key: cord-0727169-v5opykmv
authors: Clemens, Jeffrey; Veuger, Stan
title: Politics and the distribution of federal funds: Evidence from federal legislation in response to COVID-19()
date: 2021-11-13
journal: J Public Econ
DOI: 10.1016/j.jpubeco.2021.104554
sha: 170a576e8a9ad9013d664b2f08ed80041430b575
doc_id: 727169
cord_uid: v5opykmv

COVID-19 relief legislation offers a unique setting to study how political representation shapes the distribution of federal assistance to state and local governments. We provide evidence of a substantial small-state bias: an additional Senator or Representative per million residents predicts an additional 670 dollars in aid per capita across the four relief packages. Alignment with the Democratic party predicts increases in states’ allocations through legislation designed after the January 2021 political transition. This benefit of alignment with a unified federal government operates through the American Rescue Plan Act’s size and through the formulas it used to distribute transportation and general relief funds.

Fiscal transfers from the federal government to state and local governments play an important role In the US federal system. This is particularly true during downturns, when balanced budget requirements can pose sharply binding constraints on lower levels of government. During the COVID-19 pandemic, federal fiscal assistance reached unprecedented levels, with aid to state and local governments spanning four legislative vehicles and summing to almost $1 trillion. 1 The effectiveness of this fiscal assistance depends in part on how it is targeted. It is thus of interest to know the extent to which the distribution of this aid is shaped by political, as opposed to purely economic, considerations.

During the pandemic, federal fiscal assistance has been distributed through a variety of channels, including general aid to states, general aid to local governments, and aid appropriated for specific functions of state and local government. Further, direct aid has been shaped by formulas that are designed at the discretion of the US Congress. 2 We analyze the resulting distributions to gain insight into the channels through which political representation influences the distribution of federal funds.

We find evidence of pervasive small-state bias across each of the legislative vehicles. Adding across the four main pieces of legislation, we estimate that having an additional Senator or Representative per million residents predicts an additional $670 dollars in combined state and local aid per capita. While this analysis relies on cross-sectional variations in representation across states, the evidence is quite strongly suggestive of a causal role for small states' disproportionate representation. The small-state https://doi.org/10.1016/j.jpubeco.2021.104554 0047-2727/Ó 2021 Elsevier B.V. All rights reserved. q We are grateful to Philip Hoxie for excellent research assistance, and to Michael Farquharson of the Committee for a Responsible Federal Budget for assistance in the data collection process. We thank James Alt, Nora Gordon, Richard Grossman, as well as reading group and seminar attendees at the Harvard Department of Government and the UC-San Diego Department of Economics for their thoughtful comments on an earlier draft of the paper. Declarations of interest: none. bias we estimate is orthogonal to states' partisan alignment, as well as to variations in the pandemic's impact on tax revenues, the labor market, and overall economic output. It is also orthogonal to the baseline size of the state and local public sector, as well as to proxies for the costs associated with land management. By way of comparison, an advantage of $670 per capita is 12.5 times the magnitude of the annual benefit of a district representative's alignment with the party of the President in normal times, as estimated by Berry et al. (2010) .

Next, we examine the relevance of alignment with the party in power by analyzing differences in the distribution of funds associated with the March 2021 American Rescue Plan Act (ARPA) relative to legislation enacted by the previous congress. We show that arrival of unified government predicts a non-trivial increase in aid allocated to states whose delegations lean Democratic rather than Republican. A fully Democratic delegation predicts a $300 per capita increase in federal funds under unified Democratic control of the federal government relative to the previous year's divided government.

