Off-site Oversight: Who Polices the Administrative State?

Transcription

1 Off-site Oversight: Who Polices the Administrative State? Kenneth Lowande Washington University in St. Louis July 9, 2017 Abstract Scholarship on oversight of the bureaucracy typically conceives of Congress as a unitary actor. In contrast, most oversight is conducted by individual legislators informally, who forward fire-alarms directly to agencies. I develop a theory of individualized oversight and analyze its consequences for bureaucratic accountability. Leveraging the correspondence logs of 13 bureaucratic agencies from the 110 th 113 th Congress, I find ideological disagreement has a negligible impact on oversight and committee membership increases the probability of oversight. Whereas past research suggests Congress collective action problems contribute to suboptimal oversight, my findings imply individualized oversight may lead to persistent coordination failures. Postdoctoral Research Associate, Department of Political Science. lowande@wustl.edu Prepared for presentation at the Political Economy and Public Law Conference, University of Southern California, June 15-16, I thank Randy Calvert and Rachel Potter for helpful suggestions and Craig Volden for coining the title. A previous version was presented at the 2017 annual meeting of the Midwest Political Science Association, Chicago, IL.

2 In April of 2014, CNN reported that bureaucrats in the Veterans Health Administration had deliberately fabricated appointment wait times and that at least 40 veterans had died while waiting to receive medical care. 1 Response from Congress was dramatic: the House and Senate oversight committees convened special hearings, subpoenaed VA officials, demanded resignations, funded a criminal investigation, and enacted sweeping legislative reforms. But surprisingly, for years in advance of this scandal, dozens of members of Congress pulled fire-alarms they received and forwarded concerns about long wait times directly to the VA. Policy failure occurred despite legislative oversight. Monitoring unelected officials implementing public policy is a chief concern in democratic government. By overseeing that process, elected officials aim to prevent shirking, corruption, performance failures, and policy drift in bureaucracies. Not surprisingly, questions about the efficacy of this oversight has generated a vast body of theoretical and empirical research. 2 I advance this work by collecting records of over 25,000 contacts between members of Congress and bureaucratic agencies between the 110 th 113 th Congress ( ). These new data reveal activity long-theorized about, but rarely observed. In a widely cited article on bureaucratic accountability, McCubbins and Schwartz (1984) argue members of Congress use administrative procedures to set up fire alarms for agency misbehavior, which allow them to avoid performing costly police-patrols (i.e. formal oversight hearings). 3 Scholars have analyzed published oversight hearings (Aberbach 1990; McGrath 2013; MacDonald and McGrath N.d.; Kriner and Schickler 2016), whereas this paper documents pulled congressional fire-alarms. To organize the data, I develop a theory of individualized oversight, in which legislators independently audit bureaucratic agencies through informal inquiries. Departing from work that reduces Congress to a unitary actor, my model asks which legislators are more or less likely to perform oversight. Answering this question is important because this individualized oversight is, as McCubbins and Schwartz suggest, far more ubiquitous than formal oversight hearings. Moreover, the independent audits of legislators aggregate to the collective oversight of Congress so it 1 Bronstein, Scott and Drew Griffin. A fatal wait: Veterans languish and die on a VA hospital s secret list, CNN, April 23, For reviews of this work, see Moe (2012) and Gailmard and Patty (2012). 3 As of this writing, the paper has been cited 3,060 times. Google Scholar URL: 1

3 is important to understand whether that oversight is systematically biased. The theory suggests several key predictions. The first is that ideology will have a negligible impact on the propensity to perform oversight. Legislators ideologically proximate to agencies may be less likely have policy complaints ex ante, but they have an incentive to balance the auditing behavior of their distal colleagues. This logic is analogous to Austen-Smith and Wright (1994), in which groups lobby to counteract the work of the opposition. A second prediction is that legislators with relevant oversight jurisdiction in committees will be more likely to perform audits. Here, the outsized influence of these legislators renders them more willing to pay the opportunity cost of action. This notion is somewhat consistent with canonical accounts of oversight, which suggest that most is performed by congressional committees. The key caveat, however, is that this oversight is conducted off-site and is motivated by the underlying power their position affords them, not routinized police patrols performed in hearings. I find robust evidence for both of the above predictions. The data suggest ideological distance has a small or negligible impact on oversight, whereas committee status increases contact probability by around 10 percentage points. Finally, in-keeping with McCubbins and Schwartz perspective on fire-alarms, the theory implies that this kind of oversight has suboptimal general welfare consequences, though it is efficient from the perspective of legislators. In other words, the frequency of these audits should not be confused with a demonstration of effectiveness. Past work has suggested that overlapping committee jurisdictions and, in general, the multiplicity of legislative principals generates collective action problems that lead to suboptimal oversight (Gailmard 2009; Clinton, Lewis, and Selin 2014). This study empirically demonstrates that the collective oversight activity of Congress is far more diffuse suggesting both that those collective action problems are more severe and may be indicative of an additional problem: the failure to coordinate. As the VA example above suggests, when individual legislators pursue service to constituents, they sometimes fail to detect systemic policy failures. Thus, I compliment the above analyses with case studies of oversight of the VA and Nuclear Regulatory Commission (NRC) and close with a discussion of broader accountability problems. 2

