my analysis, the BCJ data also include ideal point estimates for all contribution recipients, which will allow me to measure ideological distances between appointees, presidents, and senators.16–18 Methods quite similar to that used to generate the BCJ data have been utilized to recover estimates of ideology of elected officials in the same space as unsuccessful candidates, political action committees, individual donors, and many others (Bonica 2013, forthcoming; Chen and Johnson forthcoming). In each case, the estimates for elected officials recovered from campaign finance data correlate very highly with estimates of ideology recovered from roll-call voting, with the crucial benefit of being able to recover estimates of ideology for those who were never elected (Bonica 2013, forthcoming; Bonica, Chen, and Johnson 2012).

The data collected by Gallo and Lewis (2012) include agency, bureau, and program-level characteristics, the most important of which for my purposes are the names of program managers, the PART scores of federal programs and which agencies/bureaus administered them, whether the program manager was a patronage appointee, whether the program manager was serving in an acting capacity, and whether the program manager was a PAS appointment or otherwise. As the data provide me with information on principal and appointee ideology, whether or not the appointee was appointed for patronage reasons, and the eventual performance of programs headed by different appointees and appointee types, I am well equipped to test various implications of the model.

Hypotheses

Proposition 3 presents arguably the most counterintuitive result of the model. It suggests that, contrary to what might be conventional wisdom, patronage appointments are not necessarily ideologically more compatible with the executive than are professional appointments. Indeed, there often exist trade-offs between patronage, competence, and policy utility, as illustrated by figure 3. Thus, I do not expect to find that program managers who are patronage appointees are ideologically closer to the ideal point of the president than those who are not. Hypothesis 1 is thus derived.

Hypothesis 1. Ceteris paribus, legislatively confirmed patronage appointments will be ideologically farther from the president’s ideal point than legislatively confirmed professional appointments.

Proposition 4 suggests that the executive is more willing to consider ideologically distant appointees as their competence increases, which suggests the following form for Hypothesis 2:

Hypothesis 2. Ceteris paribus, the performance of programs with legislatively confirmed managers will be increasing in the ideological distance between the president and the manager.

Of course, Proposition 1 suggests that ideologically convergent appointments will arise in equilibrium when unilateral appointments are concerned, suggesting there will be no relationship between ideology and either patronage or performance. Thus, Hypotheses 3 and 4 are derived.

Hypothesis 3. Ceteris paribus, for unilateral appointments, there will be no relationship between ideology and patronage.

Hypothesis 4. Ceteris paribus, for unilateral appointments, there will be no relationship between ideology and performance.

Finally, I examine the veracity of the assumed relationship between patronage and competence:

Hypothesis 5. Ceteris paribus, programs managed by patronage appointees will exhibit lower performance than those not managed by them.

Estimation Strategies, Data, and Results

Because all appointee characteristics are simultaneously determined in the theoretical model, care must be taken to ensure that issues arising from endogeneity are kept to a minimum. I do this in two parts—first, I use PART scores as a proxy for appointee competence. These scores were used in the George W. Bush administration as a way of measuring federal program performance according to a set of criteria.¹⁹ As my data set is constructed such that the only program managers included are those whose PART scores were determined postappointment, I can be confident that PART scores are influencing neither the ideological convergence between presidents and program managers, nor their statuses as patronage appointees, both of which are determined prior to appointment.

Second, to examine the relationship between patronage and ideology, I use simple distributional tests to examine whether the distribution of President–Appointee Distance—measured as the absolute value between the BCJ scores of the appointing president and program manager—varies for different values of Patronage Appointees, a dichotomous variable that equals one if an appointee had previously worked for the president’s party or campaign, or in the White House or Congress, and zero otherwise.

Examination of Hypothesis 1 is straightforward. Using a standard t-based difference of means test, I can determine whether the difference between $μPatronage PAS appointments$ and $μNonpatronage PAS appointments$ is significant and, more importantly, whether legislatively confirmed patronage appointments will be ideologically farther from the president’s ideal point than legislatively confirmed professional appointments. Results are presented in table 1.