A first set of contributions of our analysis is thus to add to the extensive body of work on the consequences of partisan control. Papers in this literature have investigated some of the factors we Note: This figure shows funds per capita across the four COVID-19 bills for states by type. Total education, relief, and transit funds are shown for the CARES Act, Families First Coronavirus Response Act, Response and Relief Act, and American Rescue Plan Act on a per capita basis. Panel A groups states into terciles by the number of senators and representatives per million residents, with the 1st tercile containing the largest states and the 3rd tercile containing the smallest states. Panel B groups states into terciles by the share of their congressional delegation that are Democrats, with the 1st tercile containing less Democratic states and the 3rd tercile containing more Democratic states. This figure uses data from the Committee for a Responsible Federal Budget (2021), US Federal Transit Administration (2021a, 2021b), Lewis (2021) , US Census Bureau (2020), Chidambaram and Musumeci (2021) , Medicaid and Chip Payment Access Commission (2021) , and US Office of Elementary and Secondary Education (2021). consider here, both at the federal level and at the state and local levels. The latter category includes work by Alt and Lowry (1994) , Poterba (1994) , Ansolabehere et al. (2002) , Reed (2006) , Ferreira and Gyourko (2009) , Bertocchi et al. (2020) , and Dynes and Holbein (2020) .

At the federal level, there is prior evidence on the alignment of a district or state's delegation with the President (Berry et al., 2010) and with the majority party (Albouy, 2013) . 3 There has also been significant work on presidential efforts to maximize electoral votes (Wright, 1974) and on efforts to use federal spending to aid weak incumbents (Bickers and Stein, 1996) . A substantial body of additional research has analyzed the relationship between various dimensions of political influence and the distribution of federal spending (Berry and Fowler, 2016; Knight, 2002) , economic outcomes (Hodler and Raschky, 2014; Levitt and Poterba, 1999) , and political or policy outcomes (Besley and Case, 2003; Levitt and Snyder Jr, 1997) .

The findings from our analyses of the role of alignment with unified partisan control are consistent with these earlier analyses. Our focus on unified partisan control, as opposed to some of the alternative power dynamics considered in prior research, is motivated by the specific change in political power that occurred during the COVID-19 pandemic.

Second, the findings from our analysis of disproportionate representation are also consistent with several prior analyses (Atlas et al., 1995; Lee, 1998; Hauk and Wacziarg, 2007; Knight, 2008; Rodden, 2002) . We contribute to this literature by showing that over-represented states were systematically advantaged across four high-stakes pieces of relief legislation. Prior work has focused on whether small states are advantaged by spending associated with the normal course of Congressional business, in some cases emphasizing broad categories of spending (Atlas et al., 1995; Lee, 1998) and in other cases focusing on ''earmarks" or ''pork-barrel" spending (Hauk and Wacziarg, 2007; Knight, 2008) .

Furthermore, through additional pieces of analysis, we provide several novel advances to the literature on the mechanisms through which shifts in political power translate into shifts in the allocation of funds. A unique aspect of our setting is that we are able to analyze four major pieces of legislation that serve the same general purpose: to shore up the fiscal capacity of state and local governments as they responded to the COVID-19 pandemic. This allows our analysis to focus on mechanisms other than variation in legislative priorities, which have been found to be important elsewhere. For example, Albouy (2013) finds that Republicans have a preference for defense and transportation contracts while Democrats have a preference for education dollars. Levitt and Snyder (1995) find large differences in fund disbursements at the end of a period of Democratic control and highlight that these differences can be driven by tweaked formulas, altered legislative priorities, or a combination of both.