4 Congressional Oversight and Bureaucratic Accountability Though empirical studies have been limited by investigating patterns in formal hearings, oversight has been given numerous theoretical treatments. But generally, theories of oversight treat Congress as a unitary actor for expository purposes. This is because scholars are typically interested in a set of questions about political control of the bureaucracy intended to analyze whether policy will deviate from the congressional median. Examples of such questions include does the absence of overt oversight imply a runaway bureaucracy? (Weingast and Moran 1983; McCubbins, Noll, and Weingast 1987; Calvert, McCubbins, and Weingast 1989) and do fire-alarms communicate credible information to legislators? (Lupia and McCubbins 1994; Epstein and O Halloran 1995). Even work that emphasizes bureaucracies have multiple principals often takes Congress to be unitary the other principals are the President, interest groups, and the Judiciary (e.g. Moe 1987; Hammond and Knott 1996). This means that past work does not generally address the research question of who in Congress performs oversight. The exception is studies tend to argue, generally, that oversight is conducted by-in-large in committee (e.g. Ogul and Rockman 1990), with more recent work investigating the consequences of overlapping jurisdictions (Gailmard 2009; Clinton, Lewis, and Selin 2014; Rezaee, Wood, and Gailmard 2015). The data I present suggests, however, that the broader question warrants attention. In all agencies in the sample, informal contact is far more frequent than hearing activity. For example, the most frequently contacted agency in the data, the Federal Communications Commission (FCC), was contacted about 2,800 times per Congress. In contrast, the FCC was mentioned in around 49 hearings per Congress from the 110 th 113 th. Moreover, many of these hearings were related to executive appointments, appropriations, and miscellaneous legislative measures. In a typical Congress, the House and Senate spend 3 days on official oversight of the FCC. Thus, in large agencies with frequent contact, the off-site oversight is orders of magnitude more prevalent. In agencies with narrower statutory scope, the difference is less striking but still clear. The Nuclear Regulatory Commission (NRC) in mentioned in 20.3 hearings, on average, from the 110 th 113 th Congress. It received about 30 informal contacts per Congress during the same period. But again, only a handful of those official hearings were dedicated to oversight of agency activities. Thus, alongside longstanding motivations related bureaucratic accountability, the basic empirical 3

5 fact that most oversight is conducted by individual legislators demands a theory of individualized oversight. Who Polices the Administrative State? To answer the question posed by this section, I present a simple theory of oversight in which legislators perform audits of bureaucratic agencies. The key assumptions are that legislators receive signals from constituents, which are analogous to complaints about agency action (i.e. fire-alarms ). Legislators decide whether for forward these complaints, thereby performing an audit that implies some effort or opportunity cost. Agencies, after receiving these alarms, then decide whether or not to adjust their behavior. Legislators vary on two relevant dimensions: their ideological proximity and institutional power in relation to the agency. Below, I outline the logic of the two predictions I take to the data and justify each assumption. In the stylized oversight scenario, agencies make policy choices while implementing governing statutes. 4 Their choices are sometimes thought of as falling within a discretionary policymaking window. Congress has difficulty observing policy choices because they are generally less informed. They have two broad categories of tactics to mitigate this information asymmetry (McCubbins and Schwartz 1984). The first is to perform routinized or active oversight of agency policymaking which takes place in the formal oversight schedules of committees in the House and Senate. The second is more sporadic and passive. Legislators wait for constituents to communicate concerns that expose agencies policy deviation. The second is more prevalent because it allows legislators to avoid wasting time performing active audits that do not reveal drift while giving legislators the opportunity to claim credit twice: both for policy enactment and for righting the ship. Thus, it is appropriate to restrict focus to the second tactic as constituents pull alarms and legislators decide whether or not to communicate them. Here, I use the term constituent generally, to refer to both to voters or organizations in-district, as well as public interest groups with a national presence. Past work has interpreted fire-alarm -pulling constituents both ways. 5 Moreover, around half of legislator correspondence is on behalf of individual constituents who are rarely 4 For a review of work that analyzes this scenario, see Moe (2012). 5 See Epstein and O Halloran (1995) for an interest group example, whereas (McCubbins and Schwartz 1984) repeatedly write of this kind of oversight as narrow constituency service. 4

