The Distributive Politics of the Federal Stimulus: The Geography of the American Recovery and Reinvestment Act of 2009 James G. Gimpel University of Maryland [email protected] Frances E. Lee University of Maryland [email protected] Rebecca U. Thorpe University of Washington [email protected] The geographic distribution of funds allocated under the American Recovery and Reinvestment Act (ARRA) of 2009 was poorly targeted to economic need. Areas where the recession‘s downturn was most severe did not receive any more funds on a per capita basis than areas where the recession was least severe. This paper assesses two theories to determine why funds were not successfully aimed where economic suffering was greatest. (1) We examine whether the institutional and political factors at the center of the study of distributive politics can account for the mismatch. (2) We assess the role of ―policy window‖ effects, as policy advocates steered a variety of loosely related proposals through an open window of opportunity. We find some support for both theories, but policy window effects were more important than ―pork barrel politics‖ in explaining why funds were mismatched to need. Paper prepared for presentation at the 2010 Annual Meeting of the American Political Science Association, Washington, DC, September 2-5. 1 ―Mr. Speaker, 50 years ago a Presidential candidate, John Kennedy, said the following: The Chinese use two brush strokes to write the word ‗crisis.‘ One brush stroke stands for danger; the other stands for opportunity. In a crisis, be aware of the danger, but recognize the opportunity.‖ -Rep. Anna G. Eshoo (D-Calif.)1 In theory, one of the most important responsibilities of a national government in a federal system is geographic redistribution. The national government has unique capacity to raise revenue and direct funds to states and localities otherwise unable to provide for their residents‘ needs (Peterson 1995; Callen 2009). These powers are especially important in periods of economic recession, as only the national government can engage in large-scale deficit spending to fund government safety net programs and to stimulate demand in the particular areas with the most troubled economies. In practice, however, there are many reasons to question whether the federal government will successfully perform these functions. Scholars of distributive politics have focused on the political and institutional reasons why the federal government will likely fail to match resources to need. To build legislative support for federal programs, coalition leaders frequently expand the scope of beneficiaries far beyond the neediest areas (Arnold 1979). Members of Congress in key institutional positions, such as leaders or members of the relevant committees of jurisdiction, may extract extra benefits for their constituents (Evans 1994; Frisch 1998; Lazarus and Steigerwalt 2009; Lee 2003). A majority party in Congress can use program benefits to enhance the electoral prospects of its members (Balla, Lawrence, Maltzman, and Sigelman 2002; Bickers and Stein 2000; Lazarus 2009; Lee 2003; Levitt and Snyder 1995; Rundquist and Carsey 2002). Presidents and executive agencies may direct funds to reward or woo political supporters (Bertelli and Grose 2009; Garrett and Sobel 2003; Hamman and Cohen 1997; Larcinese, Rizzo, and Testa 2006; Reeves 1 Congressional Record, February 13, 2009, H1560. 2 Forthcoming). Furthermore, the structure of representation itself is a source of bias, in that small states‘ enhanced representation in the U.S. Senate systematically advantages their residents in the distribution of federal dollars (Hauk and Wacziarg 2007; Lauderdale 2008; Lee and Oppenheimer 1999). A second source of failure in targeting to need lies in the way policymakers exploit windows of opportunity. As John Kingdon (1995) has famously theorized, policy change often rests on a confluence of conditions that, only by serendipity, come together at a particular moment in time. A problem must be widely perceived, often as a result of a crisis or ―focusing event‖ (Kingdon 1995, 94-100). A policy proposal must be ready to hand when a problem comes to attention; generally this means that policy ideas are continually being discussed and vetted among advocacy coalitions, even when immediate prospects for action are poor (Sabatier 1988). Finally, the political environment must be favorable to action, with officeholders in key elective and appointive positions willing to prioritize the issue. Windows of opportunity for major legislation are rare. Much of the time, only incremental change is possible, with public policy characterized by ―punctuated equilibria,‖ in which significant departures occur at infrequent intervals, followed by periods of relative statis (Baumgartner and Jones 1993). On occasions when windows of opportunity open, presidents, lawmakers, administration officials, and policy advocates outside government all seek to capitalize on them to achieve longheld policy goals. Policy windows thus become crowded with proposals: ―[W]hen a window does open, solutions flock to it‖ (Kingdon 1995, 176). Many of these proposals will be, at best, only loosely related to the problem on the agenda: ―Solutions become attached to problems, even though those problems themselves did not necessarily dictate those particular solutions‖ (Kingdon 1995, 177). The difficulty of legislating in an open and decentralized policy process 3 with numerous veto points incentivizes policymakers to make the most of any opportunity for breakthrough. In short, as Rep. Anna Eshoo (D-Calif.) put it in her floor speech in support of the Obama administration‘s economic stimulus package, ―in a crisis, be aware of the danger, but recognize the opportunity.‖ The logic of policy windows is also likely to redirect distributive policy away from any particular vision of funding need. Under this model, the outcomes of public policy are highly unpredictable, with unrelated or loosely related solutions ―riding the wave‖ in the wake of a crisis. Rather than public policy being the product of decisions rationally designed to solve problems, it is a ―garbage can‖ (Cohen, March, and Olsen 1972) shaped by chance and contingency, ―depending on which participants are present, which alternatives are available, and even what catches people‘s eyes‖ (Kingdon 1995, 177). Despite the wide influence of Kingdon‘s theory, the study of distributive policy has failed to grapple with the effect of policy windows on the geographic allocation of federal funds. The central question in the distributive policy literature is one of policy failure: what factors cause misallocation of federal resources? In answer, there is a vast literature on the ways that members‘ incentives to claim credit for distributive benefits, in conjunction with power inequalities inherent in representatives institutions, can misshape federal policy. But policy windows—in which unrelated policies are bundled together as policy entrepreneurs seek opportunity in crisis—have not been subject to systematic empirical investigation in the distributive politics field. This paper analyzes the American Recovery and Reinvestment Act (ARRA) of 2009 as a key test of these two theories of policymaking. Passed in the wake of the financial crisis of 2008 and an ensuing grave recession, this sweeping $787 billion legislation was quickly assembled in 4 the first weeks of the Obama presidency and signed into law on February 17, 2009. As will be shown in the following pages, the geographic distribution of funds allocated under the legislation was poorly targeted to need. Areas where the recession‘s downturn was most severe did not receive any more funds on a per capita basis than areas where the recession was mild. Then our investigation seeks to determine why funds were not successfully aimed where the economic suffering was greatest. Were the institutional and political factors at the center of the study of distributive politics a key cause of the mismatch? Or, on the other hand, were funding outcomes driven by policy window effects, as policy advocates steered a wide variety of loosely related proposals through an open window of opportunity? Our findings reveal that both causes of policy failure played a role, but that policy window effects were more important than ―pork barrel politics‖ in explaining why funds were not targeted to stricken areas. The Geography of Need According to explicit legislative language adopted by Congress, a central goal of the ARRA was ―to assist those most impacted by the recession.