The Negative Politics of Distributive Policymaking:

Transcription

1 The Negative Politics of Distributive Policymaking: U.S. Federal Grants and the Contraction of Administrative Authority George A. Krause University of Georgia and Matthew Zarit University of Pittsburgh Preliminary Draft Version 1.0 * Paper prepared for delivery at the 2018 annual meetings of the Southern Political Science Association. Hyatt Regency Hotel. New Orleans, Louisiana. January 4-6. We thank Doug Kriner for generously providing his matching codes for congressional districts and counties in his published research with Andrew Reeves. We also acknowledge Anne Joseph O Connell for her contributions to the joint work with the lead author in the creation of these bureaucratic leader latent trait measures employed in this study. Alumni Foundation Distinguished Professor of Public Administration, Department of Public Administration and Policy, School of Public and International Affairs, University of Georgia, 280G Baldwin Hall, Athens, GA gkrause@uga.edu. Corresponding Author Ph.D. Candidate, Department of Political Science, University of Pittsburgh, 4433 Wesley W. Posvar Hall. Pittsburgh, PA maz44@pitt.edu ( address). Keywords: Distributive Policymaking, U.S. Federal Grant Retrenchments, Presidential Control of Administrative Agencies; Executive Branch Coordination, Separation of Powers

2 Abstract This study offers a novel perspective of distributive policymaking that analyzes the contraction of administrative authority reflected by reductions in existing federal grants. This form of negative distributive policymaking behavior is motivated by presidents wishing to limit their own moral hazard from U.S. federal agencies. We test the empirical implications of our theory using a sample of over 842,000 grant cuts per award in a given quarter covering 21 federal agencies from1984: Q1 through 2008: Q4. The evidence offers empirical support for this logic by showing that largest retrenchments of existing grant authority occur in those agencies where presidents encounter the most severe moral hazard problems, while grant cuts tend to be smallest for those agencies that are most susceptible to presidential control. Moreover, we find that these patterns of executive grant retrenchments are largely driven by those federal agencies that are increasingly at risk of congressional influence from opposition parties under divided party government. These findings suggest that presidents strategically reduce distributive policymaking benefits for purposes of consolidating executive authority during times of presidential-legislative policy conflict.

3 Distributive policymaking entails the provision of government benefits to various stakeholder constituency groups and individuals within a political jurisdiction. Research on distributive policymaking begins with the premise that government allocates policymaking authority by expanding it through budget authority supplied in the form of grants to various governmental and non-governmental actors. Because only gains from positive policy benefits are accrued, foundational work on distributive policymaking presumes that little, if any, policy conflict arises among governmental actors (Lowi 1972; Wilson 1973). It is thus hardly surprising that distributive policymaking is an area rife with evidence showing that politicians are highly effective at distributive policymaking that accrues particularistic benefits for legislators districts (e.g., Stein and Bickers 1994; Shepsle, Weingast, and Johnsen 1981), strategically targeting resources to states and counties that are electorally important to an incumbent president s own electoral fortunes (e.g., Hudak 2014; Kriner and Reeves 2015), key members of their own party (e.g., Berry, Burden, and Howell 2010; Alexander, Berry, and Howell 2016), and agencies that are malleable to executive will (Berry and Gersen 2017; Bertelli and Grose 2009). In short, distributive policymaking enables politicians to allocate positive benefits to key constituencies, with a subset of elected officials enjoying disproportional influence over these decisions. Yet, it is not uncommon for U.S. federal grants to be partially rescinded after their initial awarding. These resource retrenchments pertain to previous grant commitments that were only to be reduced at a later date while the grant was still operable. We refer to this phenomenon as negative distributive policymaking. Take for example, The Department of Homeland Security reduced a grant to New York City for security purposes from $207.6 million to $124.5 million (Baker and Chan, New York Times, June 13, 2006, B2). This cut jeopardized the city s plans to install modern surveillance equipment, plus special police units to deal with threats posed by 1

4 domestic terrorism. The basis of this decision was that the Department of Homeland Security had not been satisfied with how these funds would reduce safety risks, and doubted the financial sustainability of some counterterrorism measures advocated by the city (Baker and Chan, New York Times, June 13, 2006, B2). In turn, this forced the NYC Budget Office First Deputy Director, Stuart A. Klein, to comment to city council members about the difficult choices that laid ahead for the city to fund their projects Since we were recently notified of the reduction, we re going through a process now, analyzing projects, working with the mayor s office and other agencies, And when a decision is made, we will forward the review of that to Homeland Security. (Baker and Chan, New York Times, June 13, 2006, B2). This illustration of grant retrenchment underscores not only the negative benefits incurred by policy stakeholders, but also the contraction of administrative policymaking authority for those U.S. federal agencies that play a determinative role in both the funding and management of federal grants. The aim of this study is to understand negative distributive policymaking decisions, based on the amount of retrenchments made to existing U.S. federal grants of a discretionary nature (i.e., non-formula grants) awarded by federal agencies. 1 We advance a theory designed to explain variations in the retrenchment of U.S. federal grants that presumes presidents seek to minimize moral hazard problems associated with bureaucratic agencies when a contraction of 1 We limit our analysis to discretionary grants since federal agencies have considerable leeway over them, and can be adjusted according to the political-administrative process, (see also, e.g., Berry, Howell, and Alexander 2016; Bertelli and Grose 2009; Bickers and Stein 1996). Formula (non-discretionary) grants constrain the alteration of distributive policymaking since statistical formulas determine allocations, and these formulas are more difficult to change. 2