6 one of the well-known interest groups who contribute to salient political debates. So this general definition in consistent with the data. This raises the question of what messages legislators will receive from constituents. A general concern is that reliance on better-informed constituents might create the same kind of agency problems legislators face when overseeing the bureaucracy. So it is important to highlight that, in general, legislators receive fire-alarms from constituents with similar preferences. This is consistent with the fact that constituents tend to select like-minded legislators, but also with classic results in signaling theory (Crawford and Sobel 1982). Lupia and McCubbins (1994) show that fire-alarms are informative when the preferences of legislators and constituents are similar. In this case, pulling a fire-alarm implies some effort on the part of constituents. Legislators, however, cannot trust that the information provided by distant constituents is genuine. So they behave strategically with the expectation that distant constituents have incentives to pull false alarms. A related concern is how these alarms arise. Alarms are pulled because agencies deviate from the policy preferences of the constituent. Since agencies are presumed to use their discretion to pursue their preferences, this implies that alarms are pulled by constituents who are ideologically distal relative to the agency. 6 By implication, complaints tend to be communicated to distal legislators. Again, this is largely consistent with existing work on legislative-executive relations. Agencies are assumed to have their own preferences and the corresponding incentive to pull policy towards those preferences. Thus, distal legislators ought to hear more alarms from like-minded constituents and be motivated to communicate those alarms. Importantly, however, legislators act individually and are aware of the incentives of their colleagues. When legislators communicate fire-alarms, they lobby agencies for corrective action. Research on interest group influence often sees lobbying efforts as providing information about a pending policy decision. This notion may apply for some fire-alarms. But another, more general 6 Some might argue that agencies would strategically adjust their behavior to avoid complaints, and therefore, no fire-alarms should be communicated (Epstein and O Halloran 1995). I disagree for two reasons. First, in practice, it is unreasonable to assume that agencies can thread the needle to avoid all complaints. Most salient regulation, for example, provokes litigation despite the procedural due-diligence of the rule-making process. Second, the mere fact that the executive branch is contacted tens of thousands of times by legislators in each Congress seems to exclude the possibility of perfect, ex ante strategic adjustment. 5

7 argument is that alarm communication provides agencies feedback about the implications of decisions already made. Likewise, these alarms could provide managers with vital information about the performance of subordinates. So the information is valuable, and agencies know that legislators pay a price to communicate it. So it could be influential; agencies might make policy adjustments based on feedback from legislators. This incentivizes proximate legislators to counteract the communication of distal ones. 7 Friendly legislators cannot risk agency adjustments away from their preferred policies, so they lobby to prevent adjustments. Ultimately, this suggests that the relationship between ideological distance and informal oversight is not straightforward. More specifically, I argue it is likely to have a negligible impact on oversight frequency because the incentives above counter balance one another. As a legislator becomes more distant, they are more likely to receive complaints and have disagreements with a given agency. But if they were more friendly, they would have the incentive to balance the behavior of those who are not. Observationally, this should manifest in an effect near zero. Legislators, of course, do not simply vary in their policy preferences. If that were the case, then the restricted focus on Congress as the congressional median would be quite warranted. In effect, the argument above suggests that in the aggregate, the median position might prevail. For that reason, it is also important to incorporate the fact that legislators vary in terms of their relative influence within Congress. More specifically, some legislators hold offices and leadership positions that privilege their communication with agencies. Committee members, through their institutional position in Congress, are in a unique position to punish bureaucratic agencies. Calling and scheduling hearings falls within their direct purview. As McGrath (2013) notes, the hiring and supervision of committee staff provides them with the expertise required to perform oversight. Congress scholarship has placed committees at the center of the lawmaking process (e.g. Fenno 1973). 8 Moreover, legislators on relevant committees tend to be more specialized, and hence, better informed about agency activities. Therefore, a second empirical expectation is that committee membership ought to make legisla- 7 In Austen-Smith and Wright (1994), unfriendly groups counteract the lobbying of friendly ones. I have reversed this usage, but the logic and implications are the same. 8 More recent work by Volden and Wiseman (2014), for example, has shown that committee chairs tend to be more effective lawmakers; the bills they introduce are more likely to clear procedural hurdles and ultimately, become law. 6

8 tors more likely to perform audits. This could operate via two non-exclusive processes. The first is that the cost of performing an audit is worth paying, because legislators are aware of their outsized influence relative to rank-and-file legislators. The second is a function of the process of committee assignment. Committee member constituencies tend to have more vested interests in the activities of the relevant agencies, and thus, may be more informed of agency actions and more likely to raise alarms. This expectation is largely consistent with past assertions about the role of committees in oversight. But, importantly, the association between committees and oversight is driven by the incentives members of Congress to perform individual audits not conducting routinized oversight in the context of the committee. Congressional Correspondence To examine the previous expectations, I analyze records of correspondence between bureaucratic agencies and members of Congress. Since the vast majority of contact records are not publicly available, these records were collected through a series of Freedom of Information Act (FOIA) requests made October 2014 (Lowande 2016). Though agencies are required by law to respond to these inquiries, their responses do not always provide information usable for analysis. For that reason, the sample of agencies and time series is largely a function of availability. In this section, I discuss concerns related to this limitation, as well as describe the nature of this off-site contact. As Table 1 indicates, I analyze correspondence from the 110 th 113 th Congress in thirteen agencies with different time-series coverage by agency. The unbalanced panel, however, is a function of idiosyncratic FOIA and record-keeping procedures and, thus, may not be related to congressional contact or the legislator characteristics of interest. I discuss this issue in greater detail when describing the analysis of the unbalanced sample. A second, related issue is the non-random sub-sample of agencies included in the analysis. Broadly speaking, this analysis over-samples independent commissions and government corporations two types of organizations typically thought of as less constrained by elected principals. However, the structural features that render agencies more or less responsive to demands are invariant within agencies during the period analyzed. The results are unlikely to be influenced by such structural features given the modeling approach discussed in the next section. On the other hand, the broader generalizability of the findings should 7