‖2 Although the recession beginning in December 20073 was unquestionably national in scope, with serious job losses all across the country, some areas of the country were hit harder than others. [Figure 1 here] As a gauge of regional variation in the depth of the recession, Figure 1 displays histograms of the frequency distribution of average percent unemployment by county for all of 2007 (before the recession was underway) and for the first quarter of 2009 (when the ARRA was 2 American Recovery and Reinvestment Act of 2009, Pub. L. no. 111-5, 123 Stat 115 (2003). The National Bureau for Economic Research dates the start of the recession in December 2007. It has not declared an end date for the recession as of this writing. 3 5 considered and adopted). The national character of the recession is evident in the scale of the downturn. Across all 3,140 counties, average unemployment grew by 54%, from 4.88% to 9.01%; nearly every county suffered job losses. At the same time, regional variability in unemployment also increased dramatically. The standard deviation around the average grew from 4.87 to 8.62. This increased variability is graphically evident in the flatter distribution of unemployment by Q1 of 2009, compared to 2007 (Figure 1). The highly peaked distribution in 2007 reveals low levels of unemployment prevailing across almost the entire country, with only a relative few pockets of high unemployment. Unemployment in early 2009, however, exhibited greater regional disparity, with some places still enjoying relatively low unemployment and others undergoing deep economic difficulty. While both distributions are right-skewed, the 2007 histogram exhibits nothing comparable to the long right tail of the 2009 distribution, indicating counties where the unemployed share of the population exceeded 15% or even 20%. Beyond unemployment, another key indicator of worsening economic fortune is the foreclosure rate. The collapse of a massive housing bubble was one of the precipitating factors in the recession, with toxic investments in mortgage-backed securities burdening the banking system and bankrupting major financial institutions, including Bear Stearns, Lehman Brothers, Washington Mutual, Fannie Mae, and Freddie Mac. For all the national impact of the housing bubble on the financial system, the problem of foreclosures was highly concentrated in geographic terms. Table 1 displays the distribution of the cumulative foreclosure rate (per 1,000 households) by county for the first quarter of 2009. As is apparent, foreclosures were concentrated in a very tiny slice of the country. Fully 70% of counties had a foreclosure rate lower than the national average of 2.2; 90% of counties had a rate below 6. By contrast, half of all foreclosures were concentrated in only 35 counties, among which the foreclosure rate was a 6 stunning 21.5 on average. Counties above the top 90th and 95th percentiles of the distribution are where foreclosures clearly spillover into a serious problem for local governments, businesses, and neighborhoods extending well beyond the individual affected residents. [Table 1 here] Regional inequalities were exacerbated by the fact that some areas severely affected by foreclosures tended also to have high unemployment rates (Pearson r=.21, p<.001). Figure 2 displays a county map of the United States shaded from light to dark to indicate areas of high unemployment, with an overlay of green houses graduated in size to denote counties with the highest rate of foreclosures. The foreclosure crisis was heavily concentrated in Nevada, California, Florida, Ohio, Georgia, South Carolina, and Michigan, commonly in locales that also suffered from high unemployment. Areas of very high unemployment also reached beyond the foreclosure centers, taking in much of Appalachia, the Deep South, and the rest of the industrial Midwest. [Figure 2 here] At the same time, large swaths of the country were afflicted by neither high unemployment nor foreclosures. A wide interior region extending across Montana and North Dakota down the Plains through South Dakota, Nebraska, Iowa, Kansas, Oklahoma, and New Mexico, along with much of Texas and Louisiana had unemployment rates well below the national average and no significant incidence of foreclosures. The Northeast also fared relatively well. In short, the pain of the recession was not experienced equally across the country. When the economy goes sour in a nation as big as the United States, just what is wrong varies by location (Chinni and Gimpel 2010, 136). A program to assist the locations most severely afflicted by the recession would need to be crafted with care and aimed with precision. 7 Targeting in the ARRA The ARRA was omnibus legislation in its most extreme form. It included $267 billion in direct spending, principally for safety-net programs such as extended unemployment benefits, food stamps, Temporary Assistance for Needy Families, health insurance subsidies for unemployed workers, Medicaid, and state fiscal relief. It provided $211 billion in tax cuts, including bonus depreciation and current loss carry-back for businesses, an increase in the Alternative Minimum Tax floor, payroll tax breaks, and tax credits for children, home buying, college, and energy efficiency upgrades. It also funded $308 billion in new discretionary spending for a host of purposes, including scientific and medical research, clean energy, transportation and water infrastructure, public housing, education, and modernization of the electric grid.4 Finally, the Act incorporated transparency provisions requiring that expenditures for grants, loans, and contracts be released to the public on a specially designed website, www.Recovery.gov. Using the information available from Recovery.gov, it is possible to track expenditures geographically to determine whether funds were directed to locations most affected by the recession.5 Unlike most scholarship on distributive politics, we examine per capita expenditures at the county level, rather than the state or congressional district. County-level data afford far 4 Estimated budget impacts of the legislation obtained from the Congressional Budget Office, ―H.R. 1, American Recovery and Reinvestment Act of 2009: Cost Estimate for the Conference Agreement for H.R. 1,‖ February 13, 2009, http://www.cbo.gov/doc.cfm?index=9989&zzz=38482. 5 We employed county-level datasets compiled by ProPublica. ProPublica drew its data from the Recovery.org website, but it then cleaned the data to correct for errors. Most importantly, ProPublica adjusted the amounts allocated for prime recipients so that they did not include amounts awarded to sub-recipients. More information on ProPublica‘s methodology is provided by LaFleur (2009). 8 better granularity for assessing how well funds are targeted to economic need, particularly in a recession that has had such disparate effects on different sectors of the economy and people at different education levels.6 We exclude state capital counties from the analysis: because these counties receive substantial allocations for programs that will subsequently be redistributed across the state, the per capita funds going to these locations are not appropriately comparable to other counties. Our data include all funds allocated through the first quarter of 2010. By that point, 62% of the total funds provided for by the ARRA had been spent (Weise 2010). It should be noted, however, that the project expenditures reported on Recovery.org do not reflect the direct spending and tax reductions in the ARRA, some of which were aimed at unemployed workers and struggling families. Recovery.org tracks the grants, loans, and contracts funded by the stimulus. Unlike the safety net programs funded by direct spending, these are the programs that were often touted as job creators. In his remarks upon signing the ARRA into law, for example, President Obama described such projects as follows: Because of this investment, nearly 400,000 men and women will go to work rebuilding our crumbling roads and bridges, repairing our faulty dams and levees, bringing critical broadband connections to businesses and homes in nearly every community in America, upgrading mass transit and building high-speed rail lines that will improve travel and commerce throughout the nation.