Estimating the Political Ideologies of Appointed Public Bureaucrats. An Application to the Senate Confirmation of Presidential Nominees Adam Bonica Department of Political Science Stanford University Jowei Chen Department of Political Science University of Michigan Tim Johnson Center for Governance and Public Policy Research Atkinson Graduate School of Management Willamette University ABSTRACT This paper uses political campaign contributions to estimate public bureaucrats’ political ideologies. Bureaucrat ideal points estimated via our method vary across time, compare meaningfully with ideological estimates in other branches of government, cover employees across a wide range of agencies, yield insight into intra-agency ideological variation, and are produced automatically from public records. To demonstrate our method, we estimate the political ideologies of appointed administrators in the U.S. federal government. We then use those estimates to test hypotheses about how U.S. presidents strategically manage the process of appointing individuals to federal bureaucratic posts requiring Senate confirmation. Public bureaucracies figure prominently in modern states (Skowronek 1982; Wilson 1989; Moe 1997; Carpenter 2001, 2005). Not only has the capacity, purpose, and authority of public bureaucracies expanded over the past century, but so too has the political clout of those government organizations (Lowi 1969; Moe 1997). As noted by various scholars, public bureaucrats have used reputation (Carpenter 2001, 2010), expertise (Gailmard and Patty 2007), delegation (Eskridge and Ferejohn 1992), and electoral influence (Moe 1997; Anzia 2011) to cultivate opportunities to shape policy. Yet, despite agreement about the public bureaucracy’s growth and influence, it remains an open question whether or not public bureaucrats use their strategic location in the policy process for personal advantage. While some scholars propose that public bureaucrats use their positions for personal gain (Niskanen 1975; Moe and Miller 1983; Moe 2007; Anzia 2011), or to shift policy in their preferred direction (Epstein and O’Halloran 1999; Carpenter 2001; Huber and Shipan 2002), others hypothesize that bureaucrats implement policy faithfully in order to foster a reputation for expertise and neutral competence (Carpenter 2010; Huber 2007). Moreover, prominent scholars hold that politicians constrain public bureaucrats’ efforts to deviate from policy orders. According to those scholars, politicians use oversight (McCubbins and Schwartz 1984; Weingast and Moran 1983; Aberbach 1990), the details of laws (Epstein and O’Halloran 1999; Huber and Shipan 2002; MacDonald 2010), procedural rules (McCubbins et al. 1987), organizational structure (Moe 1989; Hammond and Miller 1985; Hammond 1986; Lewis 2003), and staffing (Lewis 2008) to prevent public bureaucrats from pursuing policies that significantly depart from the prescriptions of elected officials. Still others argue that politicians can gain by letting bureaucrats pursue their personal political agendas: wayward public bureaucrats provide politicians a means to “shift the responsibility” for policy woes (Fiorina 1 1985) and broader delegation of policy-making provides incentives for bureaucrats to develop technical expertise (Gailmard and Patty 2007). In sum, the theoretical literature on public bureaucracy has offered numerous hypotheses about when, why, and how public bureaucrats influence policy, and how politicians seek to constrain that influence. Empirically testing these hypotheses, however, presents several challenges. The above hypotheses mainly derive from spatial models of politics. Spatial models posit that observable political behavior results from strategic actions that take into account the policy preferences of individual political actors. As a result, testing spatial models requires measures of political actors’ ideologies. While legislators, executives, and judges routinely engage in publicly observable activities that reveal their policy preferences, the same cannot be said for public bureaucrats (Clinton et al. 2012). As a result, the development of quantitative measures of bureaucratic ideology has lagged behind the measurement of other political actors’ ideologies. However, over the past decade, scholars have developed increasingly innovative methods to measure bureaucratic ideology. These methods rely on a variety of data sources, including the past political decisions of administrative officials (Nixon 2004), expert opinions (Clinton and Lewis 2008), government employee surveys (Clinton et al. 2012), the political history of agencies (Lewis 2003, 2007, 2008), and the public statements of top officials (Bertelli and Grose 2009). Although these methods have contributed greatly to efforts to empirically test the theoretical literature on bureaucracy, challenges remain. Current methods provide little insight into ideological variation within agencies and changes in agency political preferences across time. Moreover, they base their estimates on the stated values of public bureaucrats, which may contain “cheap talk,” as opposed to bureaucrats’ costly actions. Furthermore, state of the art 2 methods rely on labor intensive data collection, thus making it difficult to update ideal point estimates regularly. Together, these limitations continue to impede empirical studies of public bureaucracy. This paper seeks to address those challenges by presenting a new method for estimating bureaucratic political ideologies. The proposed method tracks—from 1979 to the present—the campaign contributions of federal employees and individuals nominated to federal bureaucratic posts. It then uses those data to produce ideological measures that (i) yield individual-level ideal points for cabinet members, executive appointees, and many career bureaucrats, (ii) provide insight into ideological variation within agencies, (iii) vary across time, (iv) can be compared with the ideal points of actors in other branches of government, (v) cover a large number of agencies, and (vi) require little effort to update.1 The paper also illustrates how these estimates can be employed to test hypotheses concerning the public bureaucracy. We enlist estimates of individual bureaucrats’ political ideologies to study presidential appointments to the public bureaucracy that require senate confirmation. By studying the influence of previously overlooked institutional factors (e.g., committees, as well as the opportunity to avoid confirmation processes via unilateral, non-recess appointment), this investigation extends previous analyses of the politics underlying Senate confirmation of bureaucratic positions (Havrilesky and Gildea 1992; Barrow et al. 1996; McCarty and Razaghian 1999; Nixon 2004; Corley 2006; Black et al. 2007; see, for a review, Aberbach and Rockman 2009). Put broadly, we find that institutions at each step of the Senate confirmation process influence the ideological distribution of successfully appointed nominees. First, we find that 1 This method also produces estimates of the typical political ideology of an agency’s bureaucrats (i.e. an aggregate measure of agency ideology), which we report in the appendix. 