5 administrative authority takes place. Because undertaking such action is both costly and targeted unevenly across active grants, presidents have an incentive to ensure that grant losses are distributed in a manner that falls disproportionately on those agencies that they find the most difficult to control. That is, the moral hazard fallout experienced by presidents from grant retrenchments will become more severe when bureaucratic leader loyalty to the president is decreasing, coupled with agency policy decisions being increasingly insulated from presidential review. Presidents will therefore have the strongest incentive to employ negative discretionary policymaking behavior as a means to limit congressional influence when effective coordination within the executive branch is most fragile under conditions of divided party government. We test these claims using approximately 842,000 discretionary non-closing U.S. federal quarterly grant cuts from the FAADS database from covering a sample of 21 U.S. federal agencies. The statistical evidence reveals that federal grant cuts are consistent with a presidential strategy of minimizing moral hazard when delivering negative policy benefits. Grant cuts are not distributed uniformly across agencies and through time. Rather, the largest (smallest) cuts in federal grants occur when agencies policy decisions are strongly (weakly) insulated from presidential policy review, coupled with chief administrators considered to exhibit weak (strong) loyalty to the appointing president. Presidential efforts at mitigating moral hazard problems in the administration of U.S. federal grants is strongest when public agencies have an opportunity to shirk in the direction of Congress when it is controlled by the opposition party. Because U.S. federal agencies play a critical role in distributive policymaking activities, presidents seek to strategically engage in negative distributive policymaking behavior that most adversely impacts those grants administered by federal agencies whom are capable of exercising bureaucratic discretion in a policy environment characterized by weak executive control. 3

6 GOING NEGATIVE: THE LOGIC OF RETRENCHMENTS IN DISTRIBUTIVE POLICYMAKING U.S. federal grants have a major impact on governmental activities. In fiscal year 2015, for example, federal grants made up nearly one-third of budgetary resources in the American states, as well as the same proportion of total non-defense discretionary spending by the federal government (Lav and Leachman, March 13, 2017, Center on Budget and Policy Priorities). While past research chronicled earlier has contributed to our understanding regarding how these positive benefits associated with federal grants are allocated based on political considerations, extremely little is known about the process that shapes the distribution of grant cuts. Rescissions of existing funds must go through Congress but are also frequently proposed at the presidential level. President George W. Bush, for example, included $2.3 billion in budget rescissions in 2005 in order to provide money for rebuilding the Gulf Coast after Hurricane Katrina. 2 As such, budgets in agencies were slashed, including grant programs. For example, in the Department of Transportation, the Recreational Trails Program, funded through block grants to states 3, were set to be cut. 4 Some stylized facts regarding negative distributive policymaking behavior can be gleaned from a sample of data from 21 major agencies that constitute a major share of retrenchment

7 activities involving U.S. federal grants. 5 These data appear below for this sample of agencies between 1984 and Figure 1A reveals that rescinding portions of existing awarded grants is commonplace. On average, 35,974 non-closing grant retrenchments occur per fiscal year (including multiple cuts per grant), covering approximately a quarter of all grant awards. 6 The greatest uptick in both the frequency and proportion of grant cuts transpires in the late 1980s and late 2000s during the second terms of the Reagan and G.W. Bush administrations, respectively. The fiscal year real dollar annual value of these grant retrenchments ranges between approximately $4.4 billion in 2000 and $21.4 billion in 2003 for our sample period. This evidence is consistent with those who claim that rescinding budgetary authority has a negligible impact on reducing government spending and budget deficits, but rather serves to alter policy priorities by selecting which programs lose benefits (Havens 1992). 5 Grant retrenchments occur in the following set of twenty agencies in our sample: Broadcasting Board of Governors, Department of Agriculture, Department of Commerce, Department of Defense, Department of Education, Department of Energy, Department of Health and Human Services, Department of Homeland Security, Department of Housing and Urban Development, Department of Justice, Department of Labor, Department of Transportation, Department of Interior, Environmental Protection Agency, Equal Employment Opportunity Commission, Federal Emergency Management Agency, Nuclear Regulatory Commission, Small Business Administration, Social Security Administration, and U.S. Agency for International Development. 6 Closing retrenchments are excluded from our sample since this type of grant cuts are performed when a grant award is expiring, and thus connote the elimination of unused surplus funds, and not a contraction of policymaking authority/benefits. 5