‖7 In the context of the congressional debate over the stimulus package, these expenditures were justified primarily in terms of their effect on employment. Many are identified around the country with signs bearing the ARRA emblem to remind Americans, in Obama‘s words, of ―our commitment . . . to investing your tax dollars wisely to put Americans to work doing the work 6 Analysis at the zip code might be even better, but much of the demographic information that we find relevant for understanding the distribution of ARRA funds is not available at that unit of analysis. 7 ―Remarks by President Barack Obama at the Signing of the Economic Stimulus Bill,‖ Federal News Service, February 17, 2009. 9 that needs to be done.‖8 Were these projects well-positioned to put people back to work in areas of concentrated unemployment? [Table 2 here] Table 2 presents frequency distributions for per capita ARRA spending by county, both for the legislation as a whole, as well as for the transportation and water infrastructure projects separately.9 The ARRA was far more universal in character than most federal programs (Stein and Bickers 1994), probably not surprising given its extraordinary scope. Every county received at least some funding from the legislation; 80% of counties obtained at least minimal support for local transportation and water projects. Despite the universalism, however, there was significant targeting to particular counties. The distributions for both total spending and infrastructure projects are highly right skewed, meaning that a relative few high values pull the mean substantially higher than the median. For all ARRA spending as of the first quarter of 2010, counties in the 90th percentile of per capita expenditures received two and a half times as much funding per resident as the median county. The funds for infrastructure were even more directed. Counties at the 90th percentile of per capita transportation and water project funding received nearly six and one half times as much as the median county. Clearly, the ARRA did target federal resources to particular locations, just the wrong ones from a need standpoint. Figure 3 displays total per capita spending by county across the United States as of the first quarter of 2010. Counties in this map are color coded using ground cover shades typically used to designate areas on a continuum from ―arid‖ to ―extremely fertile.‖ 8 ―Remarks by President Barack Obama, Transportation Secretary Ray LaHood, and Vice President Joe Biden; Subject: American Recovery and Reinvestment Act,‖ Federal News Service, March 3, 2009. 9 Transportation and water infrastructure expenditures were identified as projects managed by the federal Department of Transportation, the Army Corps of Engineers, or the Bureau of Reclamation. 10 In this case, the fertile (darker green) colors indicate counties with the highest per capita allocations under the ARRA; the desert (tan) colors reflect areas with the lowest per capita allocations. The five different colors shown reflect natural breaks in the data‘s distribution. Most counties fall into categories shown in the two desert tones, reflecting the fact that most received less than the mean per capita allocation of $870. Counties coded as lighter and darker green are scattered throughout the United States, but they are especially prevalent in the Plains and the Mountain West, in many cases in counties that had very low levels of unemployment and foreclosures. There is certainly no special concentration of funding for areas hardest hit by the recession. With the exception of Inyo County, a remote, sparsely populated county in east central California, all of California‘s counties received per capita allocations in the bottom two tiles on the map (Figure 3). Similarly, despite its high unemployment and foreclosures, nearly all counties in Florida fell into the most ―arid‖ category. There were very few concentrated allocations of money anywhere in the struggling Industrial Midwest. Multivariate Evaluation of ARRA Distribution Multiple regression offers a more definitive test of the relationship between economic need and the geographic distribution of ARRA funds. The dependent variable is the natural log of per capita ARRA funds by county.10 As is common with geographic data, standard ordinary least squares approach is inadequate here. OLS makes the assumption that the observations i on the dependent variable are independent of each other, such that E(eiej)=0. In the case of our data, however, the observations are spatially dependent, in that the per capita amounts allocated to one 10 The natural log transformation was needed to correct for the non-normal distribution of the dependent variable, as evident in Table 2. With the natural log transformation, the data are normally distributed with means and medians close together. 11 county are likely to be related to those allocated to neighboring locations. If unattended, the resulting correlated errors could result in faulty inferences based on t and F statistics. To address this statistical problem, the models include a correction for the spatially correlated errors. Additional detail on the diagnostics we employ and the specific correction method we implement is provided in Appendix A. Four indicators of economic need are evaluated in our data analysis. First, two variables measure the depth of the recession at the time the legislation was drafted by Congress and passed into law: the cumulative foreclosure rate for the first quarter of 2009 and the change in the unemployed percent between the 2007 average and the first quarter of 2009. Second, two variables identify locales with long-standing economic difficulties: median income and the percent unemployed in 2007.11 Results for the influence of these need indicators on per capita ARRA allocations are displayed in Table 3. [Table 3 here] The findings simply reveal no meaningful targeting to economic need. Areas hardest hit by the recession, as measured by the scale of their job losses or their rate of foreclosures, received no additional stimulus funds across all the projects funded by the ARRA. Indeed, the coefficient for change in unemployed % is negative and statistically significant (p<.001), meaning that areas with bigger increases in unemployment received slightly less stimulus money per capita. The model estimates that counties at the 90th percentile of change in unemployed % received $6 per capita less than counties at the 10th percentile. Low-income areas do not receive more ARRA project funds; instead, the coefficient, albeit statistically insignificant, is positive. 11 Although there is modest multicollinearity among some of these variables, regression results do not change if any particular variable is dropped from the model. Percent in poverty was tested as an alternative for median income, with no change in the results. 12 Only one indicator suggests any project funding being directed to unemployment: Areas that had high levels of unemployment before the recession began (in 2007) received more stimulus money (p<.001), but the effect is minimal. According to model estimates, counties at the 90th percentile of unemployment in 2007 received a minuscule $4 more per capita than counties at the 10th percentile. The results from the model of expenditures for transportation and water infrastructure similarly indicate little targeting to economic need. The counties with the most severe job losses during the recession were granted no additional support compared to counties with fewer freshly unemployed residents. Furthermore, locations suffering from long-standing economic difficulties did not receive any additional funding for transportation and water infrastructure. Instead, the coefficient for median income is positive (p<.05), indicating that high income areas received more assistance than low income locales, a difference of about $25 per capita between the 10th and 90th percentile. Unemployment levels in 2007 had no statistically significant effect, and the coefficient is negative. Although the results in Table 3 do suggest that areas with high foreclosures received a boost (p<.05), this finding does not hold up in the full model of infrastructure spending presented later in the paper, so we do not place much confidence in it. Taken together, the regression estimates confirm systematically what was evident visually in the map in Figure 