3 nominees for bureaucratic posts requiring Senate confirmation are more moderate than individuals appointed to posts that do not require senate confirmation. This finding illustrates that bureaucrats’ ideologies are not merely a product of presidents’ personal policy preferences; instead, the political leanings of bureaucrats also reflect the demands of Senators that wield their institutional power to block presidential nominees. Second, our investigation shows that the likelihood of Senate confirmation declines with the ideological distance between nominees and relevant committee chairs. Third, and finally, we find that the ideal points of recess appointees lean more strongly in the direction of the president’s ideal point than do nominees that face Senate confirmation. It thus appears that presidents take advantage of Senate recess periods to push through ideologically extreme nominees. In sum, political actors at each institutional stage of the Senate confirmation process influence the ideological distribution of successful appointments. In the following sections, we discuss our ideal point estimation methodology and substantive findings. We begin by reviewing past efforts to estimate bureaucratic ideology and by explaining how our method adds to those efforts. Next, we present our method, report results, and establish measure validity. With our new ideal point estimates in hand, we then illustrate their use by testing hypotheses about the Senate confirmation process. Following that illustration, we discuss how our estimates might be utilized in future studies. Previous Efforts to Measure Bureaucratic Ideology Despite their centrality in theoretical models of bureaucratic politics, researchers have struggled to produce reliable measures of public bureaucrats’ political ideologies. As articulated by Clinton et al. (2012), the mere conceptualization of bureaucratic ideology poses challenges. For instance, should bureaucrats’ political values be measured at the individual, bureau, or 4 agency level? What actions or declarations reveal bureaucrats’ genuine policy preferences? To what extent do the ideologies of bureaucratic line workers matter, relative to the values of top administrative officials? These questions make the estimation of bureaucratic ideology difficult. Nonetheless, past scholars have made significant progress in addressing those questions and, in fact, the literature on public bureaucracy now offers several methods to measure the political ideologies of government agencies and their employees. Early efforts to measure the political leanings of federal agencies relied on collections of nominal variables that captured insight into an agency’s historical roots, organizational structure, or policy mission. Scholars using this approach, for instance, might create a binary indicator that takes a value of unity when an agency was created during a Republican presidential administration and a value of nil when an agency was created during a Democratic presidential administration (Lewis 2007, 2008). Lewis (2003) and Howell and Lewis (2002) have provided a theoretical rationale for this approach; they provide evidence that the history and design of agencies leave a lasting mark on those agencies’ political orientation. As a result, indicators of agency history and structure offer insight into agency ideology. However, the historical factors that nominal variables use to capture agency ideology rarely contain the detail needed to convey ideological variation across time or within an agency. Also, since those variables place agencies into broad categories (e.g., created under a Democratic versus Republican president, or created under divided versus unified government) they do not readily lend themselves to testing theoretical predictions about bureaucratic politics that take into account the degree of ideological divergence between bureaucrats and other political actors (Ferejohn and Shipan 1990; Epstein and O’Halloran 1999; Huber and Shipan 2002). 5 Subsequent efforts to model bureaucratic ideology have addressed the latter problem by drawing on methods for determining the political ideologies of legislators, presidents, judges, and political interest groups (Poole 1998; Clinton et al. 2004; McCarty et al. 2006). Nixon (2004), for instance, identifies high-level bureaucratic officials who previously served in Congress and, then, uses estimates of those officials’ political ideologies—while in Congress— to model their political values as bureaucrats. This approach offers insight into the political orientation of a select group of upper-level bureaucratic actors. However, like nominal variables that serve as a stand-in for bureaucratic ideology (Lewis 2003, 2007, 2008), the estimates Nixon (2004) puts forward are one-shot, cover a small number of bureaucrats, and cannot be used in applications seeking to understand ideological variation within a given agency. A similar problem arises with expert surveys. Clinton and Lewis (2008) interviewed experts about the political orientation of 82 federal agencies and, with those survey data, estimated aggregate measures of agency ideology using item response models. While this method offered insight into the political orientation of a large number of agencies, it did not reveal agencies’ cross-time ideological variation, nor did it produce estimates that could be compared with ideological measures capturing the ideology of other branches of government. Subsequent survey methods produced bureaucratic ideal points that can be compared across institutions (viz. Clinton et al. 2012), but they remain constrained in the temporal period that they cover. Bertelli and Grose (2007, 2009, 2011) develop a method of estimating bureaucratic ideal points