webpage

Haphazard, Systematic, or Both? An Empirical Investigation of the U.S. Attorney Firings in 2006

Authors: Banks Miller and Brett Curry
Publication: Journal of Law & Courts 6: 379-403
Original source: jlc_published.pdf

Conversion note: this Markdown was generated by text extraction from the hosted file. Review equations, tables, notes, and citations before treating it as a final web copy.

Extracted Text

Haphazard, Systematic, or Both? AN EMPIRICAL INVESTIGATION OF THE US ATTORNEY FIRINGS IN 2006 BANKS MILLER , University of Texas at Dallas BRETT CURRY , Georgia Southern University

Abstract

In 2006, the Bush administration directed nine US attorneys to resign. This decision was a partial cause of the attorney general’s departure from the administration, and it prompted investigations and congressional hearings. Seen as largely ad hoc, we argue that theory predicts a more systematic decision-making process. We investigate this empirically andfind, consistent with literature on principal-agent theories and bureaucracy, that performance on easily monitored metrics and adverse-selection concerns predict thefirings. We explore the implications of these findings for efforts to centralize decision-making in the Department of Justice and to exert political control over US attorneys. As the federal government’s principal litigators, US attorneys (USAs) occupy a central role in the justice system. Housed within the Department of Justice (DOJ), USAs serve at the pleasure of the president. However, despite possessing this removal authority, presidents have seemingly exercised it rarely in the middle of an administration (e.g., Scott 2007). Given that history and a common suspicion that its actions represented dissatisfaction with USA behavior in a handful of specific, politically charged cases, the Bush administration’s dismissal of nine USAs in 2006 spawned significant controversy and investigation. 1 Critics alleged the removals resulted from improper considerations, while the administration proffered justifications stressing seemingly objective factors that reflected concerns We thank Trent Davis for his helpful reactions to earlier drafts of this article and gratefully acknowledge the institutional support that enabled us to obtain the data needed to complete this project. A previous version of this work was presented at the 2017 Southern Political Science Association Conference in New Orleans, where we received valuable feedback. Contact the corresponding author, Brett Curry, at bcurry@georgiasouthern.edu. 1. The nine were Todd Graves (Western District of Missouri), H. E. “Bud” Cummins (Eastern District of Arkansas), David Iglesias (New Mexico), Daniel Bogden (Nevada), Paul Charlton (Arizona), Carol Lam (Southern District of California), Margaret Chiara (Western District of Michigan), John McKay (Western District of Washington), and Kevin Ryan (Northern District of California). Journal of Law and Courts (Fall 2018) © 2018 by the Law and Courts Organized Section of the American Political Science Association. All rights reserved. 2164-6570/2018/0602-0007$10.00. Electronically published May 25, 2018. This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

about the performance of the individuals involved (Office of the Inspector General 2008, 325; Richman 2009). Although the Office of the Inspector General (OIG) issued a report on the matter re flecting qualitative conclusions based largely on accounts from those involved, to our knowledge the OIG did not analyze any performance data on the USAs. Here, we seek to test quantitatively the conventional notion that thefirings were motivated by specific cases and idiosyncratic concerns (e.g., Moynihan and Roberts 2010) against the possibility that they were consistent with a general, more extensive monitoring by DOJ principals of USAs—their agents across the country (Perry 1998). Evidence of such monitoring would speak to ongoing debates about the centralization of decisionmaking in the DOJ (Herz and Devins 2003, 1345–46; Spaulding 2011), a process that is important on its own terms but is of particular significance to the work of USAs (Suthers 2008, 112–13; Richman 2009). To be clear, in weighing the possibility that the administration’s behavior was consistent with the idea of more systematic monitoring, we do not mean to suggest that the dismissals were well executed by the administration or even politically defensible. As has been widely recognized, the firings themselves were carried out with the “reckless disregard of a choreographed train wreck ” (Iglesias 2008, 112; Suthers 2008; Ingle and Symons 2012). But we hardly think this forecloses the possibility that the administration’s underlying objectives may have been more coherent. We begin by providing a brief overview of USAs, including aspects of their duties, the mechanics of their appointment, and their term length. We then reflect on the firings themselves and, in that discussion, provide relevant descriptions of the USAs involved. Next, we consider the applicability of principal-agent theory to this study. We describe the data, the measurement of key variables, and our modeling strategy before presenting the results. Those results show that several factors systematically affected the propensity of USAs to be targeted for dismissal, and we assess the implications of them. Then, in a supplementary examination, we look at the possibility that factors consistent with principalagent theory may have been influential in other administrations. This subsidiary analysis suggests that the 2006 episode, while admittedly unique in many respects, is less aberrant than traditionally thought. Finally, we consider directions for additional research. Our findings speak to central questions of administration as they relate to law and courts. Who controls the federal government’s prosecutorial bureaucracy? What implications do trends related to centralization have for that oversight? Just how potent is the desire to avoid adverse selection —in essence, picking agents whose goals do not coincide with the principal’s—when it comes to choosing USAs? A BRIEF DESCRIPTION OF THE HISTORY AND RESPONSIBILITIES OF US ATTORNEYS The position of US attorney was established by the Judiciary Act of 1789, which provided each of the 13 original federal judicial districts an of ficer to prosecute violations 380

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

of civil and criminal law and defend the United States in civil actions. Today, there are 93 such officers, each assigned to a federal judicial district.2 The president nominates these individuals, and their 4-year term commences upon confirmation by the US Senate. Like lower court judges, they are subject to a blue-slip process that allows home-state senators to block their nominations. This selection process has been referred to as being “the greatest influence on how U.S. Attorneys perform their duties,” and, as such, scholars have noted the importance of avoiding adverse selection at this stage as a way to maximize control over these agents (Nelson and Ostrander 2016, 211). As political appointments, it is customary for USAs to submit their resignations upon a change in administration — although the incoming president may ask them to remain. Thus, while they garnered headlines, decisions of incoming presidents such as Clinton ( Johnston 1993) and Trump (Savage and Haberman 2017) to replace holdover USAs are qualitatively different from the events we examine here. Aside from departures that coincide with a change in administration or the completion of a 4-year term, both the large number of USAs as well as their often-ambitious career goals can create periodic vacancies. In fact, between 1981 and 2006, at least 54 presidentially appointed USAs left office before the completion of their term—a plurality of those became federal judges, and a number of others embarked on lucrative private sector careers or campaigns for office (Scott 2007). As a consequence, the procedure for appointing interim USAs carries significant potential implications. Between 1898 and 1986, these interim appointments were made by the respective federal district courts. In 1986, the attorney general gained authorization to make interim appointments, but those appointments were limited to 120 days. If a vacancy continued past the 120-day limit, the district court would then make an interim appointment. That changed in 2006 when Congress reauthorized the USA Patriot Act. It removed the 120-day limit and eliminated the district court ’s role in appointing interim USAs altogether, giving the attorney general plenary authority to make appointments that would remain in effect until the Senate ’s confirmation of a “permanent” successor. This was the process in place when the dismissals considered in this study occurred, although discussions in the administration about replacing USAs had begun approximately a year before this statutory change (Nelson and Ostrander 2016). 3 USAs have historically operated with significant discretion—particularly when dealing with criminal matters, which comprise the bulk of their time and attention (Perry 1998). As described by Richman (2009, 2090), “Federal criminal law … has long been characterized by extraordinarily broad substantive statutes and enforced by a relatively small bureaucracy that can pick and choose among possible targets. ” Although USAs 2. Guam and the Northern Mariana Islands share one US attorney (Scott 2007). 3. In the aftermath of the episode recounted here, Congress, in early 2007, restored the interim system that had been in place between 1986 and 2006. An Empirical Investigation of Attorney Firings