3. Although ARRA funds were targeted to some counties, the beneficiaries were not those most adversely affected by this economic downturn. Policy Windows vs. the Pork Barrel We have taken a glimpse at some key data showing that the ARRA's performance has not met its promise. Now the task is to account for the gap between aspirations and outcomes in 13 legislation intended to assist those hit hardest by the recession. Traditional scholarship on pork barrel politics would look for explanations for the ways institutional power shapes access to political benefits. An alternative theory, one that has received no attention in the literature on distributive politics, is that the geographic distribution of stimulus funds may have been diverted away from need as a result of policy window effects. The reason that the ARRA did not direct scarce resources toward the areas hardest hit by the recession is because the legislation became a vehicle for a multiplicity of unrelated policy goals as political entrepreneurs took advantage of a gaping window of opportunity. Testing the Theories One of the difficulties of testing policy windows theory is its ―indeterminacy‖ (Mucciaroni 1992, 482). In the abstract, of course, it is not possible to identify the universe of alternative policy goals that might potentially be ―coupled‖ (Kingdon 1995, 173) with the complex problem of an economic downturn. Indeed, any government spending program or tax reduction could be framed as having a stimulative effect on the macroeconomy. In particular instances, however, it is possible to identify the range of policy proposals on the national agenda at the time a piece of legislation is crafted and to employ systematic indicators to measure how those policies would likely drive the geographic distribution of funds. Use of these indicators can then afford a falsifiable, empirical test of the policy windows model. If such indicators can better explain patterns in distribution than economic need, it stands to reason that these alternative policy purposes were driving geographic allocation. From such a finding, one might reasonably infer that the economic downturn opened a policy window for political entrepreneurs to pass legislation for purposes that go well beyond putting unemployed 14 people back to work, or putting a floor under a sinking housing market. Below we identify and explain the indicators we employed to identify the alternative policy goals likely to steer the distribution of federal funds in the ARRA.12 At the time the ARRA was written, the Obama administration had a number of other policy ambitions in addition to responding to the nation‘s economic difficulties. One goal, widely touted during the 2008 campaign, was to promote research and development of clean energy sources. To the extent that this goal was incorporated into the ARRA, one would expect to find centers of science and technology around the country receiving more funds than other areas. Residents of these areas would be in a better position to apply for the research grants and to develop development project proposals capable of obtaining funding support. A second top administration priority was to foster transformations in health care technology, in particular the transition to electronic medical records. If the stimulus presented a window of opportunity to pursue these goals, one would expect funds to flow to counties in which higher proportions of the workforce are employed in computing and science or to counties with large percentages employed in health and social services. Also, if a share of research grants went to university scientists, the percent enrolled in college and universities would proxy for the presence of large higher education facilities. Another policy goal—a priority not just for the administration but for many members of Congress—was to improve the nation‘s infrastructure. Although enhancing the nation‘s infrastructure was commonly presented as a means of addressing unemployment, it also has the potential to drive the distribution of funds as an end in itself. In other words, if the policy goal is upgrading and repairing infrastructure—rather than using infrastructure projects as a tool to 12 Appendix A identifies all variables and sources used in the quantitative analysis. 15 alleviate unemployment—one would expect funding to stream toward existing public infrastructure all around the country, rather than toward infrastructure projects in areas with relatively high unemployment. Our models include five measures to capture the presence of infrastructure at the county level: interstate highway mileage, U.S. highway mileage, national parks, water area (to measure need for bridges and other water projects), and medium and small airports.13 Two other policy goals were very prominent during the development of the ARRA. A number of senators saw the stimulus as an opportunity to increase support for medical and scientific research. In the end, the legislation included fully $10 billion in new resources for the National Institutes of Health and $3 billion for the National Science Foundation. To take account of this goal, the model includes a dummy variable for the locations of the National Institutes of Health (in Maryland and North Carolina). In addition, percent enrolled in colleges and universities measures the medical and scientific research funds flowing to educational institutions, just as it gauges allocations for energy research. Finally, given budgetary shortfalls among governments at all levels, support for state and local programs, especially education, was a high priority in Congress. To measure the extent to which stimulus funds subsidized preexisting state and local government activities around the country, regardless of how seriously these areas were affected by the recession, we included the percent employed in state and local government. 13 Medium and small airports have greater needs for federal subsidies (for security and other purposes) than large hub airports, because they cannot capitalize on economies of scale. Our expectation was that the presence of these airports might attract more federal per capita stimulus dollars than large airports. However, we also experimented with including a variable for large airports in the model, as well, finding that it had no effect. 16 Given the incredible scope of the legislation, further investigation into the policy goals of key administration and congressional actors as they developed the stimulus would undoubtedly reveal many additional policy ―solutions‖ that became ―coupled‖ with the economic recession. Such an investigation would identify other relevant variables (though there are some data availability challenges in studying county-level demography). However, the measures we have included take account of a number of the most prominent and costly alternative policy goals that may account for the geographic distribution of ARRA funds. To test for a traditional pork barrel account of the distribution of stimulus funds, a series of institutional and political variables are employed. First, representation on either of the two House committees with jurisdiction over the grants, loans and projects funded by the ARRA might affect the amount a county receives: dummy variables reflect whether a member representing the county served on House Appropriations or House Energy and Commerce.14 Membership on the House Transportation and Infrastructure Committee may also have allowed members to influence ARRA infrastructure spending, even though the legislation was not referred to this committee. The ARRA did not receive formal consideration by Senate committees, but membership on key committees, Senate Appropriations or Environment and Public Works, might have permitted some senators informal sway over the legislation. Three variables are designed to measure whether the majority party reserved special benefits for itself in the legislation: % of the House delegation Democratic, % of the Senate delegation Democratic, and the county‘s Democratic share of the presidential vote ’08. Finally, two variables measure how the structure of representation might advantage or disadvantage 14 House Ways and Means and Senate Finance also considered the legislation, but had jurisdiction over the direct spending and tax programs, not the grants, contracts, and loans tracked geographically in the Recovery.org data. 17 counties. First, the greater