that vary across time and meaningfully compare across government branches. Their method uses the public pronouncements of top administrative officials to measure agency ideology. Specifically, Bertelli and Grose (2009) devise a coding scheme to categorize a vast 6 number of public testimonies made by public bureaucrats between 1991 and 2004. They construct roll call voting records for cabinet secretaries by treating announced positions as votes in support or against legislation and jointly recover ideal points for secretaries and members of Congress. While their approach is novel, they struggle to establish the face-validity of their measures. According to their results, cabinet members are unimodally distributed, with a mean near the center of the policy space, and there exists substantial partisan overlap; together, these properties suggest that appointees are considerably more moderate than members of Congress. In addition, their method produces questionable estimates for several prominent cabinet members.2 The lack of face-validity fuels concerns about inferring ideology from bureaucrats’ public statements, as opposed to their costly actions (cf. Snyder and Weingast 2000; Nixon 2004), which makes it difficult to distinguish cheap talk and strategic posturing from actual preferences. An ambitious study by Clinton et al. (2012) is the most recent attempt to measure executive agency ideology. They administered a survey to thousands of Bush-era executives to gather information on those executives’ policy preferences. The survey questionnaire included a series of policy questions carefully selected to match-up with specific Congressional roll call votes; these matches serve as bridging observations. Subsequently, they utilize a recently developed roll call scaling methodology designed to estimate ideal points for voters and legislators in a common space (Bafumi and Herron 2010; Jesse 2009, 2011; Tausanovitch and Warshaw 2012). While they report a respectable response rate of 33 percent, Clinton et al. (2012) note that high ranking appointees were less likely to respond to the survey. As a result, their sample includes only 181 political appointees, representing a small fraction of the total population of appointed officials. A major limitation of their approach is that measures are 2 For example, Republican appointees John Ashcroft and Lynn Martin locate to the left of the majority of Democratic appointees, including Janet Reno, whereas Clinton appointees Bill Richardson and Andrew Cuomo locate to the right of a majority of Republicans. 7 recovered for a single period (2005-2006). In addition, administering large-scale surveys is resource intensive and, thus, difficult to update into the future. Also, it is not clear how such an approach could be extended into the past. More problematic, however, is that anonymity requirements preclude reporting individual level estimates for survey respondents. This impedes studies from focusing on specific decision-makers such as leaders within bureaus or agencies. In sum, while previously proposed methods offer helpful tools for studying the political ideology of bureaucratic agencies and officials, limitations remain. No study has developed a measure of bureaucratic ideology that simultaneously varies across time, varies within agencies (including across individuals), compares with ideological estimates for political actors in other branches of government, and can be readily automated. In the next section of this essay, we present a method that meets these challenges. Our method captures individual-level bureaucrats’ political ideologies, which allows future researchers to pinpoint the ideologies of bureaucratic actors responsible for decision-making in a given agency. In addition, as agency composition is not static, these individual-level estimates can offer insight into how the ideological composition of the agency changes across both time and levels of the agency’s hierarchy. Moreover, our measures can be compared with the ideal point estimates of actors in other political institutions. Estimating the Political Ideology of Bureaucrats Using Campaign Contributions We construct measures of bureaucratic ideology from ideal points estimated using campaign finance data. This method enlists a dataset consisting of over 85 million contributions made during state and federal elections held between 1979 and 2012. Unlike roll-call scaling methods, which are typically confined to a single voting body, donors frequently make contributions to candidates across institutions and levels of the political hierarchy. Thus, the 8 method simultaneously estimates ideal points for a variety of elected officials and the millions of individuals who fund their campaigns. The theoretical rationale for our measurement approach involves a simple spatial model. Put simply, this spatial model posits that ideological considerations determine—at least in large part—contribution behavior. Specifically, it assumes that contributors prefer ideologically proximate candidates to those who are more distant and distribute funds according to their evaluations of candidate ideology. Formally, a contributor , considers the entire set of candidates soliciting donations, denoted as . Contributor selects the candidate,   j 1,...,J , whose ideal point, ,maximizes the contributor’s objective function: [ ( ( ) ( ) ( ) )] 1.1 where is the sum that contributor has available to donate, is candidate ’s ideal point, is contributor ’s ideal point, ( ) is a payoff function capturing the combined instrumental and expressive utility derived from donating, and ( ) is the cost function. Together, ( ) and ( ) signal contributor ’s propensity to contribute. Under this theoretical framework, contributor donates to candidate, , whose ideal point, , is closest to the contributor's own ideal point, . Ideally, we would model the determinants of contributions in a manner similar to Bonica (2012a), who uses an IRT count model to estimate ideal points from PAC contributions while controlling for non-ideological factors such as electoral competitiveness, incumbency status, or committee assignments. Unfortunately, computational costs preclude fitting a similar model to the much larger data set of individual campaign contributions, which we use here. We instead base our measures of bureaucratic ideology on Bonica’s (2012b) common space CFscores (“Campaign Finance Scores”) methodology designed to scale the much larger dataset of 9