3 8 1 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

are also responsible for overseeing civil cases, they care most about crime (Perry 1998, 139). Others (e.g., Eisenstein 1978; Glaeser, Kessler, and Piehl 2000; Lochner 2002; Whitford 2002) similarly note that USAs possess signi ficant discretion and should be expected to exercise it strategically. Although prosecutors consider “winnability” in bringing cases, setting priorities with respect to categories of cases is primarily a political choice—not a legal one (Perry 1998; Lochner 2002). And, to the extent that the association between USAs and the executive branch can be viewed as a principal-agent relationship, it is ultimately the USA’s responsibility to carry out the priorities of her political principal within legal bounds. Given this, centralization or oversight by the DOJ ( “Main Justice”) is a focus of studies of USA decision-making. It is generally recognized that such oversight occurs to “impose policy decisions that may be driven by substantive concerns, a desire for uniformity, a desire to minimize abuses of discretion, or politics,” rather than because of “superior trial experience or judgment” (Green and Zacharias 2008, 202). In other words, although Main Justice may be uninterested in routine criminal matters, it does set the parameters of national policy (Iglesias 2008, 162). Green and Zacharias (2008, 200) conclude that centralization is most sensible when the DOJ emphasizes the importance of particular issues, and there is a wealth of evidence to suggest it has become increasingly willing to do so (e.g., Suthers 2008, 112–15; Richman 2009; Ingle and Symons 2012). Countering these centripetal forces is the notion that USAs must be responsive to conditions in their districts. “The entire institutional framework of federal prosecution is premised on the normative assumption that USAs should be and will be responsive to their local communities ” (Apollonio, Lochner, and Heddens 2013, 247). Richman (2009) identifies two pragmatic reasons why Congress should want to limit the centralization of authority in Main Justice. First, preserving the autonomy of local USAs limits the president’s agenda-setting ability. Second, Richman suggests that decentralization can give members personal leverage in their respective states or districts. At the same time, USAs are subordinate to the attorney general’s authority, and there are ample reasons why they should be. There is an expectation of standardization in the nation’s basic law enforcement goals (Eisenstein 2007, 222). Such political responsiveness is a critical element of democratic accountability. The balance between centralization and decentralization has oscillated over time, by administration, and across case categories. The amount of centralization in Main Justice was especially strong during the George W. Bush administration (e.g., Driscoll 2007) until the backlash that resulted from the attorney firings of 2006; this marked a contrast with the DOJ under Attorney General Reno during the 1990s (Richman 2009). With the important exception of Eisenstein (2007), scholars have generally concluded that the forces of centralization have grown in recent years (e.g., Lochner 2002). This conventional wisdom is consistent with broader trends toward centralization in the bureaucracy, which have occurred as the White House has sought ways to limit agency discretion (Wilson 1989, 264). As former USA John Suthers (2008, 112) notes, “More and more of U.S. attorney discretion has become subject to the scrutiny of ‘Main Justice.’” 382

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

POLITICS, ACCOUNTABILITY , OR IMPUDENCE? THE US ATTORNEY FIRINGS OF 2006 The prospect of targeting a number of USAs for dismissalfirst gained traction in the weeks after President Bush’s 2004 reelection; the dismissals themselves would occur in several stages, but not until 2006. Seven of the nine were forced out on December 7, 2006, and resignations had been requested from two others earlier in the year, although accounts frequently treat the nine removals as a package (Of fice of the Inspector General 2008).4 The firings were a drastic tactic: “Although the nine fired U.S. Attorneys had served more than four years, theirfirings were both unprecedented and counter to a longstanding and widespread consensus that they could expect to continue to serve until either they left voluntarily or resigned at the end of an administration” (Eisenstein 2007, 234). It was only after Attorney General Alberto Gonzalez and White House Counselor Harriet Miers joined the administration that the proposal to dismiss a number of USAs, generally credited to the attorney general’s chief of staff Kyle Sampson, became the subject of meaningful discussion. In March 2005, Sampson produced an initial list of 14 USAs he recommended for removal. By December 2006, seven lists had been generated within the administration. At the time of the dismissals, 28 names had appeared on at least one of the lists. In many of the nine dismissals, the OIG’s report determined that the administration’s proffered reasons for the firings—all of which were styled in relation to performance — were conflicting and inconsistent. The first USA directed to resign was Todd Graves of the Western District of Missouri; Michael Battle, director of the Executive Of fice for U.S. Attorneys, called Graves and instructed him to resign on January 24, 2006 (Of fice of the Inspector General 2008, 108). H. E. “Bud” Cummins of Arkansas’s Eastern District was asked to resign in June 2006. The seven remaining USAs were informed of their dismissals on December 7, 2006. In each case, the administration set forth alleged reasons for the firings that were unique to the individual at hand. For instance, Deputy Attorney General Paul McNulty described David Iglesias, the USA for New Mexico, as being “underperforming” and an “absentee landlord” who relied excessively on a subordinate to run his of fice. Despite numerous unanswered questions, the inspector general’s report concluded Iglesias ’s removal was precipitated by complaints from Senator Pete Domenici and Representative Heather Wilson about Iglesias ’s handling of investigations into voter fraud and public corruption. Indeed, Iglesias relayed conversations he had had with Wilson and Domenici, his former mentor, before a congressional subcommittee in 2007. 5 Iglesias in particular came to exemplify the idea that political considerations may have exerted improper in fluence in the firings. 4. The 2008 OIG report contains a vivid and comprehensive account of the firings; this section relies heavily on that account, and, unless otherwise indicated, the information contained herein is derived from the report. 5. Iglesias said that, after Senator Domenici “hung up on [me] after learning [I] would not seek indictments in a criminal investigation of Democrats before the 2006 election,” he “had a sick feeling in the pit of [his] stomach … . Six weeks later I got the call that I had to move on ” (Zagorin 2007). An Empirical Investigation of Attorney Firings

3 8 3 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

The stated justifications for the removal of USA Daniel Bogden of Nevada centered on his purported resistance to the attorney general’s priorities—essentially the prioritization (or the lack thereof ) of obscenity prosecutions. An alleged lack of energy and leadership in an office “where we have the right to expect excellence and aggressive prosecution in a number of priority areas” was also mentioned as a justification for his firing (Office of the Inspector General 2008, 202). Paul Charlton, the USA for Arizona, was allegedly removed for being “insubordinate as to the Department ’s way of doing business ” on the death penalty and the taping of interrogations (221). John McKay, headquartered in Seattle, was said by DOJ officials to have been removed for performance-related issues. More cynical observers felt the firing was rooted in McKay ’s failure to prosecute alleged voter fraud in the 2004 Washington gubernatorial election, although the inspector general unearthed no direct evidence to substantiate that allegation. The firing of Kevin Ryan as USA for the Northern District of California represents an instance in which both administration officials and the OIG were in agreement —Ryan had been removed in response to his of fice management. Significant concerns about his management had been re flected in two independently administered EARS (Evaluation and Review Staff ) reviews that were“blistering”; according to theLos Angeles Times,R y a n was an administration loyalist, but his administrative and management failings were “well documented,” and he was added to the list because of them (Dolan 2007). Margaret Chiara of Michigan ’s Western District and the Southern District of California’s Carol Lam were the other USAs to be dismissed. The inspector general’s report concluded that Chiara had been fired for office management issues and noted that concern about her management style was long-standing and had, in fact, increased over time. Lam’s firing, by contrast, allegedly grew out of Main Justice ’s dissatisfaction with her office’s prosecution of firearms and immigration cases. Some believed that Lam’s firing had been precipitated by her of fice’s investigation of former Republican congressman Randy “Duke” Cunningham on allegations of corruption, but the OIG report could not corroborate that supposition. Nevertheless, the inspector general faulted the DOJ for failing to examine Lam ’s justifications for her comparatively low prosecution rates in firearms and immigration cases —namely, that her of fice emphasized the quality of such prosecution over quantity. Testifying before Congress, Principal Associate Deputy Attorney General William Moschella made clear that Main Justice was uninterested in any context or justi fication for Lam ’s “low” numbers: “Quite frankly, her gun prosecution numbers are at the bottom of the list. She only beat out Guam and the Virgin Islands in that area ” (Washington Post 2007). In her response, Lam lamented “the unfortunate bean-counting approach … that the Department came to employ. ” 6 6. The text of Ms. Lam ’s response no longer seems to be available on the House Judiciary Committee’s website, but a transcription can be found on a blog titled Tinsel Wing at https://tinselwing .wordpress.com/carol-lams-written-responses-to-the-senate-judiciary-committee/. 384