8 However, it is worth pointing out that grant cuts are not distributed uniformly. Some grant cuts are a negligible fraction of a grant award. For example, in September of 2008, $277 was cut from an $850,000 Section 8 housing grant to Leon Arms Apartments by the Department of Housing and Urban Development. In other instances, grant cuts reflect a tangible loss that adversely affects administrative policymaking authority, such as when the Department of Transportation cut $271,349 from a $432,081 Highway Planning and Construction grant to the Wisconsin Department of Transportation in The magnitude of retrenchments in U.S. federal grants reveal the extent to which policy stakeholders and beneficiaries are adversely affected by these policy choices. Negative distributive policymaking behavior contrasts with the canonical view that presumes a positive provision of governmental benefits that entails an expansion of administrative authority via policy delegation. In the latter case, presidents cede executive policymaking authority to U.S. federal agencies that are largely responsible for determining the 6

9 selection and management of grant beneficiaries (Keegan 2012; Yeh 2017). Both the sheer volume and diversity of U.S. federal grants that routinely occur in a given fiscal year covers a wide array of policy programs administered by a multitude of federal agencies. In turn, presidents capacity to control the awarding of federal discretionary grants is heavily constrained on two dimensions. From a supply-side perspective, presidents may seek to limit transaction costs by not maximizing policy control (Lewis and Rudalevige 2007; Nathan 1983). From a demand-side perspective, presidents will have an incentive to balance the competing benefits associated with policy control against administrative effectiveness (e.g. Bendor and Meirowitz 2004; Gailmard and Patty 2013). Presidents therefore have an incentive to follow a temperate path when expanding administrative authority. This path seeks to balance political control and bureaucratic discretion when it comes to the allocation of grant funding decisions made by federal agencies (Krause and Zarit 2017). An example of a disruption of the balance between political control and administrative discretion can be found in the Department of Justice during the George W. Bush Administration. The World Golf Foundation was awarded a grant to address juvenile crime despite their proposal being rated 47 th out of 104 applicants by staff within the agency. (Ross, Schecter, and Wass 2008). In this case, the agency leadership overruled bureaucratic judgments to deliver the grant to a potentially sub-par organization. The political calculus for supplying negative benefits captured by grant retrenchments should systematically differ from decisions regarding the provision of positive benefits. We maintain that presidents prefer to distribute losses associated with grant retrenchments disproportionately on those existing grants administered by federal agencies that they exert weak executive policy control over. This retrenchment of distributive benefits translates into a tangible contraction of administrative policymaking authority that differs markedly in two distinct ways 7

10 from the expansion of delegated authority associated with the awarding of federal grants. First, because cuts to existing grants are less frequent events relative to the awarding of grants, the former will be more directly targeted to those agencies whom the president will have a difficult time exerting influence over their policymaking behavior. Because amounts cut in existing grants represent a fraction of the total value of grants awarded in a given year, this activity is manageable for presidents, and their support offices, to focus their efforts on which grants to reduce from their original award status. Presidents can thus view grant cuts as a means of increasing their policy control over distributive policymaking that cannot be feasibly applied to grant award decisions. This is because the positive benefits associated with the expansion of administrative authority, coupled with the seemingly ubiquitous nature of this funding mechanism managed throughout the U.S. federal bureaucracy, do not permit presidents to maximize policy control over grant award decisions. Grant retrenchment decisions also entail allocative choices regarding how to distribute losses (i.e., negative policy benefits) among various stakeholders; whereas, the awarding of grants constitutes the distribution of gains (i.e., positive policy benefits). Research on budget cuts experienced by U.S state governments indicate that targeted cuts are acute (e.g., Bourdeaux 2016; NASBO 2009, 2010). Rescinding budget authority reflects shifting policy priorities. In testimony before Congress discussing the General Accounting Office s role in the rescissions process, Assistant Comptroller General Harry S. Havens asserted that rescissions to discretionary budget authority are politically-motivated choices regarding which programs (and by extension, agencies) to target, and the extent to reduce such authority (Havens 1992: 1, 4, 6). From a behavioral perspective, agencies experiencing discretionary grant funding retrenchments incur an endowment effect that makes political contestation regarding the 8

11 contraction of administrative authority more divisive than compared to the expansion of administrative authority (Bourdeaux 2016: 10-11). Agencies will thus be put in a tenuous intermediary position of adversely revising benefit commitments to grant recipients that were not anticipated when initial grant award determinations were made. Moral hazard problems arising from public agencies will thus be exacerbated for presidents when they must contract their policymaking authority by the formal act of administering reductions in existing grant awards. This is because budgetary cutbacks are known to induce instability and low morale within bureaucratic agencies (e.g., Lee, Johnson, and Joyce 2013: 251; Levine 1978; Pandey 2010). We assert that the largest grant cutbacks should be earmarked for those agencies that presidents experience weak executive control. Next, we elucidate the observable mechanisms at work. PRESIDENTIAL STRATEGIES FOR HANDLING GRANT RETRENCHMENTS Because reductions in existing federal grants neither occur with the same frequency nor value as the provision of grant funds, plus create difficulties for agencies charged with administering them, presidents will prefer to contract the policymaking authority of agencies that they deem least likely to comply with their policy objectives. We argue that presidents will evaluate agencies propensity for moral hazard by focusing on the extent to which the agency head is loyal to the appointing president, and the extent to which the president (and EOP units by extension) can review policy decisions made by the agency. In essence, presidents utilize this information to gauge the nature of coordination problems that arise with the shared exercise of executive authority. In 1995, President Clinton signed a law that rescinded $16.4 billion in spending from federal agencies. Roughly $6.3 billion of these cuts were made to programs at the Department of Housing and Urban Development (HUD). HUD was led by Henry Cisneros, a Clinton appointee, 9