9 be taken in careful context. I discuss this issue at length in the Supplementary Appendix. Oversight patterns may differ for a sample more reflective of the universe of executive branch agencies. For these reasons, I employ two different analyses that trade addressing some of these limitations for additional assumptions with the underlying belief that consistent results should allow more confidence in the basic empirical regularities. I defer discussion of these research designs until the section in which they are presented. Table 1 Correspondence Records Agency Equal Employment Opportunity Commission Federal Communications Commission Federal Deposit Insurance Corp. Federal Trade Commission Federal Labor Relations Commission Merit Systems Protection Board National Science Foundation National Labor Relations Board Nuclear Regulatory Commission U.S. Agency for International Development Dept. of Veterans Affairs Dept. of Energy Dept. of Interior Note: indicates data coverage for informal oversight. Off-site Oversight Before proceeding to analysis, it is useful to describe a few anecdotes of typical contact in order to verify that this contact is broadly consistent with the theory in the previous section. In some cases, contacts arise from constituents pulling the fire alarm about agency activities. Figure 1 shows the record of one such case, in which, Rep. Joseph Pitts (R-PA) contacted the Department of Energy on behalf of constituents with unsightly natural gas regulators placed near their homes. Constituents endured this externality because UGI Utilities Inc., an energy provider in western Pennsylvania, was required by agency rules. These complaints are not mere matters of curb appeal, and alarms are sometimes raised by third parties. Senator Barbara Boxer (D-CA), for example, contacted the VA after the broadcasting of an All Things Considered episode that described the families of 8

10 veterans tricked out of death benefits by MetLife, Prudential, and other life insurance providers. 9 Figure 1 Constituent complaints (Dept. of Energy) Other contacts appear to be attempts to detect wrongdoing or policy deviation. Then Senator Joseph Biden (D-DE), for example, contacted the U.S. Agency for International Development in 2007 requesting all internal policy evaluations of funded programs. These information gathering efforts do not mention a specific policy they are demands for transparency, more generally. As Figure 2 suggests, legislators with formal leadership positions also make requests of this kind though they are most often unrelated to formal oversight hearings. Members of Congress also recommend agencies take specific actions. In July 2009, Senator Chuck Schumer (D-NY) requested the Dept. of Interior reject the recommendation of the [Bureau of Indian Affairs ] Western Regional Office to allow 134 acres in Glendale, Arizona into trust for gaming purposes. 10 Members of Congress also make frequent requests that federal funds be spent in a particular way usually to benefit some specific constituent. For example, Senator Russ Feingold (D-WI) contacted USAID in February of 2007 on behalf of Marquette University to help secure an additional $50,000 for the African democracy training program run by the Les Aspin Center. The variety of contact raises the question of what does and does not constitute congressional oversight. Many of these inquiries (like Rep. Pitts natural gas regulators) are mundane aspects of daily governance. Often, they are not broad criticisms of programs but instead, specific concerns about the details of implementation left to the discretion of agencies. This includes requesting fed- 9 Life Insurance Firms Profit From Death Benefits, NPR, July 28, URL: story/story.php?storyid= Nonetheless, the Department allowed the land to be designated for gaming purposes, which led to a local controversy not resolved until November See: 9

11 Figure 2 Leadership contact (Dept. of Energy) eral funds be spent a particular way a practice known as letter-marking (Mills, Kalaf-Hughes, and MacDonald 2015). However, the conventional perspective on oversight previously described makes no distinction between the mundane, important, or distributive motivations. Instead, members of Congress attempting to gather information and influence any public policy qualifies as oversight of agencies. To reflect this generality, I investigate empirical patterns in most legislator oversight to agencies. More specifically, the dependent variables in the following analyses are two dichotomous indicators for whether the legislator contacted (1) or did not contact (0) a given agency during a particular Congress. The first indicator is any contact, excluding correspondence that did not require a response or that was directly related to committee activity. Examples of the former include thank you notes for officials attending in-district events. The later include follow-up questions for the record after committee hearings. This helps alleviate the concern that contact from committee membership will be a function of logistical necessity rather than policy oversight. The second indicator is the same, except that it excludes any grant-related contact. Grant support letters are a kind of oversight they are members attempting to guide agency action. But some may see them as far removed from the notion of fire alarms and policy deviation because they occur prior to agency decision-making. Moreover, the relative cost of what are likely form letters from members of Congress is sufficiently low that they might fall outside the scope conditions of informal oversight. Finally, it should be noted that collapsing the dependent variables to dichotomous indicators results in little information loss, because of the relative infrequency of contact by legislator-agency dyad. The baseline probability of any contact by a particular legislator for a particular agency in a given Congress is 0.36, and of those observations, 83% are five contacts or fewer. Thus, repeated contacts are very rare. All of the findings presented in the following section are robust to modeling 10

