The Weak Leviathan?Testing Partisan Theories of Political Influence on Defense Procurement in Congressional Districts: 98th-102nd Congress: The Weak Leviathan? Testing Partisan Theories of Political Influence on Defense Procurement in Congressional Districts: 98th-102nd Congress Boris Shor Harris School of Public Policy Studies University of Chicago January 17, 2006


Recent Headlines: Recent Headlines The Department of Homeland Security will provide $765 million in direct grants to high-risk urban areas for terrorism preparedness But $550 million still handed out on a formula that has a state minimum And $1.6 billion in grants to local responders at administrative discretion


Even after 9/11: The National Defense Authorization Act was considered 10 days after 9/11. Its amendments included: a transfer of land from DoD to establish the Fort Des Moines Memorial Park and Education Center, improvements to the Rocky Flats National Wildlife Refuge, and alterations to contract arrangements at the US Army Heritage and Education Center at Carlisle Barracks, PA In the year after 9/11, Alaska benefited from money for fisheries, gyms, parking garages – all located within defense appropriations bills Even after 9/11


Different Reactions: Different Reactions Journalists: It’s pork. Democrats and Republicans are equally guilty Political scientists We’re not sure politicians are able to affect defense procurement If they do, some are better than others


Why study defense procurement?: Why study defense procurement? Figures in Billons of 2001 Dollars Total: $1.8 trillion 8% Fiscal Expenditures, 2001


Why study defense procurement?: Why study defense procurement? If defense procurement is politically influenced, valuable resources are taken away from military readiness or other nondefense uses (Wheeler) Discretion is surprisingly high in procurement “When goods or services to be procured are … relatively uncomplicated, an agency generally advertises for bids … Agencies have little discretion in such cases… However, when the item to be procured is extremely complex—for example, a new weapons system—agencies need not advertise; instead they negotiate… Negotiation gives the agency considerable latitude in matters of choice.” Arnold (1979)


The Puzzle: The Puzzle How well do existing partisan theories of distributive politics explain the geographic distribution of defense procurement between congressional districts?


Time Period: Time Period 98th-102nd Congresses: 1983-1992 Chosen to avoid major redistrictings Republican presidents throughout this time Democratic House Republican Senate 98-99 congress Major defense procurement changes


Highly Skewed Distribution: Highly Skewed Distribution


Disparities in Defense Procurement: Disparities in Defense Procurement


Variation in Disparities: Variation in Disparities


Party and Distributive Politics: Party and Distributive Politics Do parties matter? Lots of theoretical debate on this point, often revolving around roll call voting and committee representativeness One empirical way to address the question is in the context of distributive politics This is good because dollars are extremely clear as a description of policy and priorities


The Negative Case against Parties: The Negative Case against Parties American elections are individualized affairs Parties may not matter because institutions that mandate competitive contracting and provide post-hoc administrative and judicial review of procurement decisions Mayer 1991 Parties may not matter because this is national security Parties may not matter because they are not cohesive enough, or they aren’t large enough “Bureaucrats do not discriminate on the basis of party, because they are interested in building and maintaining large coalitions” – Arnold 1979


How can parties matter?: How can parties matter? Majority party status provide privileged access to institutional resources—among other goods—that meet members’ electoral and policy needs (Cox and McCubbins 1993). They are durable coalitions Control of agenda Domination of committees Oversight These advantages over the minority party could influence the flows of federal expenditures Benefits can hold fractious coalitions together Cohesion is important, and it was relatively high in the 1980s (Rohde 1991) Reforms gave more power to the caucus and leadership viz a viz committees


Democrats and Defense: Democrats and Defense Do Democrats even like defense spending? Maybe not as much as Republicans (though the differences were small) But, given a specified pot of authorized monies, why not distribute to favored constituencies? Defense procurement means a lot to local economies. Obtaining a contract provides lots of jobs.