We shed additional light on the mechanisms through which funds can be targeted by analyzing multiple legislative vehicles that target the same overall priority. We show that choice of allocative formulas plays a major role. Relative to the CARES Act's population-driven formula for allocating aid, for example, the ARPA's unemployment-driven formula shifts dollars towards states with either large pandemic-driven increases in unemployment or with high baseline rates of structural unemployment. 4 The CARES Act contains funds allocated under the Coronavirus State and Local Relief Fund that go to both state and local governments, funds allocated under the Elementary and Secondary Relief Fund (ESRF) that go to local governments, funds allocated under the Governor's Emergency Education Relief Fund (GEERF) and Innovation Grants that go to states, election funds allocated to states, and transit funds allocated under sections 5307 and 5311 formulas that go to localities. FFCRA contains Medicaid matching funds that go to states. RRA contains funds allocated under the ESRF that go to local governments, funds allocated under the GEERF that go to states, transit funds allocated under sections 5307, 5310, and 5311 formulas to localities, and section 133 formula transportation funds to states. ARPA contains funds allocated under the Coronavirus State and Local Relief Fund that go to both state and local governments, funds allocated under the ESRF that go to local governments, funds allocated under the GEERF that go to states, transit funds allocated to localities using section 5307, 5310, and 5311 formulas as well as capital investment grants, and Medicaid matching increases for uncompensated care (section 9819) and community-based services (section 9817) that go to states. 5 We count independent members of congress as Democrats if they caucus with the Democrats. We count the Arizona Senate delegation as 50 percent Democratic in the 116th Congress (even though it was 100% Democratic and 100% Republican for a month or so each). We categorize Georgia's two Senate seats as Democratic in the 117th Congress. We classify CA-25 and NJ-2 as Republican seats in the 116th Congress. These states lean disproportionately Democratic. In addition, while we find that unified Democratic control predicts a substantial shift in transportation funds towards states with heavily Democratic delegations, we find no such shift in education funds, where aid formulas are linked to pupil counts. We also find the sheer magnitude of the ARPA's fiscal assistance package to be an important mechanism. Gauging the required size of the ARPA's fiscal assistance package requires drawing on analyses of the pandemic's effects on state and local government finances. The available analyses implied a need for an additional $100 to $200 billion at most (see e.g. Auerbach et al., 2020; Whitaker, 2020b) . Indeed, forecasts for states' summer 2021 budgeting processes revealed substantial surpluses (Albright et al., 2021; Miller, 2021) . While the ARPA's funding formulas are only modestly more generous to Democratic-leaning states than were the formulas from the first three relief packages, the choice to allocate $500 billion rather than $100 to $200 billion accounts for much of the Democratic states' $300 per capita advantage relative to Republican states.

Fourth, we contribute to the literature on state and local government budgets over the course of the pandemic. Initial papers in this literature sought to forecast the magnitudes of the revenue shortfalls faced by various levels of government within the United States (Auerbach et al., 2020; Veuger, 2020a, 2020b; Chernick et al., 2020; Whitaker, 2020a; 2020b) . Additional analyses have considered the pandemic's implications for spending needs (Gordon and Reber, 2020; Clemens et al., forthcoming) . Analyses have also explored the effects of initial state and local aid allocations on government employment (Green and Loualiche, 2020) . We offer the first detailed description of the determinants of federal allocations to state and local governments across all four major pieces of COVID-response legislation, which were of unprecedented scope.

The paper proceeds as follows. Section 2 presents the data. Section 3 presents our empirical framework and analysis of smallstate bias. Section 4 does the same for the consequences of unified Democratic control after the 2020 elections. Section 5 concludes.