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

However, beneath this focus on the dismissed USAs lurks another question—to what extent did being targeted for dismissal result from issues at the heart of principal-agent theory: concerns about the selection of personnel and the monitoring of their performance? The initiative for removing the USAs seems to have come first from Harriet Miers, White House counsel, who wanted to dismiss all the USAs. Sampson and others believed this to be “unwise” and instead used Miers’s idea “as a way to replace some weak U.S. Attorneys” (Office of the Inspector General 2008, 21). To this end, the initial list was styled as an informal performance review and contained recommendations that USAs be retained, be removed, or had not distinguished themselves either positively or negatively. Sampson circulated a second list in January 2006 and a third in April, which he later stated had been compiled “based on my review of the evaluations of their offices conducted by EOUSA [Executive Office for United States Attorneys] and my interviews with officials in the Office of the Attorney General, Office of the Deputy Attorney General, and the Criminal Division ” (28). The plan gained real traction in the fall of 2006, and four additional lists were disseminated between September and December. Sampson described the compilation of these lists as the product of “consensus” whereby he aggregated names and managed the process, although the OIG report disputed aspects of that account. PRINCIPAL-AGENT THEORY , POLITICAL APPOINTMENTS, AND MONITORING We have suggested that the relationship between the DOJ and USAsfits that of principal and agent (see also Perry 1998); although there is no single template capturing the wide range of such relationships (e.g., Waterman and Meier 1998; Gailmard 2014), principalagent theory has been recognized as a way to understand bureaucracies through the lens of rational choice (e.g., Moe 1984; McCubbins, Noll, and Weingast 1987; Wood and Waterman 1994). Central to that theory are the concepts of moral hazard and adverse selection (Moe 1984; Brehm and Gates 1997; Gailmard 2009), both of which are based on assumptions about human nature and relate to the lack of full and complete information that principals inevitably face (Perrow 1986). Adverse selection refers to the risk inherent in choosing agents who are not in alignment with a principal ’s goals, whereas moral hazard reflects the impossibility of observing an agent ’s every action. The most efficient way to ensure that principals and agents possess shared goals is for principals to appoint agents with care; this is of particular significance when political actors face agents with substantial levels of independence (e.g., Brehm and Gates 1997; Ennser-Jedenastik 2016). It is difficult to overstate the importance of avoiding this problem, as political appointment has been described as the single most effective —although imperfect—mechanism for controlling the bureaucracy (Wood and Waterman 1991, 801; 1994; but see Lewis 2008). USAs are no exception here (Nelson and Ostranger 2016). An Empirical Investigation of Attorney Firings

3 8 5 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Several factors might make the selection of a USA relatively (un)favorable to the administration. Given the norm of senatorial deference, in states in which the Bush administration faced Republican senators, the administration might have been more reluctant to replace a USA when it would face the potential wrath of a copartisan. 7 In fact, the association between USAs and senators from their home states has been described as “deliberately indistinct” (Iglesias 2008, 96). Conversely, where there was no relevant Republican senator, the administration would face no such constraint, and the USA would likely align more closely with the administration ’s priorities. Even given some apparent congressional apathy with respect to the USAs at this time, same-party senators were reluctant to concede authority over valuable patronage (Richman 2009, 2106). At the same time, the presence of same-party control is hardly dispositive here. For example, the effects of blue slips in the hands of opposition-party senators can be a significant constraint on an administration’s freedom of action. Furthermore, even where no same-party senator is present, other local political actors will typically play important roles. 8 Even so, with a constitutionally mandated voice in the con firmation process, senators are crucial actors, and we expect they will represent an important potential constraint on the executive. Once installed, Main Justice must then monitor USA behavior in order to limit shirking. In his typology of bureaucratic agencies, Wilson (1989, 167) describes “craft organizations” as those engaged in activities that, while processes are difficult to observe, yield results that are easy to evaluate. Like other agencies within the DOJ such as the Antitrust Division and the Federal Bureau of Investigation (Wilson 1989), USA offices closely approximate craft organizations. In such venues, metrics such as casefilings or criminal statistics are straightforward to obtain and assess, even as many of the specific activities and procedures are hidden from view. This, in turn, fosters bureaucratic accountability to the principal’s goals by providing a heuristic tool for engaging in oversight of ultimate results. We do not necessarily assert that the case statistics we employ here accurately re flect true notions of “effort,”“ performance,” or “mismanagement.” But we do believe there is ample evidence to suggest that the Bush DOJ viewed them as such. In addition to the Moschella testimony described above, scholars (e.g., Eisenstein 2007, 252), a num- 7. In one email, marked “please treat this as con fidential,” Kyle Sampson, writing to Harriet Miers, suggests that one of the reasons that Presidents Reagan and Clinton did not seek to dismiss USAs whom they had appointed after their statutory 4-year terms was that they did not wish to “receive the ‘advice’ of the home-state Senators” (Office of the Inspector General 2008, 45). The scare quotes might betray a frustration with the need to consult relevant home-state senators. Indeed, later in the email, Sampson mentions opposition to removal by home-state senators as an additional obstacle to consider in the removal process. The OIG report also indicates that Paula Silsby (Maine) and Thomas Marino (Middle District of Pennsylvania) were struck from the removal list “because both were believed to have the political support of their home-state Senators and the judgment was made not to risk a fight with the Senators over the proposed removals” (45). 8. We thank the anonymous reviewer who urged us to consider the potential role of both the blueslip process and local political actors in this decision calculus. 386

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

ber of the fired USAs themselves (Driscoll 2007; Iglesias 2008), and even some USA colleagues who were retained (e.g., Suthers 2008) have noted the Bush DOJ ’s “statistics happy” mentality. And, to underscore that point, scholars have adduced evidence that USAs sought to have their terrorism case statistics keep pace with the post-9/11 DOJ ’s prioritization of the issue, in part by reclassifying cases such as identity theft under the “terrorism” heading (see Chesney 2007). In the appendix, we display and discuss subject-specific models testing the possibility that Main Justice ’s oversight might have been focused on particular case categories. Simply put, it is dif ficult if not impossible to monitor all actions undertaken by bureaucratic agents. A host of factors accentuate this concern in the relationship between Main Justice and USAs. For one, the physical distance between Washington and USA offices makes continuous oversight cumbersome. Beyond that, despite being responsible for their offices, USAs cannot make every decision—indeed, there are different degrees of centralization over the decisions of line attorneys in the district offices (Green and Zacharias 2008, 200 –202). Since Congress gave assistant USAs civil service protections in 1988, these subordinates have grown increasingly careerist and more dif ficult for USAs to oversee (Lochner 2002; Suthers 2008, 112). In such an environment, a constant, police-patrol type of bureaucratic oversight (e.g., McCubbins and Schwartz 1984) could be appealing to principals at Main Justice. DATA, ANALYTICAL APPROACH, AND EXPECTATIONS Based on a large number of Freedom of Information Act requests to the DOJ, the data we use in this investigation come from the Transactional Record Access Clearinghouse (TRAC) at Syracuse University. These data include counts by district of the number of cases filed and completed, the number of convictions, the number of those convicted sent to prison, and the total prison term for all of those convicted and sentenced to prison. Our unit of analysis is a measure of the average for each of these categories in each district between fiscal year 2002 and fiscal year 2006 in each of the 89 districts. 9 Our dependent variable is whether a USA resigned in 2006 at the behest of the DOJ. The resignation variable is coded 1 if a USA resigned and 0 otherwise. Given the dichotomous nature of the dependent variable, we estimate logit models. Further, to account for heteroscedasticity, we estimate robust standard errors. We have six key independent variables. Four of these capture different parts of the prosecutorial process, all of which are counts (in hundreds) that we conceive of as measures of effort. Case filings (prosecutions filed ) is a measure that occurs at the earliest part of the process, and they arise in our data when an assistant USA files charges against a defendant. Case completions (prosecutions completed)f o l l o wfilings and occur when closed 9. We exclude the districts of Washington, DC, Guam/Northern Marianas, the Virgin Islands, and Puerto Rico, largely because selection pressures in these districts are not similar to those in the other 89 districts that have senators. An Empirical Investigation of Attorney Firings