Results generally support Hypothesis 1. The mean value of the distribution from which President-Appointee Distance for legislatively confirmed patronage appointments is drawn is found to be significantly larger than that of President-Appointee Distance at the 90% level. This suggests the existence of a reasonably strong trade-off between ideology and patronage for Senate-confirmed appointments.

Examination of Hypothesis 3 will be less straightforward; as it focuses on unilateral appointees and most of the program managers in the data set are legislatively confirmed, I am faced with small samples for each of my categories. As such, traditional t-tests of this hypothesis will be insufficient. Moreover, because Hypothesis 3 predicts no significant difference, it cannot be directly assessed using traditional null hypothesis rejection techniques. Instead, I employ Kruschke’s (2013) Bayesian alternative to the standard frequentist t-test. Essentially, this approach allows me to conduct t-tests while leveraging the advantages of Bayesian estimation; in the present case, the advantages I am most concerned with are the applicability of Bayesian methods to small samples, as well as the ability to determine the variance of the posterior distribution relative to zero. For reasons of parsimony, the technical specifications of the model setup—including priors—are in Appendix A.

With the sample draws, I am able to estimate whether the means of any two arbitrary distributions, $Y1∼t[0,∞)(μ1,σ1,ν)$ and $Y2∼t[0,∞)(μ2,σ2,ν)$ differ to any significant degree, where $Y1$ and $Y2$ represent the two sample vectors of President-Appointee Distance that are to be compared. Substantively, this allows me to examine hypothesis 2. Essentially, I examine whether the differences between $μPatronage non-PAS appointments$ and $μNonpatronage non-PAS appointments$ are significant. Results are presented in table 2.²⁰

Results are suggestive of no effect, as the 90% credible interval contains zero, and the point estimate is equivalent to 3% of the pooled standard deviation (SD), which is approximately 0.373, suggesting little substantive effect. However, as Hypothesis 3 posits no effect, traditional tests that rely on examining whether a confidence interval contains zero will not suffice, though it is encouraging that the null hypothesis of no difference in means is not rejected. Indeed, the failure to reject the null hypothesis cannot be considered equivalent to the acceptance or support of the null. One alternative, suggested by Kruschke (2013), Rainey (forthcoming), and others is to establish a region of practical equivalence (ROPE). Within this region, nonzero values are said to be practically equivalent to no effect, due to their lack of substantive or meaningful significance. Unfortunately, this method requires that the analyst specify a ROPE a priori.

Instead, I approach this problem from a different perspective, allowing the data to speak for themselves and the reader to draw his or her own conclusions about whether the hypothesis of no effect is supported. I estimate several ROPEs of varying lengths, framed in terms of the pooled sample SD, $σ^=sd(Y(Patronagenon−PASappointments)∪Y(Nonpatronagenon−PASappointments))$ . Table 3 provides the results.

Upon inspection, the majority of the values from which the difference between $YPatronage non-PAS appointments$ and $YNonpatronage non-PAS appointments$ is drawn lie within rather small ROPEs. Over half of the values lie within one-eighth of the pooled SD away from zero and approximately 90% lie within one-third of one SD from zero. Virtually all of the draws lie within one SD from zero. Additionally, given that the median difference is equivalent to $0.063σ^$ and the 90% credible interval is equivalent to $[−0.284σ^,0.391σ^]$ , the evidence for no practical difference seems very strong. Thus, it seems that the mean of the distribution from which President-Appointee Distance for unilaterally appointed patronage appointments is drawn is not likely to be substantively different than that of President-Appointee Distance for the unilaterally appointed nonpatronage group, suggesting little-to-no trade-off between ideology and patronage for unilateral appointments, in accordance with Hypothesis 3.

To assess Hypotheses 2, 4, and 5, I estimate a series of agency-level regression models with PART score as the dependent variable, and President-Appointee Distance and Patronage Appointment as the main independent variables of interest. Three regression models are estimated for Senate-confirmed appointments, and two are estimated for unilateral appointments.