Our analysis is centered on four major pieces of legislation during the COVID-19 pandemic, each of which directed federal relief to state and local governments. These are the CARES Act, the Families First Coronavirus Response Act (FFCRA), the Response and Relief Act (RRA), and the American Rescue Plan Act (ARPA). Readers interested in legislative histories can find a summary of key dates in Appendix Fig. A1 . For our purposes, the most crucial detail is that the CARES Act, the FFCRA, and the RRA were passed by the 116th Congress and signed by President Trump, while the ARPA was passed by the 117th Congress and signed by President Biden. Taken together, these packages constituted a massive relief effort that provided as much as $6 trillion in income support to households, a mix of loans, grants, and tax relief to firms and non-profits, additional funding for (public) health efforts, and intragovernmental grants to subnational governments. This final category includes around $900 billion in funds for state, local, territorial, and tribal governments, as well as the District of Columbia. We focus on the first two types of subnational governments here, namely those with full congressional representation. 6 Might alternative weightings of Senators and Representatives yield different results? We show in Appendix Table A2 that, when estimating equation (2), the tstatistics we obtain are quite similar when using variables that allow for alternative weightings. The predictive power of our measure of over-representation is not sensitive to our choice of functional form. We use data from the Committee for a Responsible Federal Budget (2021) to summarize the funds each bill appropriated to state and local governments. We complete these data with information from several sources. We obtain information on the distribution of transit funds for the RRA and ARPA from the US Federal Transit Administration (2021a, 2021b). Data on the allocation of ARPA assistance to non-public schools come from the US Office of Elementary and Secondary Education (2021). We obtain estimates of ARPA section 9817 matching increases from Chidambaram and Musumeci (2021) . We approximate the allocation of ARPA section 9819 federal matching funds for uncompensated care using FY2021 estimates of federal disproportionate share hospital allotments by state from the Medicaid and Chip Payment Access Commission (2021). We then present these data as funds directed to state governments, funds directed to state and local educational agencies, and funds directed to other local governments. 4 The funds can also be divided across three functional categories, namely transportation funds, education funds, and general fiscal assistance (defined here to include all other fiscal assistance). Fig. 1 provides an initial look at the distribution of funds across the four pieces of legislation. Dollar values are expressed on a per capita basis and are divided into general relief funds, transportation funds, and education funds. Panel A provides an initial look at bias in favor of small states, which benefit from overrepresentation in the US Congress. The small-state bias emerges primarily through general relief funds, which were distributed through formulas featuring floor functions. Panel B provides an initial sense of partisan advantage. It is apparent that transportation dollars skew towards Democratic leaning states, that education dollars exhibit very little partisan skew, and that general relief dollars exhibited a strong partisan skew under the ARPA, but not in the earlier pieces of legislation. The maps in Fig. A2 reinforce both of these initial impressions.

Our analysis focuses on two types of dependent variables. The first type expresses each bill's funding on a per capita basis. The second type focuses on how each states' share of each bill's funding compares with its share of the national population. We construct this proportional share of funds for each state in each bill as follows: (2021) to estimate equations of the following form:

Where Outcome c ib is funding in category c in state i and bill b. Funds per capita for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies are the dependent variables in Columns 1 to 4, respectively. In Columns 5 to 8, Outcome c ib is Proportional Share Of Bill ib , which is the ratio of state i's share of funding in category c to state i's share of the US population in bill b. Columns 5, 6, 7, and 8 show this proportional share of funds for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies, respectively. Dem:Deleg:Share ib is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i when bill b was passed. Unified b is a dummy that takes a value of 1 when the Democratic Party assumes unified control of the US House, Senate, and Presidency in 2021. We interact this dummy variable with Dem:Deleg:Share ib . k b and k i represent state and bill fixed effects, respectively. e ib is an error term. Observations are weighted by state population and standard errors are clustered by state. Average funds per capita across all four bills are $615, $399, $287 for total, state, and local funds, respectively. All proportional shares average to 1. *** p < 0.01, ** p < 0.05, * p < 0.1 

Where Outcome c ib is funding in category c in state i and bill b. Funds per capita for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies are the dependent variables in Columns 1 to 3, respectively. In Columns 4 to 6, Outcome c ib is Proportional Share Of Bill c ib , which is the ratio of state i's share of funding in category c to state i's share of the US population in bill b. Columns 4, 5, and 6 show this proportional share of funds for relief, funds to transit, and funds to education, respectively. Dem:Deleg:Share ib is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i when bill b was passed. Unified b is a dummy that takes a value of 1 when the Democratic Party assumes unified control of the US House, Senate, and Presidency in 2021. We interact this dummy variable with Dem:Deleg:Share ib . k b and k i represent state and bill fixed effects, respectively. e ib is an error term. Observations are weighted by state population and standard errors are clustered by state. Average funds per capita across all four bills are $576, $76, and $197 for relief, transit, and education funds, respectively. All proportional shares average to 1. *** p < 0.01, ** p < 0.05, * p < 0.1 In Eq. (1), Funds ib is the amount of money allocated to state i in bill b, and Pop i is the 2020 population in state i. When Proportional Share Of Bill ib is greater than 1, a state has received a disproportionately large share of the funds in bill b.