Two Recent Arguments: Two Recent Arguments Levitt and Snyder (1995) The Democratic party, through its long period of Congressional dominance in the 1960s-1980s, was able to target its constituencies with particularized benefits (social welfare spending) But not its members Rundquist and Carsey (2002) Defense procurement across states States represented by Democrats on the Armed Services committee did better


Possible Targets of Majority Influence: Possible Targets of Majority Influence Legislators themselves Alvarez and Saving 1997, Balla 2002 Partisan voters Owens and Wade 1984, Levitt and Snyder 1995 Majority party members of imp’t committees Goss 1972, Cox and McCubbins 1993, Carsey and Rundquist 1999 States Cross-district spillovers delegations as aggregation of individual districts


New Data: New Data Relatively unused data set on federal expenditures Expenditure data: CFFR - Consolidated Federal Funds Report More comprehensive than the FAADS data set for certain programs Defense procurement and wages not in FAADS Subcontracting traced to ultimate recipient


Modeling Strategy: Modeling Strategy Analysis of congressional districts as clustered in space (within states) and time (within Congresses) Integration of predictors at multiple levels of analysis Partial pooling strategy: compromise between complete and no-pooling alternatives


Data Set: Data Set 435 districts, 98th to 102nd Congresses (1983 to 1992) Explaining (logged, inflation-adjusted) defense procurement expenditures per capita in a CD for a single Congress Things like: ships, bases, hammers, coffeepots


Predictors at Multiple Levels: Predictors at Multiple Levels Individual Level: District-Congress (435x5) District level variables that change in time Partisan, committee, legislator characteristics District Level (435) District level variables that are time-invariant District demographics, district partisanship State-Congress Level (50x5) State level variables that change in time State demographics, congressional delegations, state political variables (Senate and President) State Level Indicators (50) Congress Level Indicators (5)


Partisan Hypotheses: Partisan Hypotheses H1: Districts represented by Democrats will benefit more than those represented by Republicans H2: Districts with more Democratic voters (proxied by avg pres vote in 84,88) should do better in terms of awards than those with less H3: Democratic members of Armed Services committee should differ from Republicans in providing awards to districts H4: States with House delegations (Senate or House) dominated by the majority party (Democrats) will do better in regards to expenditures


Controls: Controls Other Controls Major Military Installations DC Income Population State capitol Coastal % Urban State wealth/development State/district Area State Population Political Controls Districts Committee membership Seniority States State Presidential election returns Senate state delegation partisan composition


Bivariate Plots – Legislator Party: Bivariate Plots – Legislator Party


Bivariate Plots – Party and Committee: Bivariate Plots – Party and Committee


Bivariate Plots – District Presidential Voting: Bivariate Plots – District Presidential Voting


Bivariate Plots – State Delegation: Bivariate Plots – State Delegation


Individual Level Equation: Individual Level Equation i: District index (1..435) j: State index (1..50) k: State-congress index (1..250) t: Congress index (1..5) y: Estimated spending in a given district-congress it X: matrix of district-congress predictors B: Estimated individual level coefficients District State State-Congress Congress Linear Time Trend District-Congress Error


Group Level Equations: Group Level Equations State-Congress Intercepts U: district-level data W: state-congress level data m: estimated district-level predictors p: estimated state-congress-level predictors : indicator for district i (1 to 435) : indicator for state-congress k (1 to 250)


State and Congress Indicators: State and Congress Indicators State Indicators Congress Indicators


Predicted vs Actual $: Predicted vs Actual $


Results Summary: Results Summary H1: Districts represented by Democrats will benefit more than those represented by Republicans. H2: Districts with more Democratic voters (proxied by pres. vote) should do better in terms of awards than those with less H3: Democratic members of Armed Services committee should differ from Republicans in providing awards to districts H4: States with House delegations (Senate or House) dominated by the majority party (Democrats) will do better in regards to expenditures.


Coefficient Estimates: Coefficient Estimates


Slide35: Effect of Democratic Voters on Defense Procurement SD = 25%


Why are Republican Voters Advantaged?: Why are Republican Voters Advantaged? Democratic policy priorities advantaging nondefense spending and disadvantaging defense spending? Presidential influence? Department of Defense? Prime contractors choosing strategic subcontractors?


Slide37: Effect of Democratic Voters: Defense Employment $


Slide38: Effect of Democratic Voters: Grants SD = 20%


Conclusions : Conclusions New data and model has failed to support existing partisan hypotheses in the distributive politics literature Parties affect defense procurement, but in a counterintuitive way: by rewarding Republican districts Arnold 1979: “There are significant differences between various types of government benefits, even within … ‘distributive policies.’”