representational power of small states in the Senate is measured by the reciprocal of state population (1/state population).15 Second, more populous counties, with more House representation, might gain greater benefits. Results of Estimation The complete regression analyses of total ARRA expenditures and of infrastructure projects are detailed in Appendix B.16 Our estimates provide confidence that the findings are robust to alternative specification, even though we explain only a modest amount of the total variation in the dependent variable. Coefficient estimates from the complete models (which include all the variables tracking pork barrel and policy windows effects, along with economic need) are by-and-large consistent with those from the partial models testing the pork barrel and policy windows theories separately. Because the dependent variable is log transformed, however, the regression coefficients are somewhat difficult to interpret in substantive terms. For a better sense of model findings, we present Table 4 to show how each of the statistically significant coefficients is predicted to affect a county‘s per capita receipts (unlogged), given the range of the variables. [Table 4 here] Clearly, the traditional pork barrel account of bias in distributive outcomes fares rather poorly. In most cases, counties represented by members on the committees with jurisdictional claim to the ARRA did not obtain any extra benefits. Counties represented by House and Senate Appropriators gained no appreciable boost in funding. Representation on Senate Environment 15 Prior work has shown this to be the most appropriate functional form for estimating the small state advantage in distributive policy (Lauderdale 2008). 16 Because the models in Appendix B are for infrastructure projects, only the policy window variables relevant to infrastructure are entered. 18 and Public Works had a small positive effect, estimated at $1.08 for all stimulus programs and at $1.60 for infrastructure projects (p<.01). Representation on the House Energy and Commerce Committee generated an additional $1.08 per capita (p<.001) in the model for all ARRA programs. Small or nonexistent benefits for committee members are consistent with the perception that committee autonomy in the contemporary Congress has been weakened (Sinclair 2000). In addition, counties represented by Democrats in the House and Senate did not obtain any statistically significant funding increase. In fact, in the model of infrastructure spending, % Senate delegation Democratic exhibits a negative coefficient, albeit statistically insignificant. These across-the-board negative results for congressional representation also make sense in that the legislation included no earmarked projects, diminishing Congress‘s ability to direct funds toward particular states and districts. Notably, however, there is a partisan tilt toward counties that were stronger for the Democratic party in 2008. All else equal, counties at the 90th percentile of support for candidate Obama received between $35 and $36 more per capita in both overall funding and infrastructure projects than counties at the 10th percentile (p<.001). Although much of the early work on distributive politics found no consistent party effects, this finding corresponds with recent scholarly analysis of the more partisan contemporary period. The new work documenting partisanship in administration is especially relevant because most of the grants, loans, and contracts funded by the ARRA are in discretionary programs overseen by administrative agencies (Bertelli and Grose 2009; Garrett and Sobel 2003; Larcinese, Rizzo, and Testa 2006; Reeves Forthcoming). 19 The results also testify to the continuing relevance of the Senate‘s representational scheme for understanding distributive outcomes. Counties in small-population states received larger per capita allocations than counties in large-population states for both all ARRA funds, as well as for infrastructure spending. Between the 10th and 90th percentiles on the small state advantage measure (1/state population), the difference is just under $6 per capita. However, the 10th-90th percentile range of small state advantage understates the effect. Because there are relatively few counties in the smallest population states, the 90th percentile does not include any counties in the nation‘s 12 least populous states. If one compares counties across the full range of state population, the estimated difference between per capita allocations in California counties as compared to Wyoming counties is around $19 for both models. Consistent with previous work (Lee 1998), the states with the smallest populations (those with only one House district) gain the biggest advantages from Senate representation. Of course, even a $1 per capita reduction across a state with the population of California or New York involves a significant sum. Counties with more House representatives do not obtain any increases in per capita funding. As an account of the geographic distribution of ARRA funds, the ―policy windows‖ thesis gains considerable support from the models. Unquestionably, funds were far better aimed at existing infrastructure than unemployed persons. Counties at the 90th percentile of interstate highway mileage received $50 more per capita than counties at the 10th percentile. Counties at the 90th percentile of U.S. highway mileage gained fully $86 more per capita than counties at the 10th percentile. Extensive amounts of water area attracted additional infrastructure funds. The presence of a national park also boosted receipts by between $1 and $2 per capita in each of the models. These findings reveal that ARRA funds were better allocated for the policy purpose of 20 improving infrastructure throughout the nation than for redressing recessionary effects. The geographic targeting of funds was more toward standing investments in public infrastructure than to pockets of high unemployment or collapsing housing values. Other policy goals beyond infrastructure maintenance also exerted an effect on the geographic distribution of ARRA funds. Consistent with the expectation that the administration‘s initiatives promoting research and development in clean energy and medical technology would have an effect, the results indicate that funds were targeted to centers of science and technology. Areas with large institutions of higher education received substantially more resources. Counties at the 90th percentile of college enrollment received $17.46 more per capita than counties at the 10th percentile (p<.001). In addition, counties at the 90th percentile of employed in computing and science received $2.71 more per capita than counties at the 10th percentile (p<.001). Counties housing National Institutes of Health facilities received an additional $2 per capita. Finally, it is clear that ARRA funds were broadly distributed to sustain state and local government services in all areas, regardless of their economic distress. Counties with a larger share of residents employed in state and local government received more money, with counties at the 90th percentile receiving $11.45 more per capita than counties at the 10th percentile. Taken together, our findings support the conclusion that that both pork barrel and policy window effects drove the distribution of ARRA funds to a greater extent than the economic needs present from the recession. Remarkably, even after controlling for the political, institutional and policy factors affecting distribution, areas where the recession dug deepest did not receive additional funds. Only one indicator of need had the expected effect and in only one of the two complete models. Counties at the 90th percentile of unemployed % 2007 received 21 $4.23 per capita more ARRA funds than counties at the 10th percentile (almost precisely the same effect documented in the simple need model reported in Table 3). The areas of the country that were hardest hit by the recession in terms of job losses after 2007 or by home foreclosures in the initial months of 2009 received no additional support from the ARRA. Of the two models of distribution, then, we conclude that the policy windows thesis fares better than the distributive politics account. Congress members advantaged by committee jurisdiction or majority party membership did not procure more ARRA funds for their constituencies. To be sure, there is non-trivial evidence that allocations went more toward Democratic areas of the country, as measured by their presidential vote in 2008. And Senate representation had its characteristic effect of advantaging less populous states, in particular those with the very smallest populations. Generally speaking, though, the policy window variables have greater effects on the geographic distribution of benefits than the variables in the pork barrel model. Taken together, the findings reveal that there were systematic policy reasons for some areas receiving more ARRA funds per capita than others. However, those reasons were not that some areas were experiencing greater economic hardship. Instead, particular counties received more federal support because such a distribution better advanced policy goals in promoting clean energy, fostering medical and scientific research, repairing existing infrastructure, and subsidizing state and local government services throughout the country. Policy Windows and Distributive Politics ―I believe White House Chief of Staff Rahm Emanuel characterized this Democrat spending bill best when he said, ‗You never want a serious crisis to go to waste. And what I mean by that is an opportunity to do things you think you could not do before.