Our main independent variables relate to the distribution of power at the federal level. We use data from Lewis (2021) to construct the share of House and Senate seats held by each political party in the 116th Congress and the 117th Congress. We then average the Democratic Party's share of House and Senate seats in each state to construct the Democrats' congressional share. 5 Values for the 116th Congress map to the CARES Act, FFCRA, and RRA, and values for the 117th Congress map to ARPA. We use a second political variable that interacts the Democratic party share with a dummy that takes a value of 1 in the 117th Congress, signifying the switch to unified Democratic Party control of the House, Senate, and Presidency. Our third political variable measures small-state bias using the total number of US Senate and House seats per 1,000,000 state residents. Smaller states generally have more representatives per 1,000,000 people, which reflects their disproportionate representation in the Senate and, to a lesser degree, in the House.

A second set of independent variables describes the economic shocks and fiscal projections that contributed to estimates of states' fiscal needs as driven by the pandemic. To proxy for state-specific revenue shocks, we add Whitaker's (2020b) estimates of the realized decline in state and local government revenues in fiscal year 2020 to the projected revenue loss in fiscal year 2021. For the latter, we use estimates from Whitaker's slow recovery scenario. Note that Whitaker's combined estimate of state and local government revenue losses, spanning the 2020 and 2021 fiscal years, is $312 billion, which is far less than the $900 billion ultimately allocated by the federal government. To measure the unemployment shock from the pandemic, we purposefully adopt the ARPA's formula for distributing general relief, which is a function of the average number of unemployed persons per capita during the fourth quarter of 2020 (US Bureau of Labor Statistics, 2021). Finally, we proxy at a broad level for declines in economic activity using the percent change in total personal income between the fourth quarter of 2019 and the fourth quarter of 2020.

Our final pair of independent variables describe the outlays of state and local governments. We use the 2018 Survey of State and Local Government Finances (US Census Bureau, 2019a) to sum together direct expenditures of state and local governments. We also use the total acres of federal land by state, as reported in Vincent et al. (2020) , to proxy for needs associated with federal lands under direct control of the federal government. We report both variables on a per capita basis. Table 1 presents summary statistics on the full set of variables that are utilized in our analysis. Note, as can be seen in Fig. 1 , that not all forms of relief appeared in all four pieces of legislation. Consequently, some of the fiscal variables we analyze are associated with 150 observations, while others are associated with 200 observations. Additional details on the definitions of key variables can be found in Appendix Table A1 .

In this section we commence our empirical analysis by separately analyzing each of the CARES Act, the FFCRA, the RRA, and the ARPA. Table 2 presents descriptive evidence on the predictors of the distribution of federal funds, across states, for each of these four legislative vehicles. Specifically, it presents estimates of the following equation:

In Eq.

(2), Bill Funds b i is the total per capita funding to state and local governments in state i from bill b. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. Representatives Per Million i is the total number of US representatives and US senators divided by the population of state i. 6 This variable varies primarily with the US Congress's relative overrepresentation of small states, which is particularly strong in the Senate. S i is a vector of additional state-level covariates. These include the predicted per capita tax shortfall for state and local governments as estimated by Whitaker (2020a Whitaker ( , 2020b , the average number of unemployed persons per capita during the fourth quarter of 2020, the percentage change in personal income between the fourth quarter of 2019 and the fourth quarter of 2020, the per capita total direct expenditures of state and local governments, and the acres of federal lands per capita. e i is an error term. Standard errors are robust to heteroskedasticity. For our primary estimates, observations are weighted by state population. Appendix Table A4 reveals that we obtain similar results when we weight observations equally.

To be clear, estimates of Eq.

(2) provide descriptive evidence on the correlates of federal funding. Interpretating c as a causal effect of small states' over-representation would require that Representatives Per Million i be exogenous and hence uncorrelated with the error term, e i . This assumption could be violated, for example, if state population is systematically correlated with relevant political factors or with other factors that, in the eyes of Congress, merited additional federal funds. While the exogeneity assumption cannot be proven, what we are able to show is that the relationship we observe is orthogonal to a rich set of covariates we can add to S i . Columns 1 and 2 of Table 2 report results for the CARES Act, Columns 3 and 4 report results for the FFCRA, Columns 5 and 6 report results for the RRA, and Columns 7 and 8 report results for the ARPA. The most striking finding in Table 2 is the evidence that over-represented states have benefited disproportionately from federal dollars. Over-represented states have received more federal dollars per resident than have under-represented states in all four of the COVID-19 fiscal relief packages, though the estimates are not statistically distinguishable from zero for the RRA.