12 who had a relatively moderate level of loyalty for a top-level appointed agency head (Normalized Krause-O Connell Loyalty Score = 0.219: percentile among the subset of toplevel appointees). According to Selin s (2015) second dimension measure of agency insulation from political review, HUD was a moderately insulated agency [0.372: 64 th percentile]. In contrast, the Department of Agriculture (USDA), an agency led by a somewhat stronger Clinton administration loyalist in Dan Glickman (Normalized Krause-O Connell Loyalty Score = 1.131, percentile among the subset of top-level appointees) that was also subject to stronger political review (Selin 2 nd dimension score = 0.043, 54 th percentile) saw relatively minor cuts of $96 million 7. This example illustrates how presidents seek to distribute the contraction of administrative authority in a manner that minimizes their moral hazard problems. Presidents confront severe moral hazard problems in the administration of federal grants that can be mitigated through the strategic use of grant retrenchments. These moral hazard problems are most severe when presidents expect little loyalty from agency leaders, coupled with agency policy decisions being heavily insulated from their purview. Conversely, presidents moral hazard problems will be minimal for those agencies whose appointed leader is quite loyal to the occupant of the White House, and also where its policy decisions are easily subject to review by the chief executive. We thus offer the following testable hypothesis that predicts how U.S. federal discretionary grant cuts will vary across agencies, that is predicated on both the 7 "$16.3 Billion Cut from 1995 Spending." In CQ Almanac 1995, 51st ed., Washington, DC: Congressional Quarterly,

13 agency leader s loyalty to the president, and the president s capacity to review agency policy decisions. H1: Grant losses are decreasing when bureaucratic leader loyalty to the president and presidential capacity to review of agency policy decisions are jointly increasing. H1 implies that contractions of administrative authority will be positively associated with the scope of the president s moral hazard problem in the realm of distributive policymaking. That is, presidents prefer distributing the loss of policy benefits in a manner that minimizes coordination problems within the executive branch. This logic differs from expansions of administrative authority consonant with canonical delegation choices that positively correspond to the balancing of policy control with the exercise of bureaucratic discretion (Krause and Zarit 2017). Delving deeper into this logic, we wish to understand whether the contraction of administrative authority is rooted in policy conflict between political principals. In the presence of separation of powers conflict between the executive and legislative branches, both presidents and Congress seek to control the distribution of federal grants to reward different sets of stakeholders. When these political branches policy interests are not aligned, Congress provides an effective check on executive authority since federal grants rise in those bureaucratic agencies that are less susceptible to presidential influence. It naturally follows that presidents are most effective at balancing policy control with bureaucratic expertise when their policy interests are aligned with Congress (Krause and Zarit 2017). But as noted earlier, the political calculus should substantively differ when it comes to retrenchment of federal grants. Presidents moral hazard problems will become more challenging to grapple with when it involves the contraction of administrative authority. This is because grant losses incurred by federal agencies whose policy decisions are insulated from presidential review and are led by agency heads that exhibit little 11

14 loyalty to the president will be more inclined to become allied with Congress. Under divided party government, deeper cuts should occur for those grants administered by agencies that are less susceptible to presidential influence. Conversely, presidents can afford relatively more diffuse contractions of administrative authority under unified party government since the policy alignment between political institutions make it less feasible for hard to control agencies to become aligned with Congress. This logic generates our next hypothesis. H2: Grant losses are decreasing at a greater rate consistent with H1 under divided party government compared to unified party government. In short, H2 posits that the contraction of administrative authority in more presidentialautonomous agencies will be greater when these agencies can more easily seek refuge from opposition party majorities within Congress, than compared to when obtaining such political support becomes more difficult when both the president and Congress are controlled by the same party. In other words, presidents prefer to distribute the loss of policy benefits in a manner that insulates the executive branch from policy coordination problems that may arise from legislative interference when political institutions policy interests are not aligned. DATA AND EMPIRICAL STRATEGY To test our logic regarding how presidents strategically contract administrative authority, we analyze Federal Assistance Award Data System (FAADS) data on the 2009 constant-dollar total value of cuts made to existing discretionary (non-formula) federal grants from 1984:Q1 through 12