‖ 22 --Rep. Cliff Stearns (R-Fla.)17 The results of this study of the ARRA of 2009 point toward a fruitful synthesis of two divergent theoretical perspectives on policymaking. The study of distributive politics has centered on the way basic institutional facts—committee influence, party control, representational power—can misdirect federal policy. Scholars of distributive politics, however, have not drawn upon insights from the literature on the policy process. But just as federal distributive policy is vulnerable to pork barrel politics, it can also fail to accomplish stated policy goals because of the crowding of policy windows, as various political entrepreneurs successfully exploit available opportunities to push through legislation advancing disparate policy objectives. Work on distributive policy needs to be informed by scholarship on the public policy process, broadly speaking. In this case, the ARRA failed in terms of focusing resources on the areas of greatest economic suffering, becoming instead a vehicle for an amalgam of different policy purposes. Our findings, of course, do not speak to the substantive merit of the various public policies funded by the stimulus. They may well be worthy investments of government resources. It is also undoubtedly true that the legislation had macroeconomic effects of stimulating employment and demand, another central goal of the legislation. But analyzed in terms of distributive outcomes, the legislation did not do nearly as much as it could have to alleviate economic hardship where it was most severe. A long history of research in distributive politics has established that there are significant institutional obstacles to geographic redistribution focused on need in the U.S. system. Pork barrel politics may have played a role in the ARRA, especially with respect to the Senate‘s small state advantage and the favorable tilt toward Democratic counties. But our analysis of the 17 Congressional Record, February 13, 2009, H1580. 23 ARRA highlights how a legislative process, particularly one increasingly dominated by omnibus bills (Krutz 2001), may be especially vulnerable to the kind of unpredictable policy outcomes that concerned Kingdon (1995). Efforts to distribute federal resources to needy areas can be undermined not only by pork barrel politics, but also by the efforts of political entrepreneurs to hitch their bills to any legislative train leaving the station. At the same time, this study suggests that scholars of the policy process could benefit from more attention to distributive policy. Scholars focusing on the policy process have not taken much interest in policy decisions relating to geographic distribution, instead paying more attention to other policy types, especially regulation (e.g., Kingdon 1995, Sabatier 1988). Distributive policy, however, offers a good testing-ground for theories of the policy process. One of the advantages of this policy type is the presence of outcomes that can be compared on a metric (dollars), easing the task of systematic analysis. In this study, we found that it was possible to obtain empirical indicators that allowed us to test for the ways different policy goals drove the geographic distribution of ARRA funds. In so doing, our results yielded support for a longstanding theory of the policy process that has been difficult to empirically evaluate. Reflecting on the difficulty of empirically verifying theories of the policy process, Sabatier (1991, 153) wrote, ―The paramount task facing policy scholars during the 1990s will be to apply these theories in a variety of empirical settings.‖ A review of the subsequent literature, however, does not find many new efforts to test for the effects of policy windows, even though ―nothing has changed the direction of thinking‖ about their importance to understanding policymaking (John 2003, 482). In the case of the ARRA of 2009, policy windows theory provided more guidance in accounting for the distribution of federal dollars than the traditional factors at the heart of the distributive politics literature. 24 Finally, we hope that our findings will inform contemporary political debate over the legislation. It is ironic that the dominant criticisms of the ARRA that have characterized American political discourse since the passage of the legislation have centered on its purported politicized funding distribution. As early as February 8, 2009, the New York Times reported that the term ―porkulus‖ had appeared in nearly 70 instances in the Lexis-Nexis database since Rush Limbaugh had first used the moniker.18 The results of this analysis do not entirely discredit charges of pork barreling, but they provide considerably more support for the less influential criticism raised by Rep. Cliff Stearns (R-Fla.), quoted above. Rather than a uniquely egregious exercise in pork barreling, the ARRA brightly illuminates the politics of making the most of crisis. 18 ―Porkulus,‖ Idea of the Day: Must Reads from the Week in Review Staff, Weblog, New York Times, February 8, 2009, http://ideas.blogs.nytimes.com/2009/02/08/porkulus/. 25 Table 1: County Foreclosure Rate, per 1000 Households Mean Median Minimum Maximum Percentiles 10 20 30 40 50 60 70 80 90 95 97 98 99 2.2 0.5 0.0 54.8 0.0 0.0 0.1 0.2 0.5 1.0 2.0 3.3 5.9 9.4 12.8 15.3 21.0 26 Table 2: Distribution of Economic Stimulus Funds Across Counties Transportation and Water Funds Per Capita through Q1, 2010 3083* $200 $62 $675 13 269 $0 $19,485 $0 N Mean Median Std. Deviation Skewness Kurtosis Minimum Maximum Percentiles 10 Total Per Capita Funds through Q1, 2010 3083* $870 $570 $1,248 8 85 $15 $21,339 $271 20 $354 $0 30 $425 $19 40 $496 $40 50 $570 $62 60 $675 $90 70 $807 $126 80 $1,025 $201 90 $1,485 $401 *Counties encompassing state capitals are excluded because they receive large allocations for state programs that are then redistributed across the state. 27 Table 3: Targeting to Need in the ARRA Total Per Capita Spending Transportation and Water Funds Per Capita Coeff. (Std Err) Coeff. (Std Err) Depth of recession Foreclosure rate, Q1 '09 Change in unemployed %, '07-Q1 '09 0.004 -0.037** (0.004) (0.007) 0.019* 0.019 (0.011) (0.022) Long term economic difficulty Median Income Unemployed % '07 0.0004 0.045** (0.0001) (0.011) 0.001* -0.004 (0.0005) (0.032) Constant Lambda R-squared 6.347** 0.273** 0.065 (0.100) (0.026) 3.002** 0.297** 0.068 (0.297) (0.026) *p<0.05; **p-<0.01 Note: Dependent variable is total logged per capita ARRA allocations per county through the first quarter of 2010. N=3,084. State capital counties are excluded. Table 4: Determinants of ARRA Funding Receipts Hypothesis Pork barrel Policy window Variable Change in x (from, to) Change in expected per capita receipts for all ARRA programs Change in expected per capita receipts for infrastructure only House Transportation & Infrastructure 0, 1 n.s. n.s. House Appropriations 0, 1 n.s. n.s. House Energy & Commerce 0, 1 $1.08 n.s. Senate Environment & Public Works 0, 1 $1.10 $1.60 Senate Appropriations 0, 1 n.s. n.s. % House delegation Democratic 0, 100 n.s. n.s. % Senate delegation Democratic 0, 100 n.s. n.s. Democratic share presidential vote ‗08 24.11, 59.36 Small state