In addition to the aforementioned independent variables of interest—President-Appointee Distance and Patronage Appointment—I include a number of other control variables that may affect the performance of agencies, either directly (agency-level variables) or indirectly (manager-level variables). Divided Government is an indicator variable that equals one if the President and Senate Majority Leader are of different parties and zero otherwise.²¹Previous bureau experience, previous experience in other federal department, public management experience, and private sector management experience are dichotomous variables that equal one if the program manager possessed the relevant characteristic prior to being appointed and zero otherwise. Previous job in government is a dichotomous variable that equals one if the program manager’s position immediately prior to being appointed was also in government. Tenure as bureau chief is the number of months the program manager was in his or her capacity as bureau chief at the time of program assessment. Logged program budget is the logged size of the program budget (in millions of dollars). Agency ideology is the Clinton and Lewis (2008) measure of the ideology of the federal agency in which the program is located; higher values indicate agencies are perceived as being more conservative, and lower values indicate the opposite. Since PART scores were only issued during the George W. Bush administration, it is possible that programs in more conservative agencies are more likely to receive higher scores solely on that basis. Advanced degree is a dichotomous variable that equal one if the program manager had a masters degree or a doctorate at the time of assessment, and zero otherwise. Finally, I include fixed effects for the years in which the program was rated as well as program type.^22,23

In the models where I examine agencies with program managers appointed under PAS authority, Hypothesis 2 suggests a positive and significant relationship between President-Appointee Distance and PART score; that is, programs whose managers are ideologically divergent from the appointing president should exhibit higher performance. Additionally, Hypothesis 4 suggests no relationship between PART score and President-Appointee Distance in the model where non-PAS program managers are examined, though as before, additional interpretation will be required to assess this hypothesis of no meaningful effect. Finally, Hypothesis 5 suggests a negative relationship between Patronage Appointment and PART score in all models, suggesting programs whose managers are patronage appointees should exhibit lower levels of performance. As an additional check, I estimate a series of ordered logit models using PART rating as the dependent variable instead of PART score. This variable simply captures the categorical rating for each program, which ranges from “Ineffective” (lowest) to “Effective” (highest). Using this approach has the benefit of potentially reducing the measurement error inherent in the PART score metric; arguably, the categories themselves are measured with greater precision than the raw totals. As a final robustness check, I multiply impute the missing values of Appointee Ideology and estimate additional models based on the imputed data (King et al. 2001; Honaker, King, and Blackwell 2011).²⁴

Results of the regressions are presented in table 4.²⁵ Upon first glance, it is clear that the results generally support Hypotheses 2 and 5, and appear to support Hypothesis 4. Hypothesis 2, which posits a positive relationship between President-Appointee Distance and PART score in the PAS models, is supported by the positive and significant coefficients on President-Appointee Distance in three of the four PAS models. This suggests that when presidents are forced to appoint program managers who are ideologically distant, they are more likely to increase their utility on other dimensions and appoint those who will lead to higher agency performance.

Hypothesis 5, which posits a negative relationship between Patronage Appointment and PART score in all models, is supported by the negative coefficient on Patronage Appointment in all seven models. However, the coefficient reaches statistical significance at the 90% level in one-tailed tests only in the cases of the four models estimated using PAS data, as well as one of the three models estimated using non-PAS data. Nonetheless, these results provide some evidence in favor of the proposition that programs managed by patronage appointees tend to perform worse than those led by other types of appointees.

Hypothesis 4 is a bit more difficult to adjudicate. As before, I reinterpret the credible intervals of the coefficient estimates in the ordinary least squares models in terms of the SDs of both President-Appointee Distance and PART score, examining the effect of a one SD shift in President-Appointee Distance on PART score. For both models, the majority of bootstrapped regression coefficients lie within narrow ROPEs, as shown in Table 5. In both models, over 90% of the bootstrapped coefficients lie within one-half of one SD from zero, and a majority of bootstrapped coefficients in each model lie within one-fourth of one SD from zero. Additionally, in the model where imputed data are examined, a large majority of the bootstrapped coefficients lie within one-eighth of one SD from zero. However, the ROPEs are still relatively large considering the scale of the dependent variable, as the sample SD of PART score is approximately equal to 17.686 in Model 5 and 19.108 in Model 6. Nonetheless, this analysis provides some positive support in favor of Hypothesis 4’s prediction of no trade-off between performance and either patronage or ideology for unilateral appointments.