15 2008:Q4 for a sample of 21 major U.S. federal agencies. 8,9 The FAADS data offers individual grant-specific data from U.S. federal agencies that can be further apportioned to quarterly time and county-level cross-sectional units. The dependent variable (Grant Loss) is converted to natural logarithms to normalize the distribution of grant funding amounts. Larger grant loss amounts correspond to a contraction of administrative policymaking authority for each grant retrenchment. The key independent variables employed to test our theoretical predictions centered on presidential control of bureaucratic agencies measure the extent to which agency leaders are loyal to their appointing presidents, and the extent to which presidents can easily review agency policy decisions. In tandem, these measures account for the extent to which presidents entrust authority with specific appointed top officials charged with leading federal agencies, as well as the extent to which they can engage in executive review of agency decisions. The former measure, Loyalty, is operationalized as the normalized presidential loyalty latent trait scores for top bureaucratic leader positions in a sample of U.S. federal agencies developed by Krause and 8 We employ the FAADS data since it includes agency identification information for grant allocations dating back to 1984; whereas, the Consolidated Federal Funds Report [CFFR] data only designates agency origination for grants from 1993 onward (Berry and Gersen 2017: 1018). 9 Our analysis corresponds to roughly 80% of the 26 major agencies and slightly less than half of the 44 total agencies covered in Krause and Zarit s (2017) analysis of positive distributive policymaking. This coverage is still quite expansive when one considers the relative infrequency of grant cuts to grant awards (see Figure 1B). 13

16 O Connell (2016, 2017). 10 We employ the top appointed agency official in a given agency at the time of the grant retrenchment decision. 11 This variable is recalibrated with a minimum of zero to ensure that all baseline estimates correspond to minimum observed levels of bureaucratic leader loyalty to the appointing president in our sample. The latter measure, Review, is simply the inverted second dimension agency insulation estimates developed by Selin (2015), and hence, captures presidential capacity to review agency policy decisions. 12 Increasing values of 10 This measure is generated from a structural generalized latent trait model using Bayesian Markov Chain Monte Carlo simulation methods. The indicator variables employed to gauge presidential loyalty of bureaucratic leaders is based on six indicators: (1) whether the appointee is a member of the same party as the nominating president, (2) whether the nominee has previously donated money to the nominating president, (3) whether an appointee of the same party as the president previously served in an appointed position in a federal agency, (4) whether the appointee has elective office experience, (5) whether the appointee has experience working for the national party organization or running a state party organization in alignment with the president, and (6) whether the appointee worked in state government when the nominating president served as governor. 11 For the sample of agencies under investigation here, the Krause-O Connell Normalized Loyalty scores among top-level appointed agency officials has a mildly attenuated range of ( 1.733, 2.329) compared to a range of ( 1.733, 2.732) for their full set of agencies. 12 For the sample of agencies under investigation here, the Selin second dimension has an attenuated range of ( 0.563, 1.922) compared to a range of ( 0.988, 4.1) in those agencies excluded in our sample. 14

17 this inverted measure denote increasing presidential capacity to review agency policy decisions (i.e., declining agency insulation). This measure is also recalibrated to a zero minimum observed value so that baseline estimates correspond to the most insulated agencies in our sample. Because our broader focus is on the distribution of executive authority, it remains possible that grant cuts will be less severe in those agencies where presidents policy goals are aligned with policy activities of bureaucratic agencies. To account for this possibility, we account for the president s party affiliation (President s Party: + 1 Republican Presidents, = 1 Democratic presidents) and the ideological classification of agencies policy activities (Agency Ideology: conservative [+1], neutral [0], and liberal [ 1]) employing the 95% credibility estimates developed by Clinton and Lewis (2008). The interaction term, President s Party ⅹ Agency Ideology should possess a negative coefficient sign since grant cuts should be smaller when these institutional actors become increasingly aligned. We also account for the managerial skills (Managerial Competence) and policy expertise (Policy Competence) of bureaucratic leaders using the latent trait measures developed by Krause and O Connell (2016, 2017a, 2017b). The posited effect on grant retrenchments is ambiguous. On one hand, both of these qualifications-related traits are valuable for administering executive authority, and hence, should be inversely related to grant losses. On the other hand, there is a discernible tradeoff between bureaucratic leader loyalty and competence (Krause and O Connell 2016, 2017), especially relating to policy competence, that makes it plausible that grant retrenchments are larger in agencies headed by appointed leaders who possess greater policy expertise to effectively handle such contractions of administrative authority. In addition, we incorporate additional covariates established in the literature may impact U.S. federal grant decisions. For instance, we account for many factors that explain variations in 15

18 presidential and congressional particularism behavior considered by Kriner and Reeves (2015). 13 First, the standard presidential particularism covariates include qualitative characteristics associated with key political-geographical constituencies operationalized as a series of binary indicators. Swing states are states where the losing candidate for president reached at least 45% of the two major parties in the preceding three elections (Kriner and Reeves 2015: 94). Core states are where the president s party averaged 55% or more between the two parties in the preceding two elections, and core counties must meet the same threshold (Kriner and Reeves 2015: 143). Similarly, core states and counties offer attractive opportunities for presidents to reward and maintain their base for upcoming elections. We also include an indicator for election year (Election Year). Similar to Kriner and Reeves, we also incorporate several interaction effects to examine swing and core constituencies. We include indicators for core counties in swing states (Core County x Swing State), core counties in core states (Core County x Core State), core counties in election years (Core County x Election Year), core states in election years (Core State x Election Year), and swing states in election years (Swing State x Election Year). The expectation in each case, based on Kriner and Reeves findings, is that each interaction will be inversely associated with larger cuts in grant funding. 13 Data available from: Kriner, Douglas; Reeves, Andrew, 2016, "Replication Data for: Presidential Particularism and Divide-the-Dollar Politics", doi: /dvn/hqxkj9, Harvard Dataverse, V1, 16