advantage (1/state pop) .411, 6.56 Total # of House members $35.40 $35.96 $5.84 $5.80 1, 4 -$2.86 n.s. National park 0, 1 $1.22 $1.78 Medium & small airports 0, 1 n.s. n.s. Interstate mileage 0, 50.15 $50.22 $50.40 U.S. highway mileage 0, 86.07 $86.21 $86.60 Water area (logged %) 2.33, 6.91 National Institutes of Health 0, 1 College enrollment (%) 8.34, 25.58 Employed in health & social services (%) 7.85, 14.45 Employed in computing and science (%) .48, 3.09 n.s. $4.89 $2.00 -- $17.46 -- n.s. -- $2.71 -- 1 Need Employed in state and local government (%) 8.97, 20.20 Change in unemployed %, ‘07-Q1 ‘09 1.36, 7.23 Foreclosure rate (per 1000), Q1 ‗09 $11.45 -- -5.78 n.s. 0, 5.79 n.s. n.s. Unemployed (%), 2007 2.96, 7.04 4.23 n.s. Median income 32,019, 57,602 n.s. n.s. Note: The entries in the second column reflect the values at the 10th and 90th percentiles of the continuous variables and 0 and 1 for dichotomous variables. The entries in the third column employ the regression coefficients reported in the full model of Table 4 to predict the change in total per capita receipts for all ARRA programs associated with the changes shown in the second column. The entries in the fourth column employ the regression coefficients reported for the full model in Table 5 to predict the change in per capita receipts for infrastructure associated with the changes in the second column. Figure 1: Frequency Distribution of County Unemployment Rate, 2007 and 2009 Figure 2: Unemployment rate and Foreclosures, by County, First Quarter of 2009 1 Figure 3: Total ARRA Funds through Q1 2010 Per Capita, by County Appendix A: Spatial Error Estimation Proper model specification in ordinary least squares can often account for spatial dependency in error terms, but not always. When unobserved factors exert influence on the dependent variable, the inability to account for these latent forces will yield spatially correlated errors (LeSage and Pace 2009; Anselin 1988). The procedure to test for spatial dependency is to start by estimating the regression model with ordinary least squares and examine the resulting residuals using the test statistic, Moran's I (Anselin 1988, 16-19). A positive and significant Moran’s I indicates clustering in space of similar values. Such tests assume that the researchers have first explored the data and specified a particular neighbor structure to gauge spatial dependency. In our tests, we utilize a binary contiguity between adjacent units, expressed as 1, 0 values. The contiguity is first order so that if two spatial units have a common border, they are assigned a value of 1, and if non-contiguous, a value of 0.19 This makes sense for county-level data analysis because boundaries between counties are arbitrarily drawn, and the stimulus funds distributed to adjacent counties are likely to be related. Commonly capital projects, for instance, serve more than just a single jurisdiction, and their construction may draw upon local human and capital resources from several neighboring jurisdictions.20 Starting from the basic linear regression model: 19 By convention, the diagonal elements of the spatial weights matrix are set to zero, and to facilitate interpretation, the elements of each row are standardized such that they sum to one. 20 Choice of this spatial structure was dependent upon exploratory data analysis, and higher order contiguity matrices in which more distant neighbors are considered linked, were also considered, but ultimately found to be unnecessary to correct the least squares estimates. An additional merit of the first order autoregressive structure is its simplicity. 1 the spatially autoregressive error can be written as: where λ is the spatially autoregressive coefficient, W is the binary contiguity weights matrix, and µ is an error term that satisfies the requirements of independent and identical distribution (i.i.d.) and homoscedastic variance (Anselin 1988, p. 100-101). Tests of our OLS residuals for each of our models showed modest but statistically significant positive spatial dependency. In other words, the per capita amount allocated to nearby locations could be used for prediction, meaning that the data generation process is spatially autoregressive (Anselin and Bera 1998). Related diagnostics indicated that the appropriate course of action was to estimate with maximum likelihood the following linear regression model with a spatially autoregressive error: Where λ is the coefficient for the spatial error lag. The value of the dependent variable for each location is therefore influenced by the errors at neighboring locations through the term (I - λW)1. The resulting coefficients will have appropriate standard errors. 2 Appendix A Table 1. Variable Definitions, Means and Standard Deviations Variable Name Mean Std. Deviation Natural Log of Total Cumulative Spending Per Capita through Q1, 2010 6.4 0.8 Natural Log of Total Infrastructure Spending Per Capita through Q1, 2010 includes Highway Projects through U.S. DOT and Water Projects through U.S. Department of Interior, and U.S. Army Corps of Engineers. County Represented by a House Transportation Member 3.6 2.2 0.4 0.6 County Represented by a House Approps Member 0.4 0.6 County Represented by a House Energy and Commerce Member 0.3 0.5 County Represented by a Senate Environment and Pub Works Member 0.3 0.5 County Represented by a Senate Approps Member 0.6 0.5 Total Members of House Serving County 2.4 1.3 % Democratic House Members Serving County 47.9 37.7 % Democratic Senators Serving County 49.3 42.2 % Democratic Presidential Vote 2008 41.3 13.7 % Unemployment 2007 4.9 1.8 Foreclosure Rate, Quarter 1, 2009 2.1 4.3 Change in Unemployment Rate 2007 to Q1 of 2009 4.2 2.4 43,985 11,436 NIH Facilties in County 0.0 0.0 Major National Parks in County 0.0 0.1 Medium and Small Hub Airports in County 0.0 0.2 Mileage of Interstate Highway in County 18.7 31.7 Mileage of Major US Highway in County 43.4 37.3 Natural Log of % Water Area in County 4.5 1.8 % College+Grad School Enrollment 2000 16.0 9.6 % Health & Social Services Employment 11.1 2.7 1.6 1.4 % State & Local Gov't. Employment 14.0 5.0 % Unemployment Quarter 1, 2010 10.4 3.8 % Unemployment Quarter 4, 2009 9.1 3.4 % Unemployment Quarter 1, 2009 9.0 3.5 63,761,534 249,376,860 Median Household Income % Computer & Scientific Employment Cumulative Total Amount Awarded for All Programs through Q1, 2010 3 Cumulative Total Amount Awarded Per Capita through Q1, 2010 Infrastructure Total Amount Awarded through Q1, 2010 870.0 1,248.1 10,180,998 44,350,934 200.4 675.4 Infrastructure Total Amount Per Capita through Q1, 2010 Total N =3084 Data Sources For award amounts: ProPublica. http://www.propublica.org/ Unemployment statistics: Bureau of Labor Statistics. Local Area Unemployment Statistics (LAUS). Foreclosures: RealtyTrac. http://www.realtytrac.com/ Committee Memberships: Almanac of American Politics. Highway Mileage: U.S. DOT. Bureau of Transportation Statistics. U.S. National Transportation Atlas. http://www.bts.gov/publications/north_american_transportation_atlas_data/ Airports: U.S. DOT, Federal Aviation Administration. Air Carrier Activity Information System; and the U.S. Geological Survey and National Atlas of the United States at NationalAtlas.gov, provided by the U.S Department of the Interior, U.S. Geological Survey. National Parks: U.S. Department of the Interior, U.S. Geological Survey. Water Area: U.S. Department of Agriculture. Economic Research Service. National Institutes of Health: U.S. Department of Health and Human Services, National Institutes of Health. Congressional District Data: Clerk of the U.S. House of Representatives. For 111th Congressional District political boundaries. U.S. Department of Commerce, Bureau of the Census. Geography Division. Employment and Income Data: U.S. Department of Commerce, Bureau of the Census. For 2008 Presidential Vote: David Leip's Atlas of U.S. Presidential Elections. http://www.uselectionatlas.org/ For GIS boundary files and coverages: Environmental Systems Research Institute, Inc. ESRI Data & Maps DVD licensed for use with ESRI® software. 