Overall, the empirical analyses reported here support all of my hypotheses to varying degrees. Foremost, the balance of empirical evidence supports the notion that presidents desiring to make appointments subject to legislative confirmation are forced to make trade-offs between policy, patronage, and performance. Senate-confirmed program managers who are patronage appointees tend to be ideologically more distant from the president than those who are not patronage appointees, as defined by previous work for the president’s party or campaign, suggesting a trade-off between patronage and policy. Federal programs headed by Senate-confirmed managers who are ideologically divergent from their appointing presidents tend to perform better (at least on the PART score metric) than those headed by ideologically convergent managers, suggesting a trade-off between policy and performance. Finally, federal programs headed by patronage appointees tend to exhibit lower levels of performance than those not headed by them, suggesting a trade-off between patronage and performance. Additional suggestive evidence is provided to support the notion that these trade-offs are not present when unilateral appointments are considered.

DISCUSSION AND CONCLUSION

The notion that executives desire a bureaucracy that is both competent and responsive to their needs is neither counterintuitive nor surprising. However, this notion is at odds with the normative ideal that the bureaucracy should be neutral and carry out its directives without regard to politics or party. In this article, I developed a formal model and examined the conditions under which executives will pursue responsively competent agencies and neutrally competent agencies. As presented, the model represents an advance in the bureaucratic appointments literature due to its explicit incorporation of bureaucratic competence. Foremost, I find that when executives and legislatures are motivated solely by policy and possess complete information about the preferences and competencies of those nominated to agency posts, perfect agency competence results in equilibrium. Indeed, executives want agencies to be both politically responsive and administratively competent. However, when the executive is constrained by a legislature with whom she disagrees on policy, then such an agency may be unattainable. In this case, the executive’s next-best strategy may be to sacrifice some political responsiveness and maintain maximal agency competence, effectively inducing (relatively) neutral competence at the agency level. Further ideological compromises may be needed if patronage appointments are to be made.

Given the initial prediction of perfect agency competence and the obvious disconnect from empirical observation that this theoretical result poses, I examine other factors to explain why instances of bureaucratic incompetence abound in the literature. I focus on nonpolicy incentives accorded to the executive by patronage demands and show that these factors are one possible explanation for the existence of less-than-perfect agency competence. I then show that when the president desires to make patronage appointments to legislatively confirmed positions, they will often have to sacrifice political responsiveness to ensure confirmation. This is largely due to the legislature not receiving any benefits from patronage while being saddled with relatively incompetent appointees; in order for patronage appointments to be made, the executive will often have to reach a compromise with the relevant legislators by providing them with increased policy utility.

Empirical analysis supports the model’s predictions with regard to the circumstances under which patronage appointments occur and how the political environment affects agency ideology. Importantly, the empirical results support the model’s predictions that legislatively confirmed patronage appointees will be more ideologically divergent from the executive than similarly situated nonpatronage appointees. Moreover, they support the notion that executives, faced with constraints imposed by the legislative confirmation requirement, may be forced to choose between policy congruence and performance, or between patronage and performance. Finally, the theoretical model, as well as the empirical results, supports the notion that these trade-offs will not exist for unilateral appointees. The notion that patronage appointments may not necessarily be loyalists or lackeys to the president is at odds with much of the conventional wisdom regarding patronage appointments, yet it helps to explain why the legislature would be willing to confirm those who are less competent and provide the president with some nonpolicy benefit.

With regards to the model itself, while it represents a significant advance in the study of appointments, it still suffers from the common flaw that besieges all formal models. That is, it is merely an abstraction of the dynamics at play and contains unavoidable simplifications. Importantly, other representations of the process are possible, and they may yield different insights into the appointments process. Additional work on this topic would benefit from examination of these alternative representations, not the least of which is the possible exploration of a more dynamic context. The environment presented here is one in which only one nomination is considered at a time, and the game ends after the nomination is confirmed or rejected. Obviously, this is not a wholly accurate presentation of the environment in which nominations take place; instead, it is almost always the case that multiple nominations are pending at any one time and legislators’ decisions are, at least in part, based on what future nominees might be placed before them.²⁶ Relatedly, one might consider the fact that most nominations are not rejected; rather, legislative disapproval often manifests itself in the form of confirmation delay (McCarty and Razaghian 1999; Shipan and Shannon 2003).