3 8 7 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

either by conviction (typically plea bargain) or by failure to convict. Obviously,convictions measures the number of completed cases resulting in a conviction. The measure of the number of defendants sentenced to prison is a count of the number of defendants given a prison term because of conviction. We use these general measures of effort because they comport well with the notion, echoed multiple times in the OIG report and in later justifications from the administration, that the fired USAs did not manage their offices well. 10 Generally, then, we expect that higher numbers of case filings and the like will result in a decreased likelihood of resignation. There is some reason to expect that measures of effort from earlier in the prosecutorial process (filings and case completions) will be more predictive of resignation than will measures from later in the process (convictions and defendants sentenced to prison) because these earlier measures are less likely to be affected by other actors (see Wilson 1989) and might be viewed as more accurate indicators of USA effort by actors in Main Justice. Aside from effort, our second major concern centers on USA selection. Normally, adverse selection refers to the danger that a principal is unable to select agents with preferences that match its own. Here, we have taken the stylized position that the president (and Main Justice) is the principal —stylized because in reality the USAs have multiple principals that include Congress, specific senators, and other district-specific actors. This traditional approach to adverse selection suggests that the administration should be more likely to seek resignations from agents that it was most constrained in selecting initially. Here, we measure this initial constraint on selection as a dummy variable equal to 1 if there was at least one Republican senator in a state at the time the USA was selected to serve and 0 otherwise ( constrained appointment). This re flects the fact that, because norms of senatorial courtesy would be operative in such a situation, the administration’s preferences would be expected to accommodate those of its copartisan. Nevertheless, initial adverse selection may not be a powerful explanation for the decision to seek resignations because it ignores the fact that the administration, by December of 2006, had had at least 5 years, with the exception of Lam, for whom they had had only 4 years, to observe performance regardless of initial selection constraints. 11 Therefore, a savvy principal might be prospective rather than retrospective in thinking about selection. To capture this prospective position, we have created a variable coded 1 if no Republican senator is in place in 2006 (when the firing decisions were made) and 0 otherwise ( unconstrained replacement). Because of how we have coded them, we expect both measures to have positive coefficients. In the case of an appointment constraint, the presence of a Republican senator reduces the freedom of the administration. In the case of uncon- 10. Please note that we revisit the issue of whether subject-speci fic performance was indicative of resignation below in the results section. 11. Our results remain consistent even if we constrict our sample to include only fiscal years 2003, 2004, and 2005. 388

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

strained replacement, the absence of a Republican senator clears the path for the administration’s ideal candidates to ascend to the position. Further, comparatively, we expect that concerns related to prospective selection (unconstrained replacement) will be more powerful predictors of the USA firings than will retrospective measures, given other evidence of effort. In addition to the variables of primary theoretical interest, we include a number of control variables to account for district-speci fic differences that might affect our measures of total effort output. First, we include a count of the number of referrals from investigative agencies that a district receives ( referrals received, measured in thousands of cases)—higher levels of referrals should make filing more cases easier and so on. Second, districts that are more populous could potentially generate more cases for USAs (district population, measured in millions). Third, as a proxy for the size of a USA ’so ffice we include a count of the number of court-related work hours in a district ( court work hours, measured in thousands). 12 Court work hours represent the amount of time attorneys in a district spent in court or preparing for court. For instance, in fiscal year 2001, the majority of work hours were devoted to time actually arguing in court, while most of the additional hours were committed to witness preparation (Executive Of fice for United States Attorneys 2001). Of fice size obviously indicates the capacity of an of fice to produce case outputs, but there are additional theoretical reasons to include this measure. Eisenstein (1978) focused on district size as a key explanation of the independence of USA offices. It may be, then, that larger offices are less amenable to policy directives— therefore, USAs in them are a better target for Main Justice. But, since Eisenstein wrote in the late 1970s, the number of assistant USAs has skyrocketed, and it is plausible that assistant USAs are allocated to many of fices based, in part, on how amenable a particular USA is to administration policy. 13 There are diverging expectations about of fice size. If larger offices are more independent, they may be attractive candidates for reform. At the same time, if the administration has played favorites in allocating resources, then larger offices are those already preferred by Main Justice. USAs in districts that bordered Mexico were obviously at higher risk for being fired than others. This is clear from the simplest descriptions of the firings that three of the nine dismissed were in border districts (Arizona, New Mexico, and the Southern District 12. Because the DOJ stopped reporting the number of assistant USAs in 2003, we must rely on a proxy for office size rather than a direct count of the number of assistant USAs. We have data that begin in fiscal year 1986, and from 1986 through 2002 the correlation between the number of assistant USAs and court-related work hours is quite high ( r 5 .86). 13. There are limits to the expectation. While it may be the case that resources are allocated on the basis of a USA’s fidelity to administration priorities, it is also true that the resources provided to the largest offices (e.g., the Southern District of New York, the Northern District of Illinois, and the Eastern District of New York) are unlikely to vary as a function of such fealty. We are indebted to an anonymous reviewer for bringing this important point to our attention. An Empirical Investigation of Attorney Firings

3 8 9 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

of California).14 Therefore, it is important to account for the influence of bordering Mexico in the models. However, the introduction of a variable to denote the presence of a border district induces quasi-separation in the logit models that we present in table 2 (see, e.g., Zorn 2005). Thus, in the appendix, we present two approaches to help account for the absence of this variable and to ensure the robustness of the results that we present in the main text. First, we present a set of alternative models that exclude all border districts (table A1). Our results excluding border districts mirror those presented in table 2, with the noted exception that measures of effort after the filing of a case are no longer statistically significant. The models in table A1 use a penalized likelihood approach to account for the fact that the reduction in N from 89 to 84 induces quasi-separation in the models (Kosmidis and Firth 2009). Second, we use a penalized likelihood approach to “solve” the quasi-separation problem for models that include all districts and an indicator for whether a district is on the border. Table A3 contains results for these models. Here, too, our results are very similar to those presented in table 2. Therefore, we are confident that the need to exclude an indicator variable for border districts in the main models in table 2 does not affect our conclusions. In addition, the case mix from district to district is likely to vary considerably. To help control for these variations, we include two control variables focused on the types of referrals USAs receive. First, we include a measure that tracks the average rate at which USAs decline cases referred to them by investigative agencies ( declination rate). Second, we include a measure of the percentage of cases in each district that are what we call easy cases ( percentage of referrals easy cases), measured as the percentage of total cases that are immigration, narcotics, or weapons cases. 15 Together, these variables help control for the flow of cases into a USA ’so ffice. Table 1 displays descriptive statistics for the included variables. RESULTS Across the life cycle of a case, we estimated a series of regressions to track the effects of monitoring and adverse selection on the decision to seek a resignation. In all, we have regression models for the number of casesfiled, the number of cases completed, the number of convictions, and the number of those convicted sentenced to prison. Each model fits the data relatively well, with statistically significant Wald x 2 statistics and a high area under the Receiver Operating Curve (ROC) values. In general, models using effort early 14. Notably, the two border districts not caught up in the firing scandal were in Texas, Bush’s home state. The USA for the Western District of Texas at this time, Johnny Sutton, was previously the director of criminal justice policy for Bush when he was governor of the state. Sutton was chair of the Attorney General’s Advisory Committee of US Attorneys. He was also legendary University of Texas baseball coach Cliff Gustafson’sf a v o r i t ep l a y e r . 15. This characterization mirrors that of Richman (2009, 2100), who describes a “virtually inexh a u s t i b l es u p p l yo fr e l a t i v e l ye a s i l ym a d ec a s e s” in these areas; in contrast to terrorism and white-collar cases, in immigration, narcotics, and weapons cases, proving the defendant’s state of mind is straightforward. 390