12 the complete counts. Findings I have argued that ideological disagreement between individual legislators and agencies will have a negligible effect on oversight propensity because of proximate legislators incentive to counteract the auditing of distant ones. Conversely, I argue that institutional roles in particular, committee membership should increase oversight because of the superior leverage these legislators have over the agencies they oversee. The results are consistent with these arguments, and are robust to a variety of different specifications and modeling assumptions. I begin with an analysis that leverages a presidential transition to analyze oversight variation within-district, then examine how consistent these results are after the inclusion of additional agencies and Congresses. Oversight and the Transition from Bush to Obama I compare changes in oversight of a particular agency by a particular district over the course of the transition from presidencies of George W. Bush and Barack Obama. This research design has a variety of nice features. First, it allows me to isolate the over-time variation in agency-ideology attributable to a change in the presidency. As I describe below, this will primarily be a function of staff turnover. Second, since Democrats maintained majorities in the House and Senate, all committee turnover will be unrelated to partisan changes in the executive branch. This restricts the sample to the seven agencies for whom I have data for this period, shown in Table 1, and is implemented with the linear probability model below, 11 Contact ijt = β 0 + β 1 Distance ijt + β 2 Committee ijt + ζx + γ i + δ j + φ t + ɛ i where γ i represents district fixed effects which accounts for differences in legislative constituencies, δ j denotes agency fixed-effects which accounts for the fact that some agencies perform functions 11 Since baseline probability of contact is 0.36, it is not surprising that the results of a logistic regression are nearly identical. Another concern of LPM estimation is that it yields predicted values outside the bounds of 0 and 1. In this case, few ŷ values (about 16%) fall outside this bounds. Thus, I report the LPM results for ease of interpretation. 11

13 more likely to be overseen by Congress, 12 and φ t is Congress fixed-effects, accounting for exogenous shocks common to a particular Congress. 13 ζ represents the effect of time-varying characteristics within districts and agencies I discuss later in this section, and ɛ i represents residuals clustered by district. The effect of ideological distance is driven by two sources of variation changes in legislator ideology driven by electoral turnover, and changes in agency ideology driven by personnel turnover. The effect of committee membership is driven by district representatives moving onto or off of jurisdictionally relevant committees. Distance is a measure of the ideological distance between the legislator and the agency. Development in this area of measurement is ongoing and vulnerable to persistent problems, so I rely three measures. The first is the absolute Commonspace DW-NOMINATE distance between the agency and legislator in a given Congress. Legislator ideal points come from Carroll et al. (2015). Agency ideal points were estimated by Chen and Johnson (2014). This provides a time-variant, measure of disagreement on an interpretable and often-used scale. Chen and Johnson use campaign donations and staff composition to place legislators and executive agencies on a common scale. Though this method is useful and innovative, it has several drawbacks. Agency ideal points vary by administration despite the fact that there are meaningful fluctuations in staff composition reported by quarter. 14 Also, since the scores are based on staff composition, many agencies are clustered toward the center of the ideological scale. So these estimates may reflect disagreement between an agency s employees and legislators, as opposed to legislator disagreement with an agency s mission. A second measure uses the commonspace estimates developed by Richardson, Clinton, and Lewis (2017). These time-invariant agency estimates are driven by survey results that ask agency personnel to take positions on particular bills for consideration in Congress (Clinton 12 Agency mandates may dictate higher rates of comment and recommendations on significant rules. Some agencies outlay billions in competitive grants which increases the probability of congressional support letters. Agencies vary meaningfully in size and scope, which may promote congressional incentives to oversee their activity. A brief, descriptive look at oversight patterns suggests this is the case. In general, agencies in the top panel of Figure B7 have fewer staff and less statutory authority than the departments and commissions in the bottom panel. 13 This includes differences between end-of-term and beginning-of-term levels of oversight, or changes in macroeconomic conditions that might influence legislative activity. 14 See the Fedscope database from the U.S. Office of Personnel Management: 12