19 Legislative particularism covariates are matched to the congressional districts to each county according to Kriner and Reeves s (2015: 195) 14 assignment based on which Member of Congress represents the majority of a county s population. Following prior studies analyzing particularism using federal grants data (e.g., Berry, Burden, and Howell 2010; Kriner and Reeves 2015; Berry and Gersen 2017), we also include a series of binary indicators indicating whether the district is represented by a member of the president s party (President s Party). The expectation is that this will be inversely related to grant retrenchments. We also include whether the Member of Congress is a member of the House majority party (House Majority), whether they are a committee chair, a member of the House Appropriations Committee (House Appropriations), or if they are a member of the House Ways and Means Committee (House Ways and Means). Each of these covariates is expected to have a negative effect on U.S. federal grant reductions. We also incorporate covariates on demographic information that varies across both counties and years employed by Kriner and Reeves (2015: 131). These measures include the natural log of the county population (ln[county Population]), along with the poverty rate for the county (County Poverty Rate), and the per capita income in 2009 constant dollars (County Per Capita Income). These variables account for time-varying county-specific characteristics that may alter the grant funding levels obtained by counties through time. Finally, we account for the amount of grant funds cut due to scale effects associated with a particular grant in our sample. For a given lifespan of a particular grant, some are cut multiple 14 We thank Douglas Kriner for generously providing us with these data in correspondence (January 18, 2017). 17

20 times. 15 We measure this using a covariate Grant Cut Frequency. We surmise that the more frequent cuts to a particular grant will be associated with smaller grant retrenchments. Grant Amount is a measure of the total value of a particular grant that is subject to a reduction. The total grant value is determined by summing the values that were paid to a grantee within our database. In some cases, this results in negative values where grants in the early years of our data see more cuts to their funds than payments. To reduce skew and kurtosis, this measure is subject to an inverse hyperbolic sine transformation that is applicable in the presence of non-positive values. Grant values that are larger will be more prone to being retrenched for purposes of minimizing costs imposed on these awards and the recipients benefitting from them. The general forms of the estimating equations in the respective baseline and partisan government regime model specifications employed to predict variations in U.S. federal government retrenchments take on a semi-logarithmic functional form to account for leptokurtosis in the distribution of the constant-dollar amount of grant losses from the retrenchment of U.S. federal grants. Larger (positive) values indicate large reductions in a specific grant s value for a given quarter: ln( Grant Loss ) Loyalty Re view Loyalty Re view X (1) ijt 1 jt 2 j 3 jt j k kijt ijt Re Re ln( Grant Loss ) Loyalty R e view Loyalty Re view ijt 1 jt 2 j 3 jt j UPG Loyalty UPG view UPG 4 t 5 jt t 6 j t Loyalty view UPG X 7 jt j t k kijt ijt. (2) 15 Approximately 22% (21.67% or 182,626) of the grant cuts in our sample were previously cut at least once. Of these prior retrenchments, 83% (151, 561) only occurred once, while 99.5% (181,687) of them occurred four times or less. 18

21 H1 is supported by the data when the value of federal grant retrenchments (Grant Loss) is at their apex when Loyalty and Review take on their lowest values, while being at their minimum when these covariates take on high values such that β1, β2 > 0, and β3 < 0. Similarly, H2 is supported by the data when the magnitude of these grant cuts increases at a greater rate as bureaucratic agencies charged with administering these policy decisions become less susceptible to presidential influence under divided party government (γ1, γ2 > 0, γ3 < 0), as well as compared to times of unified party government (γ1 > γ5, γ2 > γ6, γ3 < γ7). The model specifications contain a vector of additional covariates (X) described above, plus a disturbance term εijkt ~ (0, σ 2 I). Because the truncated version of this variable only covers grant losses, the sampling design can induce bias in OLS estimates since conditional mean of the truncated dependent variable is higher than the untruncated analog by construction. Therefore, we estimate models of the form of Equations (1) and (2) employing a truncated normal maximum likelihood regression technique. This method accounts for the nature of sampling bias in the form of estimating an ancillary nuisance parameter correction (σ). This method is ideally suited to address truncation when its nature is known to the researcher by data design since they choose to restrict the value of the outcome variable of interest to particular subset of the population i.e., choice-based sampling (Wooldridge 2003: ). The estimated models are augmented by inclusion of two-way fixed effects by county and quarter to account for unobserved heterogeneity that may arise in the form of between-county variations in U.S. federal grant cuts and common-period national-level shocks in the retrenchments of federal grants that impacts both agencies and 19