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

in the process (filings) are more predictive of firing, judging by the values of the area under the ROC measures. The amount of constraint imposed on the administration ’s replacement decision (unconstrained replacement) is the single most important predictor in all models. Table 2 displays results for each of the models for all districts. Further, we summarize the critical results for our measures of effort at each stage infigure 1, including results from all districts (table 2) and nonborder districts (table A1). The left panel in figure 1 displays results across all districts for the effort variables, while the right panel displays the nonborder district results. The vertical lines represent 95% con fidence intervals. In figure 1, we map the effect of increasing case filings, completions, convictions, or prison sentences by shifting effort at each stage from the 10th percentile (low effort) to the 90th percentile (high effort) and show that such an increase significantly decreases the likelihood of being asked to resign by the Bush administration in 2006. Increasing the number of case filings from 276 cases (the 10th percentile) to 2,252 cases (the 90th percentile) decreases the probability of resignation by 29 [ 248, 211] percentage points (95% confidence intervals are in brackets throughout). More concretely, a USA with an average of 276 cases filed per year would, holding all other variables at their means, face a 1 in 3 chance of being asked to resign in 2006, while a USA averaging 2,252 cases per year has closer to a 1 in 12 chance of being asked to resign. The effect of increases in effort with respect to cases completed (26-percentage-point decrease [246, 27]) and convictions (29-percentage-point decrease [247, 211]) is similar to the effect for case filings. Increasing the number of defendants sentenced to prison results in a 9-percentage-point decrease [ 218, 1] in the likelihood a USA is fired. Note that the results for the number of defendants sentenced to prison are not statistically significant, lending some support to the notion that performance measures from later in the process are less influential in the evaluation process. Further, the right panel of figure 1, presenting results from nonborder districts, suggests that only the results for case filings

Table 1. Descriptive Statistics

Mean SD Min Max Asked to resign .10 .30 0 1 Prosecutions filed (hundreds) 13.69 25.92 1.27 185.33 Prosecutions completed (hundreds) 12.92 24.55 1.22 177.81 Convictions (hundreds) 11.27 23.01 1.02 168.14 Sentenced to prison (hundreds) 9.37 20.32 .80 14.37 Constrained appointment .64 .48 0 1 Unconstrained replacement .33 .47 0 1 Referrals received (thousands) 18.40 27.39 2.83 194.74 District population (millions) 3.31 2.74 .51 18.06 Court work hours (thousands) 8.22 7.43 1.46 39.02 Declination rate .46 .07 .31 .69 Percentage of referrals easy cases .40 .14 .18 .91 An Empirical Investigation of Attorney Firings

3 9 1 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table 2. Logit Regression Results

Model 1 Model 2 Model 3 Model 4 Coefficient Robust SE Coef ficient Robust SE Coef ficient Robust SE Coef ficient Robust SE Effort: Prosecutions filed (2) 2.36** .15 … … … … … … Prosecutions completed (2) … … 2.34** .16 … … … … Convictions (2) … … … … 2.44** .19 … … Sentenced to prison (2) … … … … … … 2.12 .10 Selection: Constrained appointment (1) .98 .98 .93 1.01 1.28 1.19 .85 1.03 Unconstrained replacement (1) 2.67** .96 2.48** .94 2.73** 1.05 2.12** 1.00 Controls: Referrals received .35** .15 .32** .15 .40** .17 .10 .07 District population 2.12 .22 2.13 .22 2.12 .23 .01 .19 Court work hours 2.05 .08 2.05 .09 .14 .11 2.05 .08 Declination rate 21.35 4.54 2.58 4.38 .66 4.45 2.88 4.63 Percentage of referrals easy cases 6.55* 3.74 6.09* 3.65 7.49 3.36 5.74 3.74 Constant 27.05 3.44 26.97 3.34 28.76 3.98 27.83 .36 Wald x 2 29.48 (.00) 24.29 (.00) 19.83 (.00) 14.41 (.07) Area under Receiver Operating Curve .82 .82 .84 .81 Note.—The dependent variable is asked to resign. N 5 89. * Significant at p < .10 (two-tailed). ** Significant at p < .05 (two-tailed). This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

are robust. Excluding border districts, case completion, convictions, and prison sentences no longer have statistically signi ficant effects. This lends further support to the notion that effort earlier in the prosecutorial process is more important for evaluations by the DOJ. We also expected selection concerns to figure prominently in the decision-making of Main Justice in the firings. Our results are interesting not only because they support this interpretation but also because they highlight the fact that the administration appeared prospective and not retrospective in its concerns. Our variable measuring whether the initial appointment of the USA in question was constrained by the need to defer to a sameparty senator is positive but never statistically significant. This suggests that the DOJ did not view initial selection constraints as relevant 4 years after the fact—a result that seems logical, given the presence of additional information on which to assess the USAs. Conversely, there is strong evidence to indicate that the administration paid attention to whether, in naming a replacement USA, it would be subject to the constraints of senatorial courtesy. Our estimates for the unconstrained replacement dummy variable are consistently positive and statistically significant across models. In results averaging across the four stages, USAs who could be replaced without deference to a Republican senator were 19 [2, 36] percentage points more likely to be asked to resign. A USA in a district in which there was no Republican senator to consult was about six times as likely to be asked to resign (30% possibility) as one in a district with at least one Republican senator present (5% possibility). Note that the effect of replacement constraints is equal to a sub-

Figure 1. Change In Probability Of Resignation With Increasing Effort, By Stage

An Empirical Investigation of Attorney Firings

3 9 3 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

stantial increase in prosecutorial effort in casefilings, suggesting that this factor was as important as case statistics in the administration’s decision-making.16 Two of the control variables are significant in a number of models. First, the number of referrals a USA’so ffice receives correlates with the likelihood of dismissal, when controlling for effort. Put differently, if the number of cases referred to a USA is high but effort is relatively low, then the models suggest that this is a signal to the DOJ to look more closely at performance. Similarly, in two models in table 2, as the percentage of referrals that are easy cases increases, again controlling for effort, so too does the likelihood of dismissal. The logic here is likely similar to that for overall referral numbers —more of these easyto-prosecute cases should result in more prosecutions. In addition to these models using measures of the overall level of effort at a given stage of the process, we also created subject-specific measures based on issue areas that the OIG report highlighted as important to the Bush administration. These include immigration and weapons cases from the justifications offered for the firing of Carol Lam, corruption prosecutions offered as justifications and speculation in the firing of David Iglesias (and possibly John McKay), and the terrorism cases long thought to be the central concern of the Bush-era DOJ (Suthers 2008; Richman 2009). We estimated a series of regressions similar to those reported in table 2 (reported in table A2 in the appendix). In most of them, the subject-specific measures of effort are not significant. The one exception to this general trend is with respect to terrorism prosecutions. This is not surprising from the perspective of Main Justice, as we know that the Bush administration and the DOJ in particular obsessed about terrorism cases, to the extent of perhaps encouraging USA of fices to lump more routine immigration and identity-theft cases into the category (Chesney 2007). Our results suggest that an increase in the number of terrorism prosecutions from 0 (the 10th percentile in the data) to 21 (the 90th percentile) reduced the likelihood of being asked to resign by 20 percentage points—an effect smaller than the increase in total prosecutions. Notably, our results for overall effort remain unchanged if we include terrorism filings in a model with total case filings, suggesting that effort in terrorism cases is not a substitute for measures of more general effort. There is no indication that there was a systematic evaluation of USA performance on corruption, immigration, or weapons prosecutions. 16. Ironically, many of the vacancies created by the firings remained unfilled by permanent USAs for the remainder of the Bush presidency. Only four of the nine positions would be filled by a Senateconfirmed USA prior to 2009, and those took approximately a year to install. Interestingly, all but one of the USAs to win Senate con firmation were those from states where there was at least one Republican senator—the exception occurred in California’s Northern District, where the Senate confirmed a veteran USA to lead the of fice. None of the other positions from states with two Democratic senators were filled until the Obama administration. To us, this suggests at least two conclusions. First, it illustrates that yet another aspect of the administration’s calculus in the episode proved faulty. Second, as the majority party, Democrats were perhaps willing to allow GOP colleagues to exercise senatorial courtesy. However, Democratic ire about the firings and the administration’s evident desire to circumvent senatorial courtesy appear to have steeled them to oppose virtually all Bush nominations in their states. 394