How large is the small-state bias we estimate? An additional representative per million residents predicts an additional $670 to $780 per state resident across the four pieces of legislation. This is the advantage, for example, of Montana's roughly 1 million residents, who enjoy representation from 2 Senators and 1 representative, relative to Arkansas's roughly 3 million residents, who enjoy only 2 representatives per million on account of their 2 Senators and 4 representatives. Note that $670 per capita is quite large in comparison with related effects as estimated in the literature on distributive politics. For example, it is 12.5 times the magnitude of the annual benefit of a district's alignment with the party of the President, as estimated by Berry et al. (2010) . 7 The estimates are suggestive that 5 We count independent members of congress as Democrats if they caucus with the Democrats. We count the Arizona Senate delegation as 50 percent Democratic in the 116th Congress (even though it was 100% Democratic and 100% Republican for a month or so each). We categorize Georgia's two Senate seats as Democratic in the 117th Congress. We classify CA-25 and NJ-2 as Republican seats in the 116th Congress. 6 Might alternative weightings of Senators and Representatives yield different results? We show in Appendix Table A2 that, when estimating equation (2), the tstatistics we obtain are quite similar when using variables that allow for alternative weightings. The predictive power of our measure of over-representation is not sensitive to our choice of functional form. 7 To what extent can we provide evidence on the potential biases discussed above? A comparison of the sparse specifications in Table 2 with the specifications that include additional covariates reveals that the correlation between over-representation and federal dollars is orthogonal to states' revenue shocks, economic shocks, the size of their public sector, acreage of federal land, and population density. While we cannot fully rule such factors out, analyses in our appendix materials provide additional evidence that variations in states' needs are not a leading factor behind the small-state biases we estimate. 9 The sparse specifications in Table 2 reveal that federal fiscal assistance has tended to be positively correlated with the Democratic Party's share of each state's congressional delegation. Column 7 shows that this correlation was particularly strong in the ARPA. This set of correlations, however, is sensitive to whether we include the more extensive set of covariates in the regression. Among these other covariates, the strongest and most consistent correlate of federal dollars is the unemployment variable, which has particularly strong predictive power for dollars allocated through the ARPA. These correlations are consistent with a targeting of states which had high unemployment rates, which may have been amplified by the newly increased power of those states' representatives. It is to the role of the shift in federal power that we turn in the following section.

Our analysis of the relationship between partisan political representation and the distribution of fiscal assistance across states takes advantage of the sharp change in partisan control that occurred following the November 2020 election and subsequent Senate runoff elections. In these elections, the Democratic Party secured control of the Senate and the White House, in addition to the House of Representatives. This change enables us to estimate panel specifications that control for all time-invariant factors that may differ across states. The estimates presented in Table 3 and 4 are of equations of the following form:

In Eq. (3), Outcome c ib represents fiscal outcomes in funding category c in state i and bill b. We analyze two outcomes of interest, namely federal dollars per capita and the Proportional Share of Bill variable described by Eq. (1).

The primary coefficient of interest in Eq. (3) is c, which describes the degree to which alignment with the unified Democratic majority predicts increases in federal funding. That is, estimates of c will reflect the degree to which ARPA's funding priorities and/or funding formulas allocated dollars towards states with Democratic delega-tions beyond what one would have predicted based on the legislative vehicles that were passed under divided government. Note that our interest is not in whether one set of allocations aligns better with either objective assessments of need or subjective assessments of socially desirable priorities. Instead, our interest is in understanding the extent to which alignment with unified government predicts changes in states' allocations. Table 3 sorts funds according to the type of state or local government that received the funds. The categories of funds in Table 3 are the total funds for state and local governments (Columns 1 and 5), state governments (Columns 2 and 6), counties and municipalities 10 (Columns 3 and 7) , and educational agencies (Columns 4 and 8) . The dependent variables in Columns 1 through 4 are expressed in dollars per state resident. The dependent variables in Columns 5 through 8 are expressed as the ratio of the state's share of funds in category c relative to its share of the country's total population. Table 4 follows a similar structure, but with funds allocated according to their functional categories: general relief funds (Columns 1 and 4) , transit funds (Columns 2 and 5), and education funds (Columns 3 and 6) .