22 counties, respectively. 16 Robust estimates of the variance-covariance matrix for these statistical models are cluster-adjusted by county since it is the most granular unit of analysis in our data design, and also contains a large number of clusters to ensure consistency of these standard error estimates (e.g., Petersen 2008). EMPIRICAL FINDINGS The truncated normal maximum likelihood regression estimates appear in Table 1. Given that both models are estimated in semi-logarithmic form, the coefficients and corresponding standard errors are transformed into percentage change form (100 (e β 1)). Ideologically moderate agencies typically experience approximately a 40% cut, while the cut for agencies whose policy activities are ideologically aligned with presidents are cut by about 38% (2 19%) more compared to ideologically opposed agencies. Although the direction of this effect is opposite of expectations, it remains possible that presidents prefer grant retrenchments in those agencies that they are less concerned about moral hazard risks, and thus place most of the burden on agencies whose policy activities are ideologically aligned with their own. The only covariate related to geographic targeting that seems to predict grant retrenchments in a meaningful way is Election Year. That is, the average grant loss that occurs for grants linked to non-core counties, non-core states during an election year are a factor of 9.24 and 7.60 times greater, on average, than compared to these key geographical areas in election 16 Because the Insulation covariate only exhibit between-agency variations, we omit agency fixed effects from these model specifications. Accounting for the baseline Insulation effect on federal grant allocations across agencies is critical for testing our theory s empirical implications. 20

23 years, as well as non-key geographical areas in non-election years. This suggests that grant losses fall disproportionately on non-targeted, safe electoral jurisdictions that can act as a buffer from losses being incurred in electorally and partisan valuable geographic areas where it could adversely affect the incumbent president s party re-election chances. The small percentage change coefficient impacts, combined with lack of statistical significance, associated with the key legislators often viewed as being most influential in the exercise of distributive policymaking suggests that these representatives districts do not bear the burden of grant retrenchments. Further, demographic considerations do not significantly impact the burden of grant losses on specific counties. Yet, the frequency of cuts made to a given grant award in a given fiscal year translates into smaller grant losses for a particular grant in a given quarter, thus suggesting that successive grant reductions in a short time frame will be smaller. Although the volume of grants awarded by an agency and the bureaucratic leader s managerial competence have no significant impact on the value of grant retrenchments, the latter individual s policy competence is positively associated with the magnitude of grant losses experienced by the typical award. This finding suggests that agencies headed by an individual with considerable policy expertise will be better equipped at dealing with grant retrenchments, or instead to offset the positive grant benefits obtained by agencies headed by a leader that exhibits relatively higher levels of policy expertise (Krause and Zarit 2017). The covariates of primary interest, Loyalty, Review, and Loyalty Review in the baseline model specification possess the correct hypothesized signs (β1, β2 > 0, and β3 < 0) indicate that grant losses are decreasing when the president s moral hazard problems become less severe due to increased loyalty of their appointed agency head, as well as president s greater capacity to review agency policy decisions consistent with H1. The evidence from the Divided-Unified 21

24 TABLE 1: Baseline and Divided-Unified Government Models Covariates Expected Sign Baseline Model Agency Constraints Loyalty (β 1, γ1) *** (10.81) Review(β2, γ2) *** (22.36) Loyalty x Review (β3, γ3) *** 22 Divided-Unified Government Model 358.9*** (11.77) *** (24.77) *** (7.990) (7.258) Unified Party Government (γ 4 ) *** (56.00) Loyalty x Unified Party Government (γ 5 ) +/0/ *** (14.48) Review x Unified Party Government (γ 6 ) +/0/ *** (39.17) 114.6*** (10.61) Loyalty x Review x Unified Party Government (γ 7 ) President-Agency Ideological Alignment Party of the President +/0/ 41.83*** (7.790) Agency Ideology +/0/ (5.351) Party of the President x Agency Ideology 18.53*** (2.502) Geographic Targeting Swing State (5.581) Core State (8.557) Core County (5.413) Election Year *** (19.18) Core County x Swing State (5.496) Core County x Core State (9.799) Core County x Election Year (3.680) Core State x Election Year (6.821) Swing State x Election Year (4.247) Political Actor Targeting Member of the President s Party (3.003) Majority Party (4.833) 38.98*** (7.894) (5.163) 19.60*** (2.462) (5.563) (8.498) (5.453) 659.7*** (19.16) (5.522) (9.707) (3.685) (6.829) (4.241) (2.991) (4.857)