14 et al. 2012). The third measure is the quintile distance ( Binned, ranging from 0-4) between the agency and the legislator. Here, I bin legislator DW-NOMINATE and Clinton and Lewis (2008) agency ideal point distributions into quintiles, with 1 indicating most liberal, and 5 indicating most conservative. I then use the absolute difference in these values as a measure of disagreement. In contrast to the previous measure, this relies on the estimates of Clinton and Lewis, who use the judgements of experts to assign agencies on an ideological scale. I bin these estimates into quintiles and take the absolute difference, since the two measures are on difference scales. I choose to bin these metrics to avoid assuming strict cardinality, but normalizing both measures to some scale (e.g. [0,1]) and taking the absolute difference produces similar results. In general, though there remain serious limitations to measuring agency ideology, the results across difference measures are strikingly similar which suggests some confidence is warranted. Committee is a dichotomous indicator for whether the district is represented by a member of a jurisdictionally relevant oversight committee for the agency in question. Relevant oversight committees were determined through agency staff responses to the 2014 Survey on the Future of Government Service (Lewis and Richardson 2015; Richardson 2015). An insignificant number of respondents reported more than one relevant committee, and these committee jurisdictions do not meaningfully differ from oversight plans released by Congress. 13

15 Table 2 Oversight in the 110th & 111th Congresses Variable: Measure: Dependent variable: Contact Non-Grant Contact Non-Grant Contact Non-Grant (1) (2) (3) (4) (5) (6) Distance Chen & Johnson (0.018) (0.018) Richardson et al (0.023) (0.024) Binned (0.004) (0.004) Committee (0.013) (0.013) (0.013) (0.014) (0.013) (0.013) Constant (0.157) (0.126) (0.128) (0.130) (0.156) (0.158) N 7,685 7,685 7,090 7,090 7,685 7,685 Linear probability estimates of dichotomous indicator for informal legislator oversight of agency; robust standard standard errors clustered by district in parentheses; all models include over-time controls (logged agency budget and president s party indicator) along with agency, Congress, and district fixed effects. I also include indicators for presidential co-partisanship, and agency budget. Note that because of the restricted time series, there are no changes in majority party or party control of a district that are not synonymous with the change in presidential co-partisanship. Agency budgets accounts for legislative changes that surely promote oversight. This is particularly important for the time series considered, since it includes the enactment of the American Recovery and Reinvestment Act (2009). Among other provisions, the stimulus funded renewable energy research grants, as well as a more general increase in the NSF budget of $3 billion, 15 which could plausibly influence Congress collective propensity to oversee. Table 2 reports the results across all measures of ideological disagreement and oversight contact. Overall, the estimates are consistent across specifications and indicative of the expectations described earlier. Ideological distance appears to have a negligible effect on oversight, whereas committee membership increases the likelihood of oversight. By negligible effect, I do not mean the mere absence of statistical significance (Rainey 2014). In model 1, the upper bound of the 15 Release of these funds drew some controversy. In 2010, for example, Republican Senators released a list of wasteful spending under the stimulus. See GOP slams stimulus plan with list of 100 worst projects, CNN, August 3, 2010; url: ArbgvoCEUuW 14

16 95% confidence interval would suggest that a standard deviation change in distance is associated with a two percentage-point increase in the probability of oversight. Across all measures, this is the largest (in magnitude) effect size within a confidence interval, suggesting the effect of distance is near zero. In contrast, committee membership is associated with an 8 percentage-point increase over the unconditional contact probability of Thus, the results are consistent with the theory of individualized oversight in a balanced sample restricted to leverage variation from the presidential transition. Oversight in the 110th-113th Congress Though the previous analysis provided estimates with a balanced panel and clearly identified sources of variation, it is limited by the restricted time series and sample of agencies. The results may be a function of idiosyncratic factors related to the 110th-111th Congress, presidential transitions in general, or the specific agencies included. To address these issues, I extend the previous analysis to the complete sample of agencies in Table 1. This, however, trades off generalizability for additional assumptions. Most obviously, for the estimates to remain unbiased, the unbalanced panel requires we assume the factors leading to inclusion in sample are unrelated to oversight. Some clearly meet this requirement. For example, the data were obtained via FOIA request, and some agency-congresses were not included because of agency protocols to transfer such records to the National Archives and Records Administration (NARA) after a set date. However, a FOIA request is citizen-performed audit of an agency that requests information. So it is possible that agency responsiveness to the requests for data are associated with legislator s propensity to contact them. The models are similar to those in the previous section, with the exception that additional variation warrants the inclusion of more time-varying characteristics of legislators in districts. The House of Representatives, for example, now includes a change in majority party. So I include indicators for presidential co-partisan, majority party status, and Republican. In addition, because there are sufficient within-majority party changes in committee chair and ranking minority legislators, I include indicator variables for both committee roles. I have argued that committee membership gives legislators leverage over agencies that makes performing audits more attractive. Thus, I expect that these positions increase that leverage and the likelihood of contact. 15