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

DISCUSSION AND SUPPLEMENTARY ANALYSIS We have found evidence to suggest that the Bush administration ’s decision to fire nine USAs in 2006 was more systematic than most have previously thought. Speci fically, our results are consistent with the idea that the administration used two criteria predicted to be important in principal-agent theories: avoiding adverse selection and monitoring effort. Notably, we also investigated whether there was anything systematic about the creation of the lists and found no evidence that there was. 17 Placement on a list was clearly not indicative of performance, nor was the potential of having to accommodate a Republican senator an influence in the construction of the lists. Given that there were multiple lists generated by numerous sources over multiple years, it is not surprising to us that the listing process appears unsystematic. Nonetheless, the far more consequential decision— which USAs to let go—seems to have been a potentially rational reaction to the difficulties of managing bureaucratic agents. Those USAs ultimatelyfired were not necessarily performing inadequately, nor was it necessarily appropriate for the administration tofire them. Additionally, USAs are actors uniquely situated between national and local forces, and their behavior must respond to both. A level of nuance is needed in assessing USA behavior, because there can be dramatic differences in crime demographics and other characteristics across districts. This requires some level of flexibility and discussion with state and local of ficials (Suthers 2008, 113). But centralization and top-line monitoring are typically inimical to this sort of nuance. Therefore, we do not want our argument about the rationality of the administration’s behavior to be confused with a suggestion that it was normatively correct. In addition, we investigated whether the DOJ appeared to treat various indicators of effort differently according to how directly USAs controlled the effort indicator monitored. To this end, we found that effort earlier in the prosecutorial process—particularly the decision to file a case—was a more potent predictor of the decision tofire a USA than was the later-stage indicator of the number of defendants sentenced to prison. Evidence in favor of filings being a particularly important indicator of effort also comes from the relative robustness of that measure to the exclusion of border districts in model 1a. We think this is important evidence of some sophistication in the way that Main Justice looked in on its agents. This evidence of monitoring couples nicely with the evidence suggesting that Main Justice was primarily concerned with the potential of adverse selection and not how it might have affected agent selection in the past. Given evidence of performance, it is logical that the administration would not heed past adverse-selection issues. Our results are consistent with the exercise of rational behavior by the DOJ and the White House, but, as is generally the case with observational data, those results cannot 17. This investigation included regressions mirroring those reported in table 2 and uses Heckman selection models, which indicated no need to simultaneously estimate inclusion on a list and ultimate removal. An Empirical Investigation of Attorney Firings

3 9 5 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

definitively establish that such rationality actually guided the decision-making —that is, while we may be observing rational behavior, it is also possible that we have simply observed the appearance of rational behavior. At the same time, the clarity of our results and evidence suggesting that the process’s initial stages attempted to identify the“weakest” USAs (Office of the Inspector General 2008, 21) buttresses our confidence in these empirical results. Although these results are important in their own right, we are also interested in whether similar considerations might have been operative in other administrations. This is a dif ficult question, because the firings in 2006 were unusually brazen and therefore unusually easy to observe. As explained earlier, all nine of the USAs removed in 2006 had served more than a full 4-year term. However, beyond the prominence they garnered, their departures were inconsistent with a clear norm: USAs in those circumstances would anticipate they could remain in place until the administration ’s end if they desired to do so. To determine whether the twin concerns of naming a successor and effort monitoring—as predicted by our theory and evidenced in our results —may have also factored into attempts to control USAs in other administrations, we must first develop a plausible definition of when we observe a resignation that occurred under some pressure from the administration. This is not straightforward, so we wish to emphasize how tentative our conclusions on this front must remain. We proceed in two steps. First, we offer ad efinition for when a USA might have been asked to leave office, rather than leaving by choice. Then, using that definition, we analyze whether this group had characteristics as suggested by our theory. Defining those USAs who may have left of fice before they wished to do so requires that we gather additional data on what a USA who left of fice did subsequently and on the timing of when USAs entered and exited of fice. We undertook this somewhat time-intensive endeavor for all USAs in the Clinton administration and the Obama administration. Focusing on these administrations offers us a number of advantages. First, both Clinton and Obama served two full terms, allowing us to make comparisons to the George W. Bush administration. Second, they are temporally proximate to the Bush administration. Third, they offer us the chance to see whether the theory we explain in the article is limited by partisanship. We will use the term“potentially dismissed” to define those USAs who may have been asked to leave, rather than leaving on their own volition. We readily admit that we have no way of knowing their actual reason for leaving —indeed, this is the value of looking so closely at the 2006 firings—but we think it is useful to probe for any indications that other administrations may have had similar considerations in mind. The timing of a USA’s exit is crucial to our de finition. From the complete list of USAs serving in the Clinton and Obama administrations, approximately 270 in total, we focus on those who leave office after having served at least 4 years but before normal politics suggests a resignation. A normal political resignation occurs near the end of the president’s 8-year term. So, we excluded from our list of potentially dismissed USAs those who left of fice the year be- 396

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

fore the relevant impending elections, so either after 1999 or after 2015, because those leaving so close to a known political transition were likely leaving because of partisan politics. This is a somewhat arbitrary cutoff, because it is possible that some USAs would like to stay on longer assuming a copartisan wins the elections in 2000 or 2016. So, for instance, a USA appointed to office in 1993 who then left office in 1998 would be on our list, but one appointed in 1994 who left in 1996 would not. This set of criteria allows us to make a tighter comparison to those USAs fired in 2006, all of whom served at least their initial 4-year terms. Were we to code the date of departure more leniently, our list would undoubtedly become overinclusive, causing a tripling in the number of USAs whose exits we deem to be potentially suspicious (and this strikes us as facially invalid). From the list of USAs who left within our specified time frame, we then searched for information on what job the USA took after leaving of fice. 18 We did this by searching a host of secondary sources, including newspaper articles, obituaries, law- firm press releases, Wikipedia pages, and the like. We then kept on our list only those USAs who went into private practice. 19 This means we exclude from our potentially dismissed list those USAs who left office because of appointment to a federal office—either as a federal judge or to work in an executive agency. Such a subsequent appointment indicates anything but displeasure with the work of a USA. We also exclude a small number of USAs who leave to run for elective of fice (including as a state judge). This process leaves us with 22 USAs of interest across the two administrations. To reiterate, a number of the USAs on our list undoubtedly left by choice to earn more money in the private sector, and we cannot distinguish those who did so from those who left less willingly. The point is that our list is likely to contain those who left unwillingly, if any did so. Comparing the USAs on our list of potential dismissals to the remaining group of USAs in both administrations might shed light on whether the difficulty of replacement and effort monitoring are factors that administrations other than that of George W. Bush considered. With respect to what we have called unconstrained replacement—the ability of the executive to replace a USA without having to consult a copartisan senator —there is some suggestive evidence that those on the potentially dismissed list were easier to replace than other USAs. We find that 35% of the USAs on the dismissal list could be replaced without consulting a senator, compared to 30% of the remaining USAs in each administration. The evidence for effort, measured in terms of prosecutions filed, is also suggestive. Here, we find that USAs on the dismissal list filed an average of 269 fewer cases per year than did other USAs in the Clinton administration (the difference is statistically significant at p 5 .02) and 999 fewer cases in the Obama administration (the 18. We hasten to add that our cutoff in 2015 allows suf ficient time for those USAs leaving office to find subsequent employment and for that employment to be reported and available to us. 19. This corresponds with what we know about the subsequent career patterns of the USAs fired in 2006. For the eight USAs on whom we could gather information (excluding Chiara), seven went into private practice immediately after exiting. Iglesias is the lone outlier—he became a law professor. An Empirical Investigation of Attorney Firings