Conclusions - Methodology: Conclusions - Methodology Reality is messy; we want to accommodate some of that reality by employing a multilevel modeling design Group-level predictors are important – the technique of multilevel modeling allows us to do this easily


Slide41: “The 2003 Defense appropriations billed included a provision that would make it easier for senators to send out postcards to constituents to notify them of town meetings. The amendment was sponsored by Senator Arlen Specter, Republican of Pennsylvania would eliminate wording that limited the taxpayer-financed postcards to counties of fewer than 250,000 people. ‘This is directly related to the war effort,’ said Mr. Specter … ‘Meeting with the people of Pennsylvania is an important part of our job, including informing them of the war effort.’” - NY Times, 4/2003


Slide42: [In the defense authorization bill] you can find the amendment offered by Democratic Sen. Max Baucus for a grant to Rocky Mountain College in his state of Montana for three Piper aircraft and a simulator, and Republican Sen. Rick Santorum’s $3 million add-on for an unbudgeted artificial lung device for the Army. By the time Congress had finished with the bill in July, House and Senate members had added more than 2,000 of these earmarks …. None, though, had been included in the defense budget put together by DoD and the Office of Management and Budget (OMB). - Winslow Wheeler, Washington Post, August 2004


Full Results: Full Results


Roadmap: Roadmap Large disparities in defense procurement expenditures exist Discuss the existing distributive politics literature Description of my approach Results


Multilevel Modeling: Multilevel Modeling Brief Definition: Direct combination and modeling of data at different levels of analysis That is, taking into account context effects. Advantages: This is more the way the world really looks (nesting) More efficient regression estimates (giving more structure to the data) Better accounting of uncertainty


Pooling Strategies for TSCS Data: Pooling Strategies for TSCS Data TSCS data is inherently structured State-year observations are not independent across space and time What pooling strategy to follow? Complete pooling: strong independence assumption Are states really so similar? Suppresses variation that can be important No pooling: no relationships between observations? Are states really so different? Estimates are too variable Partial pooling is a good alternative. Avoids extremes. Partial dependence of observations


Pooling Differences: Pooling Differences Complete Pooling: μ is fixed No Pooling: μ is unconstrained


Bayesian Estimation: Bayesian Estimation Brief Definition: Prior information + Sample Data = Posterior Distribution of each parameter Advantage: easily estimate multilevel models. But, very computing intensive. Advances in computing power in the last 10 years allow for the estimation of Bayesian models


Further Advantages for TSCS Data: Further Advantages for TSCS Data TSCS data usually required fixed effects but problematic in that they eat up degrees of freedom and do not explain group-level variation (Steenbergen 2002) High collinearity between group-level indicators and time-invariant predictors not an issue (Western and Jackman 1994)


Bayesian Multilevel Models for TSCS Data: Bayesian Multilevel Models for TSCS Data Beck and Katz (2001): Monte Carlo evidence that random coefficient models may be best for TSCS data Western (1998) estimated a similar model for TSCS data to analyze economic growth in OECD countries across time Bayesian approach advantages more accurate forecasts more accurate estimates of time-series effects more realistic accounting of uncertainty Limitations Did not address 3 common TSCS problems


Model Outline: Model Outline Dependent variable is influenced by state-year (“unit level”) predictors Model has varying intercepts: states and years exert “group level” effects of shifting intercepts up and down Group-level variances are estimated, not fixed


Difficulties in the Distributive Politics Literature: Difficulties in the Distributive Politics Literature Little consensus Atomistic view of congressional districts Observations of individual CDs are not independent Common effects of other representatives, Senators, and presidents are shared


Addressing Serial Correlation: Addressing Serial Correlation Serial correlation a significant problem in budgetary data Inclusion of state-specific time trends (Gelman et al, 1995) Other techniques


Heteroscedasticity: Heteroscedasticity Source of it is frequently unit heterogeneity Random intercepts and group-level (state and year) predictors to explain unit differences


Contemporaneous Correlation: Contemporaneous Correlation Year intercepts capture “time shock” effects Group-level predictors get at spatial correlation


Stylized Facts on the Budget: Stylized Facts on the Budget Years are fiscal years, running from October 1 to September 30 Note: Current law prevents outgoing president from transmitting a budget for next fiscal year Revisions to current fiscal year possible as well (Reagan 1981) through reconciliation Sequence: Presidential request, budget resolutions passed, 13 appropriations bills, reconciliation