The results in Table 3 and 4 provide evidence on a nuanced set of channels through which political representation can shape the distribution of federal funds. First, the estimate in Column 1 of Table 3 indicates that the ARPA directed more funds towards states with heavily Democratic delegations compared to the CARES Act, the FFRCA, and the RRA. Relative to states with entirely Republican delegations, residents of states with entirely Democratic delegations are predicted to receive, on average, over $300 per capita more through ARPA than we would have predicted had control over Congress and the executive branch not been unified in Democratic hands. 11 Through what mechanisms did the ARPA shape the distribution of federal funds relative to earlier pandemic relief legislation? The additional funds could be driven by the ARPA's sheer magnitude, by the formulas through which those funds were distributed, or by its allocation across functional categories and levels of government. We provide evidence on these mechanisms through several pieces of analysis.

In Columns 2 through 4 of Table 3 , we divide federal fiscal assistance into funds directed to state governments, funds directed to local governments within each state, and funds directed to educational agencies for distribution to school districts, charter schools, and other recipients. We find that the state and non-educational local components of federal aid drive the overall ARPA advantage for states with more heavily Democratic delegations. Funding for educational agencies, in contrast, do not differ significantly between states with different rates of Democratic representation.

In Columns 5 through 8 of Table 3 we analyze the extent to which states received a disproportionate share of each bill's fiscal relief. That is, we estimate Eq. (3) using the dependent variable described by Eq. (1). This variable is increasing in the extent to which a state's share of federal funds exceeds its share of the national population. Column 5 of Table 3 reveals that, on a proportional basis, the distribution of ARPA funds was only modestly more tilted towards states with Democratic party delegations than were previous bills. In the aggregate, this indicates that the significant increases in relief funds associated with unified party control were driven primarily by the magnitude of ARPA relief relative to relief in the earlier bills. Although the aggregate masks substantial shifts in the distribution of funds within key functional categories, Democratic states did not receive dramatically larger shares of ARPA funds than they received from the earlier bills. 8 To arrive at this number, we first calculate the absolute deviation of each state's averaged number of members of Congress per million residents from the weighted mean across states. We multiply this number by the coefficient in Column 7 of Appendix Table A5 to arrive at the impact per resident per bill. We then aggregate across bills and states to estimate a grand total of $29.9 billion.

9 Table A3 provides evidence that RepresentationPerCap i and Dem:Deleg:Share are roughly independent of one another. Table A4 reveals that the small-state bias is not particularly sensitive to whether we weight observations according to states' populations. For Table A5 , we stack observations for the four pieces of legislation and demonstrate that our results are insensitive to adding each of the covariates from our more heavily controlled specifications one at a time. Fig. A3 shows further that the relationship between representation per capita and fiscal assistance per capita is quite strongly linear (Panel A), while the relationship between population and fiscal assistance per capita is decidedly non-linear (Panel B). Tables A6 and A7 use a logged representation variable to provide evidence that the estimates in Tables 2 and A5 are not driven by the representation variable's skewness. 10 This category also includes other local governments that are not recipients of funds from state educational agencies, such as utility districts. 11 Appendix Table A8 shows that we estimate a moderately smaller differential of just under $200 when we weight observations equally rather than weighting according to state population.