3 9 7 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

difference is not quite statistically significant at p 5 .11).20 In sum, both the replacement measure and the effort measure point in the same direction: those on our potentially dismissed list tend to be those that principal-agent theory suggests might be pressured to leave office. Although necessarily tentative, this suggests to us that, while the visibility of the Bush administration’s actions was indeed sui generis, the actual operation of them may not be. CONCLUSION In many ways, the 2006 firings punctuated a trend toward centralization in Main Justice during the Bush presidency (Eisenstein 2007), a trend in which the 9/11 attacks and the administration’s response to them played an integral role (e.g., Suthers 2008; Richman 2009). Of course, the desire for centralization and the monitoring it necessitates typically take other, less drastic forms. Indeed, in the supplementary discussion above, we found some preliminary, if tentative, evidence that such considerations have been relevant in other circumstances. Further, this study has several implications for broader investigations of these topics in relation to USAs and Main Justice ’s oversight of them. For one, it suggests national principals will be most likely to monitor markers of USA effort early in the prosecutorial process—especially at thefilings and completion stages. We suspect the waxing of some issues (namely, immigration) and the waning of others (such as narcotics) as targets of national interest may be reflected in other aspects of USA behavior and its responsiveness to central actors and their priorities, as suggested by the Bush administration’s apparently close monitoring of terrorism filings. The availability of longitudinal data across time and administrations will facilitate such investigations, and we plan to undertake our share. Finally, as described above, we believe it is important to consider the possible relevance of nonnational factors in structuring USA behavior. We should study these local forces, both on their own terms and with an eye toward how they condition the attention of USAs to national priorities. Given their centrality in federal criminal justice policy, and the expanding federalization of crime, USAs are central actors in understanding how the state applies force against its own citizens. Our study examines one extreme episode in the attempt to exert political control over these actors, but much more work remains in explaining subtler and more regular forms of influence.

Appendix

Here we provide supporting evidence for a number of claims made in the article. First, we present evidence from the regressions that exclude border districts in table A1. As in the regression results presented in table 2, clear patterns of monitoring and adverse-selection concerns emerge. Note that the results from the models in table A1 are reported in figure 1 in the main text. The regressions in table A1 are penalized to account for the problem of quasi-separation caused by reducing the number of districts in the regressions 20. The combined difference across both administrations is 315 fewer cases ( p 5 .20). 398

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table A1. Logit Regression Results, No Border Districts, Penalized Likelihood

Model 1a Model 2a Model 3a Model 4a Effort: Prosecutions filed (2) 2.35** … … … (.19) Prosecutions completed (2) … 2.28 … … (.19) Convictions (2) … … 2.44 … (.28) Sentenced to prison (2) … … … 2.29 (.37) Selection: Constrained appointment (1) 1.54 1.63 1.59 1.41 (.93) (1.08) (1.07) (.98) Unconstrained replacement (1) 2.56** 2.38** 2.28** 2.28** (.91) (.94) (.99) (.94) Controls: Referrals received .39** .32** .38** .19 (.17) (.16) (.19) (.13) District population .24 .27 .26 .46** (.21) (.21) (.22) (.23) Court work hours 2.28** 2.26** 2.31** 2.27** (.14) (.13) (.14) (.13) Declination rate 2.49 2.06 1.94 4.89 (5.07) (4.79) (4.88) (4.55) Percentage of referrals easy cases 23.69 23.78 22.34 2.80 (3.56) (3.79) (3.96) (4.49) Constant 23.84 24.99 25.33 26.77 (3.09) (3.23) (3.35) (3.28) Note.—The dependent variable is asked to resign. Robust standard errors in parentheses. N 5 84. * Significant at p < .10 (two-tailed). ** Significant at p < .05 (two-tailed). from 89 to 84. Quasi-separation occurs when a variable is a perfect predictor for the outcome, causing issues with how the model generates coefficient estimates for that variable (Zorn 2005). To handle this problem, we turn to the recommendation of Kosmidis and Firth (2009) to estimate models with penalized log likelihoods. In addition, we estimated a series of regressions to account for the possibility that Main Justice cared more about effort in speci fic, highly salient areas of law rather than overall effort. We present results of regressions in specific areas of law in table A2. These measures also tend to induce quasi-separation, so we use a penalized likelihood approach here as well. Note that cases are sorted into issue areas by the coding assigned to the lead charge by the assistant USAs who initially process a case, according to TRAC. We focus on those issue areas that the firings highlighted: corruption, immigration, and weapons, as well as the central issue for the DOJ at this time, terrorism. As noted in the main text, we find some evidence that subject-specific performance was important in the decision-making process, as more terrorism prosecutions reduce the likelihood of being asked to resign. An Empirical Investigation of Attorney Firings

3 9 9 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table A2. Subject-Specific Logit Regressions, Penalized Likelihood

Model 5a Model 6a Model 7a Model 8a Effort: Corruption prosecutions filed (2) 2.06 … … … (.05) Immigration prosecutionsfiled (2) … 2.00 … … (.00) Terrorism prosecutions filed (2) … … 2.15** … (.06) Weapons prosecutions filed (2) … … … .00 (.01) Selection: Constrained appointment (1) .46 .64 .30 .49 (.83) (.88) (.87) (.85) Unconstrained replacement (1) 1.76** 1.97** 1.79** 1.87** (.85) (.91) (.89) (.93) Controls: Referrals received .01 .04 .05** .01 (.01) (.03) (.02) (.01) District population .16 .08 .35* 2.00 (.19) (.18) (.20) (.02) Court work hours 2.02 2.06 2.08 2.00 (.07) (.08) (.07) (.06) Declination rate 1.16 2.39 1.20 3.11 (4.45) (4.33) (4.55) (4.26) Percentage of referrals easy cases 2.83 4.02 2.16 3.20 (2.89) (2.89) (2.84) (2.73) Constant 24.86 26.43 24.97 26.41 (2.76) (2.79) (2.76) (2.80) Note.—The dependent variable is asked to resign. Robust standard errors in parentheses. N 5 89. * Significant at p < .10 (two-tailed). ** Significant at p < .05 (two-tailed). No other subject seems to have mattered for Main Justice. To be as concise as possible, we present results only from models using casefilings because our results from later stages of the prosecutorial process are not substantially different from those we present here. We also use penalized regression for these subject-specific models to avoid problems caused by quasi-separation. As noted in the main text, there is evidence that more terrorism prosecutions had a potentially prophylactic effect for USAs, reducing the probability of being asked to resign by 20 [227, 215] percentage points (moving from the 10th percentile to the 90th percentile in the data). This evidence coincides with the emphasis that the Bush DOJ placed on terrorism after 9/11. Finally, to check the robustness of the results we present in table 2, we reestimate the models but include the border district variable, accounting for quasi-separation using the penalized likelihood technique described above. Results are presented in table A3, with the only notable difference being that the effect of case completions is somewhat larger than in the model presented in table 2. 400

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table A3. Penalized Likelihood Including Border Districts

Model 9a Model 10a Model 11a Model 12a Effort: Prosecutions filed (2) 2.49** … … … (.16) Prosecutions completed (2) … 2.77** … … (.25) Convictions (2) … … 2.48** … (.17) Sentenced to prison (2) … … … 2.31** (.12) Selection: Constrained appointment (1) 1.23 1.06 1.50 1.44 (.83) (.99) (1.03) (.99) Unconstrained replacement (1) 2.52** 2.09** 2.26** 2.31** (.84) (.95) (.96) (.95) Controls: Referrals received .44** .66** .40** .19** (.14) (.22) (.14) (.08) District population .18 .36 .29 .45** (.20) (.23) (.21) (.21) Court work hours 2.21** 2.52** 2.31** 2.28* (.10) (.20) (.13) (.12) Declination rate 23.49 24.16 1.58 4.64 (4.41) (4.73) (4.21) (4.41) Percentage of referrals easy cases 23.45 21.67 21.92 21.34 (3.38) (3.77) (3.83) (4.12) Border district 11.44** 19.68** 11.56** 13.27** (3.51) (6.06) (3.97) (4.04) Constant 22.13 21.75 25.19 26.47 (3.88) (2.99) (3.08) (3.29) Note.—The dependent variable is asked to resign. Robust standard errors in parentheses. N 5 89. * Significant at p < .10 (two-tailed). ** Significant at p < .05 (two-tailed).