4 Appendix B, Table 1: Predicting Total ARRA Receipts, Policy Windows and Pork Barrel Models National Institutes of Health National park Medium and small airports Interstate mileage US highway mileage Water area (logged %) College enrollment (%) Employed in computing & science (%) Employed in health & social svc (%) Employed in state and local gov’t (%) Lambda Constant House Transportation & Infrastructure House Appropriations House Energy and Commerce Senate Environment & Public Works Senate Appropriations % House delegation Democratic % Senate delegation Democratic Democratic share presidential vote ‘08 Small state advantage (1/state pop) Total # of House members Lambda Constant Lambda Policy Windows Pork Barrel Full Model Coeff. (Std Err) Coeff. (Std Err) Coeff. (Std Err) 0.791* 0.225* -0.042 0.001* 0.001*** 0.003 0.006*** 0.0268* 0.011 0.025*** 0.264*** 5.70** (0.363) (0.103) (0.085) (0.000) (0.000) (0.008) (0.002) (0.012) (0.005) (0.003) (0.026) (0.080) 0.016 0.035 0.069* 0.115* -0.050 0.00002 0.0001 0.010*** 0.582*** -0.036** 0.163*** 5.885*** (0.025) (0.026) (0.029) (0.032) (0.117) (0.0004) (0.0004) (0.001) (0.045) (0.013) (0.028) (0.057) 0.695* 0.196* -0.067 0.001** 0.002** 0.006 0.007** 0.039** -0.002 0.0186** (0.350) (0.100) (0.083) (0.001) (0.000) (0.008) (0.002) (0.014) (0.005) (0.003) 0.024 0.014 0.074** 0.098** -0.034 0.0001 0.001 0.004** 0.567** -0.048** (0.331) (0.025) (0.028) (0.031) (0.032) (0.0004) (0.0004) (0.001) (0.044) (0.025) 0.123** (0.028) 5 Constant 5.602** Unemployed % ‘07 Foreclosure rate, Q1 ‘09 Change in unemployed %, ’07-Q1 ‘09 Median income N R-squared 3,084 0.148 3,084 0.122 (0.133) 0.035** (0.001) -0.003 (0.004) -0.014* (0.007) -0.003 (0.002) 3,084 0.169 *p<0.05; **p<0.01; ***p<.001 Note: Dependent variable is total logged per capita ARRA allocations per county through the first quarter of 2010. N=3,084. State capital counties are excluded. Appendix B, Table 2: Predicting Expenditures on Infrastructure Projects, Policy Windows and Pork Barrel Models National parks Medium and small airports Interstate mileage U.S. highway mileage Water area (logged %) Lambda Constant Policy Window Pork Barrel Full Model Coef (Std Err) Coef (Std Err) Coef (Std Err) 0.772* 0.165 0.007*** 0.006*** 0.092*** 0.304*** 2.733*** 0.574 0.048 0.005** 0.006** 0.066** (0.304) (0.250) (0.002) (0.001) (0.025) 0.053 0.027 -0.046 0.467*** 0.005 0.012 0.000 -0.002 0.020*** 0.590*** (0.518) (0.084) (0.095) (0.107) (0.109) (0.047) (0.001) (0.001) (0.004) (0.154) Unemployed (%), 2007 Foreclosure rate (per 1000), Q1 ‘09 Change in unemployed %, ’07-Q1 ‘09 Median income -0.036 -0.001 0.03 0.005 (0.033) (0.939) (0.022) (0.005) Lambda Constant 0.268*** (0.026) 1.629*** (0.033) 3,084 0.111 Transportation and Infrastructure House Appropriations House Energy & Commerce Senate Environment & Public Works Senate Appropriations Total # of House members % House delegation Democratic % Senate delegation Democratic Democratic share presidential vote ‘08 Small state advantage (1/state pop) Lambda Constant N R-squared *p<0.05; **p<0.01; ***p<.001 (0.304) (0.249) (0.001) (0.001) (0.024) (0.026) (0.130) 0.014 -0.007 -0.075 0.533*** -0.125 0.132** -1.256 -0.002 0.023*** 0.671*** 0.265*** 2.118*** 3,084 0.095 (0.083) (0.083) (0.095) (0.107) (0.105) (0.043) (0.001) (0.001) (0.004) (0.150) (0.026) (0.026) 3,084 0.092 Note: Dependent variable is total logged per capita allocations for transportation and water projects per county through the first quarter of 2010. N=3,084. State capital counties are excluded. Bibliography Anselin, Luc. 1988. Spatial Econometrics: Methods and Models. Dordrecht, The Netherlands: Kluwer Academic Publishers. Anselin, Luc and Anil K. Bera. 1998. "Spatial Dependence in Linear Regression Models with an Introduction to Spatial Econometrics." Pp. 237-289 in Handbook of Applied Economic Statistics, edited by A. Ullah and D. E. Giles. New York: Marcel Dekker. Arnold, R. Douglas. 1979. Congress and the Bureaucracy: A Theory of Influence. New Haven: Yale University Press. Balla, Steven J., Eric D. Lawrence, Forrest Maltzman, and Lee Sigelman. 2002. "Partisanship, Blame Avoidance, and the Distribution of Legislative Pork." American Journal of Political Science 46:515-525. Baumgartner, Frank R. and Bryan D. Jones. 1993. Agendas and Instability in American Politics. Chicago: University of Chicago Press. Bertelli, Anthony M. and Christian R. Grose. 2009. "Secretaries of Pork? A New Theory of Distributive Public Policy." Journal of Politics 3:926-945. Bickers, Kenneth N. and Robert M. Stein. 2000. "The Congressional Pork Barrel in a Republican Era." Journal of Politics 62:1070-1086. Callen, Zachary. 2009. The Seams of the State: Infrastructure and Intergovernmental Relations in American State Building. Ph.D. Dissertation. Department of Political Science. University of Chicago, Chicago, Illinois. Chinni, Dante and James G. Gimpel. 2010. Our Patchwork Nation: The 12 Community Types that Make Up Our Nation. New York: Penguin Books. Cohen, Michael, James March, and Johan Olsen. 1972. "A Garbage Can Model of Organizational Choice." Administrative Science Quarterly 17:1-25. Evans, Diana. 1994. "Policy and Pork: The Use of Pork Barrel Projects to Build Policy Coalitions in the House of Representatives." American Journal of Political Science 38:894-917. Frisch, Scott A. 1998. The Politics of Pork: A Study of Congressional Appropriation Earmarks. New York: Garland. Garrett, Thomas A. and Russell S. Sobel. 2003. "The Political Economy of FEMA Disaster Payments." Economic Inquiry 41:496-509. Hamman, John A. and Jeffrey E. Cohen. 1997. "Reelection and Congressional Support." American Politics Quarterly 25:56-74. Hauk, William R. and Romain Wacziarg. 2007. "Small States, Big Pork." Quarterly Journal of Political Science 2:95-106. John, Peter. 2003. "Is There Life After Policy Streams, Advocacy Coalitions, and Punctuations: Using Evolutionary Theory to Explain Policy Change?" The Policy Studies Journal 31:481-498. Kingdon, John W. 1995. Agendas, Alternatives, and Public Policies. New York: HarperCollins. Krutz, Glen S. 2001. Hitching a Ride: Omnibus Legislating in the U.S. Congress. Columbus: Ohio State University Press. LaFleur, Jennifer 2009. "How We Compiled and Analyzed Stimulus Spending." ProPublica. Larcinese, Valentino, Leonzio Rizzo, and Cecilia Testa. 2006. "Allocating the Federal Budget to the States: The Impact of the President." Journal of Politics 68:447-456. 1 Lauderdale, Benjamin E. 2008. "Pass the Pork: Measuring Legislator Shares in Congress." Political Analysis 16:235-249. Lazarus, Jeffrey. 2009. "Party, Electoral Vulnerability, and Earmarks in the U.S. House of Representatives." Journal of Politics 71:1050-61. Lazarus, Jeffrey and Amy Steigerwalt. 2009. "Different Houses: The Distribution of Earmarks in the U.S. House and Senate." Legislative Studies Quarterly 34:347-373. Lee, Frances E. 1998. "Representation and Public Policy: The Consequences of Senate Apportionment for the Geographic Distribution of Federal Funds." The Journal of Politics 60:34-62. —. 2003. "Geographic Politics in the U.S. House of Representatives: Coalition Building and Distribution of Benefits." American Journal of Political Science 47:713-727. Lee, Frances E. and Bruce I. Oppenheimer. 1999. Sizing Up The Senate: The Unequal Consequences of Equal Representation. Chicago: University of Chicago Press. LeSage, James and R. Kelley Pace. 2009. Introduction to Spatial Econometrics. Boca Raton, FL: Chapman and Hall/CRC. Levitt, Steven D. and James M. Snyder, Jr. 1995. "Political Parties and the Distribution of Federal Outlays." American Journal of Political Science 39:958–80. Mucciaroni, Gary. 1992. "The Garbage Can Model & the Study of Policy Making." Polity 24:459-482. Peterson, Paul E. 1995. The Price of Federalism. New York: The Twentieth Century Fund, Inc. Reeves, Andrew. Forthcoming. "Political Disaster: Presidential Disaster Declarations and Electoral Politics." Journal of Politics. Rundquist, Barry S. and Thomas M. Carsey. 2002. Congress and Defense Spending: The Distributive Politics of Military Procurement. Norman: University of Oklahoma Press. Sabatier, Paul A. 1988. "An Advocacy Coalition Framework of Policy Change and the Role of Policy-Oriented Learning Therein." Policy Sciences 21:129-168. —. 1991. "Toward Better Theories of the Policy Process." PS: Political Science and Politics 24:147-156. Sinclair, Barbara. 2000. Unorthodox Lawmaking: New Legislative Processes in the U.S. Congress. Washington, DC: CQ Press. Stein, Robert M. and Kenneth N. Bickers. 1994. "Universalism and the Electoral Connection: A Test and Some Doubts." Political Research Quarterly 47:295–317. Weise, Karen 2010. "Stimulus Spending Likely to Make Administration's Goal." ProPublica, August 3.
© Copyright 2026 Paperzz