25 Committee Chair (7.738) House Appropriations (4.919) House Ways and Means (8.743) Demographic ln[county Population] (28.95) County Percent in Poverty (1.287) County Per Capita Income (0.566) Grant Characteristics Grant Cut Frequency (Same Fiscal Year) *** (1.310) Grant Amounts in Fiscal Year (IHS Transformed) (0.140) Additional Bureaucratic Leadership Traits Managerial Competence +/0/ (2.890) Policy Competence +/0/ 85.63*** (5.180) Truncation Bias Correction *** (1.831) (7.689) (5.009) (8.687) (28.64) (1.293) (0.577) *** (-1.28) (0.141) (3.096) 86.78*** (5.334) *** (1.825) Pseudo R AIC 3,909, BIC 3,910, N 842, ,863 Party Government model specification reveals that this pattern holds only under divided party government regimes (γ1, γ2 > 0, γ3 < 0). Eras of unified party government reveal a distinct pattern, as increases in presidential capacity to control agencies through both means results in a relatively higher level of grant losses compared to divided party government (γ7 > 0). As a matter of fact, grant losses are almost 11 times larger under unified party government than compared to divided party government when presidential capacity to control agencies is constrained to its weakest observed levels (γ4 = 992.0). This pattern contrasts with the decline in positive grant awards by federal agencies under unified party government occur when both the president and 23

26 Congress confront bureaucratic agencies that are either extremely susceptible to, or independent of, political control based on these same considerations (Krause and Zarit 2017). To obtain a more precise leverage over analyzing these estimates, Figure 2 presents the conditional marginal effects of each agency constraint on the average grant loss per award in a given fiscal quarter that is estimated from both models appearing in Table 1. This provides a means for us to evaluate our first hypothesis (H1) that grant losses will be declining when both bureaucratic leader loyalty to the president and presidential capacity to review of agency policy decisions are increasing. Once again, we convert both the conditional coefficients and corresponding standard errors used to compute these linear combinations of coefficients using the percentage change formula used for the tabular entries. Figure 2A displays the marginal effect of bureaucratic leader loyalty to the president (Loyalty) on the percentage change in the average grant loss in our sample, conditional in increasing presidential capacity to review agency policy decisions (Review) from its minimum through the maximum value observed value in our sample at various percentile ranks. At the observed minimum value of Review (Nuclear Regulatory Agency, ), a one unit change in Loyalty is associated with a 259% (3.6 times) greater grant loss, while the effect of bureaucratic leader loyalty to the president declines to 155% (2.55 times) and 106% (2.06 times) between the 10 th and 75 th percentiles, before eventually falling to 73% (1.73 times) and 11% (1.11 times) higher for the 90 th percentile and maximum values of Review (90 th : Department of Agriculture, ; Max: Broadcasting Board of Governors, ). Similarly, Figure 2B presents the same effect using presidential capacity review agency policy decisions (Review) as the main effect on grant loss, while being conditioned on the bureaucratic leader loyalty s to the president (Loyalty). Presidents capacity to review agency 24

27 decisions results in a larger contraction of authority when bureaucratic leaders are not considered to be very loyal to the current administration. Yet, as bureaucratic leader loyalty rises, in conjunction with presidential capacity to review agency policy decisions, grant awards in those agencies will experience much less drastic cuts. Specifically, enhanced presidential review of agency policy decisions yield corresponding declines in grant losses that fall from 639% (7.39 times) for the weakest presidential loyalist bureaucratic leader observed in this sample (NASA Administrator Daniel Goldin: Loyalty = ) to 123% (2.23 times) for the median bureaucratic leader in terms of loyalty to the administration (Secretary of Transportation, James Burnley IV, Loyalty = ), all the way down to 25% (1.25 times) for the most loyal bureaucratic leader (Secretary of Energy, Federico Pena, Loyalty = ). This evidence suggests that presidents are considerably less inclined to contract administrative authority when they have sufficient means to control administrative agencies. The next set of graphs based on the Divided-Unified Party Government Model distinction are analyzed to evaluate our second hypothesis (H2) that grant losses are decreasing at a greater rate as presidential control over agencies is increasing under divided party government compared to unified party government. We obtain clear evidence consistent with H2 that is borne out in both graphs since the conditional slope in grant losses under divided party government is always steeper in a downward manner compared to the conditional slope in grant losses under unified party government. Figure 2C displays the marginal effect of bureaucratic leader loyalty to the president on grant loss, conditional on increases in presidential capacity to review agency policy decisions (expressed in percentage change form). Grant losses fall by 354% (4.54 times) due to a one-unit 25

28 FIGURE 2 The Marginal Effects of Conditionally Increasing Political Control on Grant Retrenchments (Based on Baseline and Divided-Unified Party Government Model Estimates) Figure 2A: Marginal Effect of Loyalty on Grant Loss, Conditional on Review Baseline Model Min 10th 25th 50th 75th 90th Max Presidential Capacity to Review Agency Policy Decisions Figure 2C: Marginal Effect of Loyalty on Grant Loss, Conditional on Review Divided-Unified Party Government Model Figure 2D: Marginal Effect of Review on Grant Loss, Conditional on Loyalty Divided-Unified Party Government Model Min 10th 25th 50th 75th 90th Max Presidential Capacity for Reviewing Agency Policy Decisions Divided Government Unified Government 1, Min 10th 25th 50th 75th 90th Max Presidential Loyalty of Bureaucratic Leader Divided Government Unified Government Note: Marginal effect estimates are denoted as dots based on percentage change conditional effects and corresponding standard errors are represented by shaded 95% confidence bands. 26