References

Apollonio, Dorie, Todd Lochner, and Myriah Heddens. 2013.“Immigration and Prosecutorial Discretion.” California Journal of Politics and Policy 5:232–51. Brehm, John, and Scott Gates. 1997. Working, Shirking, and Sabotage: Bureaucratic Response to a Democratic Public. Ann Arbor: University of Michigan Press. Chesney, Robert M. 2007. “Federal Prosecution of Terrorism-Related Offenses: Conviction and Sentencing Data in Light of the ‘Soft-Sentence’ and ‘Data-Reliability’ Critiques.” Lewis and Clark Law Review 11:851–901. Dolan, Maura. 2007. “Bush Loyalist Was Added to Purge List Late.” Los Angeles Times, March 22. Driscoll, Sharon. 2007. “The Gonzales DOJ Firings: A Conversation with Former U.S. Attorney Carol Lam. ” Stanford Lawyer , no. 77. https://law.stanford.edu/stanford-lawyer/articles/qna -legal-matters-with-carol-lam/. Eisenstein, James. 1978.Counsel for the United States: U.S. Attorneys in the Political and Legal Systems. Baltimore: Johns Hopkins University Press. An Empirical Investigation of Attorney Firings

4 0 1 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

———. 2007. “The U.S. Attorney Firings of 2006: Main Justice’s Centralization Efforts in Historical Context.” Seattle University Law Review 31:219–63. Ennser-Jedenastik, Laurenz. 2016. “The Politicization of Regulatory Agencies: Between Partisan Influence and Formal Independence. ” Journal of Public Administration Research and Theory 26:507–18. Executive Office for United States Attorneys. 2001. United States Attorneys’ Annual Statistical Report. Washington, DC: Department of Justice. https://www.justice.gov/sites/default/files/usao /legacy/2009/06/08/01statrpt.pdf. Gailmard, Sean. 2009.“Multiple Principals and Oversight of Bureaucratic Policy-Making.” Journal of Theoretical Politics 21:161–86. ———.2 0 1 4 .“Accountability and Principal-Agent Theory.” In The Oxford Handbook of Public Accountability, ed. Mark Bovens, Robert E. Goodin, and Thomas Schillemans. Oxford: Oxford University Press. Glaeser, E. L., D. P. Kessler, and A. Morrison Piehl. 2000. “What Do Prosecutors Maximize? An Analysis of the Federalization of Drug Crimes.” American Law and Economic Review2:259–90. Green, Bruce A., and Fred C. Zacharias. 2008.“‘The U.S. Attorneys Scandal’ and the Allocation of Prosecutorial Power.” Ohio State Law Journal 69:187–254. Herz, Michael, and Neal Devins. 2003. “The Consequences of DOJ Control of Litigation Authority on Agency Programs.” Administrative Law Review 52:1345–75. Iglesias, David. 2008. In Justice: Inside the Scandal That Rocked the Bush Administration. Hoboken, NJ: Wiley. Ingle, Bob, and Michael Symons. 2012. Chris Christie: The Inside Story of His Rise to Power .N e w York: St. Martin’s. Johnston, David. 1993.“Attorney General Seeks Resignations from Prosecutors.” Washington Post, March 24. Kosmidis, Ioannis, and David Firth. 2009.“Bias Reduction in Exponential Family Nonlinear Models.” Biometrika 96:793–804. Lewis, David E. 2008. The Politics of Presidential Appointments: Political Control and Bureaucratic Performance. Princeton, NJ: Princeton University Press. Lochner, Todd. 2002.“Strategic Behavior and Prosecutorial Agenda-Setting in United States Attorneys’ Offices: The Role of U.S. Attorneys and Their Assistants.” Justice System Journal 23:271– 94. McCubbins, Mathew D., Roger G. Noll, and Barry R. Weingast. 1987.“Administrative Procedures as Instruments of Political Control.” Journal of Law, Economics, and Organization 3:243–77. McCubbins, Mathew D., and Thomas Schwartz. 1984.“Congressional Oversight Overlooked: Police Patrols versus Fire Alarms.” American Journal of Political Science 28:165–79. Moe, Terry M. 1984. “The New Economics of Organization.” American Journal of Political Science 28:739–77. Moynihan, Donald P., and Alasdair S. Roberts. 2010.“The Triumph of Loyalty over Competence: The Bush Administration and the Exhaustion of the Politicized Presidency.” Public Administration Review 70:572–81. Nelson, Michael J., and Ian Ostrander. 2016.“Keeping Appointments: The Politics of Confirming U.S. Attorneys.” Justice System Journal 37:211–31. Office of the Inspector General. 2008. An Investigation into the Removal of Nine U.S. Attorneys in 2006. Washington, DC: US Department of Justice. https://oig.justice.gov/special/s0809a/final .pdf. Perrow, Charles. 1986. Complex Organizations: A Critical Essay. 3rd ed. New York: McGraw-Hill. 402

JOURNAL OF LAW AND COURTS

F ALL 2018 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Perry, H. W. 1998. “United States Attorneys—Whom Shall They Serve?” Law and Contemporary Problems 61:129–48. Richman, Daniel. 2009. “Political Control of Federal Prosecutions: Looking Back and Looking Forward.” Duke Law Journal 58:2087–124. Savage, Charlie, and Maggie Haberman. 2017. “Trump Abruptly Orders 46 Obama-Era Prosecutors to Resign.” New York Times, March 10. Scott, Kevin M. 2007.U.S. Attorneys Who Have Served Less Than Full Four-Year Terms, 1981–2006. Washington, DC: Congressional Research Service. Spaulding, Norman W. 2011.“Independence and Experimentalism in the Department of Justice.” Stanford Law Review 63:409–46. Suthers, John W. 2008. No Higher Calling, No Greater Responsibility: A Prosecutor Makes His Case. Golden, CO: Fulcrum. Washington Post.2 0 0 7 .“House Judiciary Committee, Subcommittee on Commercial and Administrative Law Holds a Hearing on the Dismissal of U.S. Attorneys. ” March 6. Waterman, Richard W., and Kenneth J. Meier. 1998. “Principal-Agent Models: An Expansion?” Journal of Public Administration Research and Theory 8:173–202. Whitford, Andrew B. 2002. “Bureaucratic Discretion, Agency Structure, and Democratic Responsiveness: The Case of the United States Attorneys.” Journal of Public Administration Research and Theory 12:3–27. Wilson, James Q. 1989.Bureaucracy: What Government Agencies Do and Why They Do It. New York: Basic. Wood, B. Dan, and Richard W. Waterman. 1991. “The Dynamics of Political Control of the Bureaucracy.” American Political Science Review 85:801–28. ———. 1994. Bureaucratic Dynamics: The Role of Bureaucracy in a Democracy. Boulder, CO: Westview. Zagorin, Adam. 2007. “Why Were These U.S. Attorneys Fired?” Time,M a r c h7 . Zorn, Christopher. 2005.“A Solution to Separation in Binary Response Models.” Political Analysis 13:157–70. An Empirical Investigation of Attorney Firings

4 0 3 This content downloaded from 141.165.060.165 on August 22, 2018 05:56:08 AM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).