17 Table 3 Oversight in the 110th-113th Congress Variable: Measure: Dependent variable: Contact Non-Grant Contact Non-Grant Contact Non-Grant (1) (2) (3) (4) (5) (6) Distance Chen & Johnson (0.013) (0.013) Richardson et al (0.016) (0.016) Binned (0.003) (0.003) Committee (0.010) (0.010) (0.009) (0.009) (0.009) (0.009) Constant (0.129) (0.128) (0.151) (0.150) (0.126) (0.125) N 14,747 14,747 19,175 19,175 19,118 19,118 Linear probability estimates of dichotomous indicator for informal legislator oversight of agency; robust standard standard errors clustered by district in parentheses; all models include over-time controls (chair and ranking status, logged agency budget, president s, majority, Republican party indicators) along with agency, Congress, and district fixed effects. District fixed effects unique following redistricting. I report these estimation results in Table 3. Overall, the findings are consistent with the restricted sample. Ideological distance, by most measures, appears to have a negligible effect while committee status increases oversight. For models 1-2 and 5-6, the confidence intervals never include an effect size larger than 1 percentage point for a standard deviation increase in distance. The exception, however, are models 3-4, which use the Richardson, Clinton, and Lewis (2017) measure of commonspace ideological distance. Here, the 95% confidence intervals include a 4 percentage point reduction in the probability of contact for a standard deviation increase in distance. Since the agency ideology measure is time-invariant, this effect is driven entirely by electoral turnover within districts. Since the models account for status within committees, they estimate the effect of changes in audit patterns as legislators become or lose ranking and chair status. I report these findings in Figure 3. Not surprisingly, the reliance on fewer legislators for over time changes in ranking and chair status results in less precise estimates, but overall, the results follow the intuition above. Chair status increases the probability of oversight, over and above mere committee membership In most specifications, post-estimation tests indicate these coefficients are not distinguishable by convention (p = 0.15), so here, I do not mean that the effect of membership is necessarily smaller than the effect of Chair status. 16

18 In most specifications, there is similar effect for ranking minority legislators and the data suggest we cannot reject the possibility that they have the same effect. In other words, the highest ranking committee officials for each party do not appear to behave differently. Figure 3 Oversight and Committee Roles Member Ranking Minority Chair Marginal Increase in Probability of Oversight Note: Estimates from models reported in Table 3. Membership and institutional roles on relevant committee determined by oversight jurisdiction reported by surveyed officials (Richardson 2015). Individualized Oversight and Coordination Failures The previous analyses provide empirical support for the argument that individual incentives motivate legislators to perform off-site oversight of the bureaucracy. However, those results cannot speak to broader questions about the effectiveness of this oversight for facilitating accountability in the executive branch. While there is no obvious outcome comparable across agencies that would allow a more systematic test of this question, the data provide an alternative perspective on salient cases of oversight in the last decade. To investigate these broader questions, this section provides case studies of the 2014 VA wait-time scandal and the Nuclear Regulatory Commission s (NRC) response to the Fukushima Daiichi nuclear disaster. Overall, these incidents suggest that individualized oversight is parochial, reactionary, and may promote coordination failures. Importantly, these potentially undesirable features are consistent with what past scholarship has implied that But all Chairs are also members, so it represents the additional increase in probability above membership. 17

19 abundant oversight does not necessarily imply effective oversight. In the case of the VA, records suggest there were numerous fire-alarms pulled in advance of the scandal. In 2010, then-senator Jim Webb (D-VA) contacted the Department of Veterans Affairs after receiving complaints about the Washington, DC VA Medical Center. General complaints about claim processing delays are common in the VA log, but between 2010 and April 2014, ten legislators made inquiries about delays in medical care. But these inquiries are not all constituent casework as some voice general concerns about Department procedures. According to the log, in a 2011 inquiry, Senator Patty Murray (D-WA) Chair of the Senate Veterans Affairs Committee, voiced the following: Concerns about an internal VA memorandum indicating that gaming strategies are used at VA facilities to improve internal scoring measures at the expense of Veterans seeking VA health care. Request from VA how the VA has eliminated these improper scheduling practices... Murray refers to an April 2010 memo that indicates VA management was aware of the recordkeeping procedures that led to the 2014 crisis. So the key Senator tasked with overseeing the Department raised concerns about the exact problem a full three years in advance of the first CNN report. This oversight by individual legislators was accompanied by Government Accountability Office (GAO) reports in 2000, 2004 and 2013 that indicated reported wait times were unreliable. 17 But collective action despite the action of individual legislators, was largely absent. The lone exception is a hearing on March 14, 2013 in the House Veterans Affairs Subcommittee on Oversight and Investigations, where the problems with reporting wait times were voiced by VA officials and acknowledged by legislators. This illustrates the central point made by McCubbins and Schwartz (1984), that fire-alarms predominate over police patrols, while also suggesting the potential inadequacy of both. I do not mean to suggest that oversight could have prevented policy failure in this instance, or that legislators knew about policy failure and did not act. In fact, legislators engaged in considerable oversight. But, as a theory of oversight performed by individual legislators would suggest, much of this activity is in service of the narrow concerns of a few constituents. Collectively, this activity suggests the more widespread problems that were later revealed. But addressing those problems required coordination among legislators which 17 SOURCE 18