Moreover, while the models presented here include the possibility of nonpolicy utility in the form of patronage, they blackbox the method by which it is generated. A modification of the model might explicitly model the dynamics of how this utility is derived. For example, if patronage incentivizes future work on behalf of the party, and thus indirectly increases turnout, an executive might be more willing to utilize patronage appointments if the end result would be a legislature whose preferences are more ideologically in line with her own.

Finally, the models here assume both the executive and the legislature are completely informed about the policy preferences and competence of agencies. Indeed, administrations may have incentive to equivocate or simply lie about the preferences and abilities of those put forth for consideration. Alternatively, potential nominees may have incentive to lie to both executive administrations and legislatures about their own preferences and abilities, or executives may even be unsure of their own preferences. None of these possibilities are considered here, and the possibility that bureaucratic incompetence arises as a result of incomplete information is fertile ground for future work.

APPENDIX A. SETUP OF KRUSCHKE’S BAYESIAN T-TEST

While my end goal is to estimate whether two distributions, $Y1∼t[0,∞)(μ1,σ1,ν)$ and $Y2∼t[0,∞)(μ2,σ2,ν)$ , differ to any significant degree—where Y₁ and Y₂ represent the two sample vectors of President-Appointee Distance that are to be compared—the Bayesian method I employ requires a little bit of setup before analyses can be performed.²⁷ Like all Bayesian methods, Kruschke’s (2013) test requires that I assign priors to the parameters of interest. Thus, I assume that the means of the distributions of interest (i.e., the t distributions from which various values of President-Appointee Distance are drawn, conditional on the type of appointment under analysis) are drawn according to a normal distribution that is left truncated at zero, with mean equal to that of the pooled data, and SD equal to 1,000 times the SD of the pooled data; this latter assumption ensures that the prior is relatively uninformative and will exert minimal influence on the final assumptions.²⁸ I further assume that the prior distribution for the SDs of the t distributions of interest are uniform distributions whose lower and upper limits are respectively set to one-thousandth of the SD of the pooled data and one thousand times the SD of the pooled data. Finally, the shared degrees of freedom parameter (minus one) for the t distributions of interest is characterized according to an exponential distribution with $λ=129$ . These priors have the effect of assuming the two distributions from which the values of President-Appointee Distance are equal. In summary, the prior distributions of the parameters of interest are displayed as follows:

Following specification of the priors, repeated Markov Chain Monte Carlo [MCMC] sampling using the Gibbs sampler allows me to generate distributions of the parameters of interest.²⁹ Because of the small sample sizes, I allow the MCMC sampler to run for 150,000 draws, disposing of the first 50,000. To ensure independence between draws, I use a thinning interval of 10, only keeping every 10th draw. This leaves me with a sample of 10,000 draws for each parameter of interest.

APPENDIX B. MODEL FORMALIZATION

Note that in both models, $ω^=ω$ with probability c and 0 with probability $1 −c$ .

DIRECT EXECUTIVE APPOINTMENT MODEL

After observing $ω^$ , the agency sets a policy $p∈ℝ$ , which it chooses in order to maximize $EuA(pω^)=−(p+ω^−xA)2$ . Clearly, the agency will set $p*(ω^)=xA−ω^$ , as shown in Remark 1. Importantly, less-competent agencies have greater variance in the policies they implement, as they are less consistently able to correct for the true policy shock ω.

Remark 1. In equilibrium, the agency’s optimal policy choice is $p*(ω^)=xA−ω^$ .

Working up the game tree, the executive’s optimal appointment is determined given her own preferences over policy and the equilibrium policy that will be chosen. Importantly, as agency responses are conditioned on $ω^$ , the expected value of which is a function of c, the executive must take this into account and determine her expected utility accordingly.

Lemma 1. The executive’s expected utility $EuE$ is given by

Clearly, the executive’s utility is strictly increasing in c.

Remark 2. For either type of appointment, the executive’s expected utility will be maximized at $c=cˉτ$ .