20 was eventually generated by the salience of the scandal, not foreknowledge of the problems that generated it. The VA and NRC cases also support the notion that this oversight is most often reactionary. The tsunami following the Tōhoku earthquake damaged the Fukushima Daiichi nuclear plant, resulting in three meltdowns and the release of radioactive material (Holt, Campbell, and Nikitin 2012). As Figure B6 indicates, prior to the incident, the NRC rarely heard from members of Congress. The NRC is among a set of peer agencies whose comparatively small budgets and staff sizes means they often attend few general oversight hearings per year usually before subcommittees on appropriations. This, of course, changed dramatically following the developments in Japan when the NRC received over 50 inquiries from representatives. The trend is similar at the VA. Following the publication of the CNN report, there were 38 inquiries related to wait times for medical care in This suggests congressional oversight functions the way past scholarship describes: as primarily reactionary, rather than preventative. But again, the reactions of legislators do not always correspond to collective activity that results in general policy change. In both cases, legislative oversight reflects parochial concerns. The flood of VA inquiries follow a predictable pattern with legislators making inquiries about whether VA facilities in their districts were affected. Oversight after Fukushima was similar, as legislators with nuclear plants in their districts particularly those on the west coast asked what the NRC was doing to prevent similar disasters from occurring at those U.S. facilities. Persistent inquiries eventually led to the publication of reports detailing lessons learned and regulatory differences between the U.S. and Japan. It is also worth noting that in this instance, oversight was generated by an event seemingly unrelated to the actions of the NRC. Whereas most theoretical models suggest oversight is meant to prevent policy deviations, in this case, oversight came in response to an exogenous change in the state of the world. Moreover, the demands of legislators appear to have shifted the NRC s priorities and added to their existing workload, even as the NRC s final reports indicate that existing planning standards for severe incidents made U.S. facilities more resilient to flooding ex ante. These cases illustrate several features of legislative oversight that are difficult to observe without the data presented in this paper. Scholars have suggested that much of Congress oversight activity goes unobserved as legislators forward fire-alarms from constituents. In these instances, that 19

21 Figure 4 Fire Alarms and the Fukushima Nuclear Disaster Informal Contact Workload Note: workload is the number of inquires that remain to be closed out on a given day in the Nuclear Regulatory Commission. Dashed line denotes March seems to be the case. But they also suggest that oversight performed by individual legislators does not always lead to the collective action required for problem-solving. That oversight is often reactionary, and narrowly focused on the concerns of constituents. Conclusion The scope and effectiveness of legislative oversight of the bureaucracy is a perennial issue in American politics and modern government, more generally. Agencies are tasked with the details of policy implementation, and are increasingly tasked with complex problems that require delegated authority and policy discretion. Since the publication of McCubbins and Schwartz (1984) work, political scientists have argued that much of legislative oversight of the bureaucracy occurs off-site. But without record of those interactions, we cannot test basic empirical predictions about this larger universe of interactions between Congress and the bureaucracy. Moreover, this information permits analysis of individual-level incentives of members of Congress to conduct oversight. Past work has been limited to analyzing Congress as a unitary actor most often its implicit floor or committee median. In contrast, this article suggests congressional oversight is far more diffuse and ubiquitous. Hearings, resolutions, and statutory changes are exceptions to the oversight that occurs day-to-day. In light of that fact, I have modeled factors influencing legislators likelihood of performing that oversight. I find that ideological disagreement has a neg- 20

22 ligible impact on oversight, whereas committee status increases the probability of oversight. Both of these findings diverge from standard models of Congress an institution. Whereas Congress as a whole would, intuitively, seek to oversee agencies it disagrees with competing legislators have an incentive to counter-act the oversight of their colleagues. Whereas committees perform oversight as a function of routine distributions of labor in Congress, committee members take advantage of their positions to enhance their own individual oversight goals. In addition, this article highlights a qualitatively different collective action problem than the one pointed out by (Gailmard 2009). Whereas that work focused on redundancy as generating sub-optimal level of oversight induced by the free-rider issues, this is primarily a coordination problem. Because the vast majority of oversight is conducted by individual legislators privately, the institution is sometimes unable to observe systemic problems. Moreover, the general empirical patterns revealed suggest that both problems may be worse than previously thought, as 535 principals rather than several committees with overlapping jurisdictions attempt to police the administrative state with the parochial concerns of constituents. References Aberbach, Joel Keeping a Watchful Eye: The Politics of Congressional Oversight. Brookings Institution Press. Austen-Smith, David, and John R Wright Counteractive Lobbying. American Journal of Political Science 38(1): Calvert, Randall L., Matthew D. McCubbins, and Barry R. Weingast A Theory of Political Control and Agency Discretion. American Journal of Political Science 33(3): Carroll, Royce, Jeff Lewis, James Lo, Nolan M. McCarty, Keith Poole, and Howard Rosenthal Common Space DW-NOMINATE Scores with Bootstrapped Standard Errors.. Chen, Jowei, and Timothy Johnson Federal employee unionization and presidential control of the bureaucracy: Estimating and explaining ideological change in executive agencies. Journal of Theoretical Politics (mar). 21