Given these three results, the results in the main text can be derived.

LEGISLATIVE CONFIRMATION MODEL

First note that Remark 2 can be generalized to the legislature, implying the legislature’s utility is conditionally maximized at $c*=cˉτ.$

Moving up the game tree, I solve for the executive’s optimal proposal. Without loss of generality, assume $0≤xL<xE$ . As previously shown, the executive’s utility is conditionally maximized when the agency’s ideal point is $xE$ and competence is $cˉτ$ . However, due to the legislative confirmation requirement, this is sometimes impossible. If this is the case, she will instead pursue the second-best solution of neutral competence. Thus, she will choose to propose a conditionally competent agency $(c*=cˉτ)$ with whom she disagrees to a certain extent $(xA*∈(xL,xE))$ , where the equilibrium agency ideal point is determined by solving for the proposal along the legislature’s indifference plane that maximizes the executive’s utility.

However, before the equilibrium results can be presented, one notational modification needs to be made. First, denote $xB,τ$ as the constrained proposal of type τ, which is determined by setting $EuL(xA,cˉτ)=EuL(xQ,cQ)$ and solving for $xA$ , focusing on the positive roots since $xL<xE$ .

Lemma 2. The optimal boundary proposals $xB,τ$ are defined as

Given this notation, there exists the possibility that for a given set of parameters, $xB,L$ might not exist. This will occur if $xQ$ is sufficiently close to $xL$ . In particular, by setting $(xQ−xL)2+Ω2(cˉL−cQ)3$ equal to zero and solving for $xQ$ , it can be shown that $xB,L$ will not exist in ℝ if $xQ∈(xL−Ω2(cQ−cˉL)3,xL+Ω2(cQ−cˉL)3)$ . Interestingly, this interval only exists if $cQ≥cˉL$ . From this restriction, Remarks 3 and 4 are derived.

Remark 3. Patronage appointments are not confirmable if induced agency competence will be sufficiently low and the status quo agency preferences will be sufficiently close to those of the legislature.

Remark 4. Professional nominees are always confirmable.

Further note that since $cˉL<1$ , it follows that $xB,L<xB,H$ . In other words, when the legislature’s constraint binds, and the status quo agency’s ideal point is sufficiently close to the legislature’s, patronage appointments result in ex post agencies farther from the executive’s ideal point than do professional appointments. Formally, if $cQ≤cL$ and $xQ∈(xL−(min{xB,H,xE}−xL)2+Ω2(cQ−cˉL)3,xL+(min{xB,H,xE}−xL)2+Ω2(cQ−cˉL)3)$ , or if $cQ≥cLandxQ∈(xL−(min{xB,H,xE}−xL)2+Ω2(cQ−cˉL)3,xL−Ω2(cQ−cˉL)3)∪(xL+Ω2(cQ−cˉL)3,xL+(min{xB,H,xE}−xL)2+Ω2(cQ−cˉL)3)),$ this condition will hold. Intuitively, this is due to the fact that agency incompetence induces variance into the policymaking process, which has strictly negative utility consequences for the legislature. Thus, Proposition 3 in the main text is derived.

Equilibrium outcomes are presented in Remark 5.

Remark 5. Where it exists, the executive’s optimal proposal $(xA*,c*,τ*)$ in the legislative confirmation model with patronage is given by

The executive will nominate $(xA*,c*,τ*)$ if $xA*∈[xE−(xQ−xE)2+Ω2(c*−cQ)3,xE$$+(xQ−xE)2+Ω2(c*−cQ)3]$ and will make no nomination otherwise. The legislature will confirm $(xA*,c*,τ*)$ if $xA*∈[xL−(xQ−xL)2+Ω2(c*−cQ)3,xL+(xQ−xL)2+Ω2(c*−cQ)3]$ and will reject otherwise.

Proposition 4 can be shown via direct examination of Remark 5.

APPENDIX C. BAYESIAN DIAGNOSTICS

APPENDIX D. SUMMARY STATISTICS

APPENDIX E. REPRESENTATIVENESS OF APPOINTEES WITH IDEAL POINTS

REFERENCES