Because Democratic states are modestly advantaged by the totality of state and local aid, a larger bill will mean more dollars per resident of Democratic states than of Republican states. Interpreting the magnitude of ARPA relief requires drawing on estimates of state and local government needs. By the time the ARPA had been drafted, essentially all analysts had arrived at the conclusion that only modest additional fiscal relief was needed (see e.g. Auerbach et al., 2020; Whitaker, 2020b; DeGroot et al., 2021; Lincicome, 2021; Walczak, 2021) . Estimates of remaining need would have implied a relief package similar in magnitude to the state and local relief found in the CARES, RRA, and FFCRA. The ARPA's $500 billion in aid thus exceeded even the largest estimates of remaining need. These analyses have been borne out by subsequent forecasts for states' budgets. The ARPA's magnitude should, in this sense, be interpreted as a political choice, and one that previous analyses in the literature on distributive politics would have struggled to detect. This is because analyses in the literature do not typically have directly applicable measures of the amount of need associated with specific legislative priorities.

Columns 6 through 8 of Table 3 reveal that the ARPA's overall distribution masks substantial shifts in the distribution of funds for educational agencies versus other state and local government entities. The ARPA's state government relief and, to a lesser extent, the funds distributed to counties and municipalities, shifted substantially towards Democratic states relative to funds from the earlier bills. In contrast, the ARPA's relief for educational agencies shifted slightly towards Republican states relative to funds from earlier bills. Table 4 presents results in which we divide federal relief funds according to their functional purpose rather than according to the government entity that received them. 12 Results in Columns 1 and 4 describe the distribution of general relief funds, while Columns 2 and 5 describe transportation funds and Columns 3 and 6 describe education funds. Columns 1 through 3 analyze spending expressed in dollars per state resident while Columns 4 through 6 are expressed in terms of each state's share of funds relative to its share of the country's population.

Columns 1 and 2 reveal that the shift in the distribution of dollars was driven primarily by general relief funds. Columns 4 and 5 convey, in contrast, that the proportional distribution of transit dollars shifted much more heavily towards Democratic states under the ARPA than did general relief funds. These results are tied together by the fact that general relief funds account for a much larger share of overall fiscal assistance than do transportation funds, as shown earlier in Fig. 1 . In contrast with general relief and transportation dollars, we find no evidence of a shift in the partisan skew of education dollars. Indeed, as shown earlier in Fig. 1 , education dollars were distributed evenly across the states in each of the legislative vehicles in which they appeared.

What legislative mechanisms drive these shifts in the distribution of funds? An inspection of the legislation reveals that shifts in fiscal relief were largely driven by formula design. With respect to general relief funds, the ARPA's unemployment-driven formula steered dollars to states with high levels of unemployment, which reflect a mix of pre-pandemic and pandemic-driven factors. Among the nation's most populous states, for example, Democraticleaning New York and California have had unemployment rates well above the national average, while Republican leaning Texas has been quite close to the national average and Florida has been well beneath it. The allocation of transportation funds is more complicated, as it reflects a combination of formula-driven and discretionary allocations. The Section 5307 Urbanized Area Apportionment formula, for example, was used to allocate a large share of CARES Act, RRA, and ARPA transportation dollars. This formula is driven to a significant degree by estimates of bus mileage and skews quite strongly towards states with Democratic party delegations. The lack of a partisan distribution in education dollars similarly reflects the underlying allocation formulas. These formulas, principally those of Title I, Part A of the Elementary and Secondary Education Act, place a heavy weight on headcounts of eligible children, who are typically children from low-income households.

An important question in both the political economy and political science literatures is how changes in the distribution of federal funds are achieved. That is, what are the legislative mechanisms through which changes in the distribution of funds emerge? As past research has pointed out, key mechanisms can include agenda-setting power, standard legislative logrolling, and a governing party's ability to advance agendas with policy priorities that target their constituents' needs and pocketbooks. Our context is of interest in part because it provides an opportunity to assess distributive politics in the context of multiple, salient pieces of legislation with substantial funds dedicated to the same broad end, support for state and local governments.

Our analysis reveals an important role not just for the distribution of funds in specific legislative vehicles, but also for their size. That is, we find that the change in federal control shifted the distribution of ARPA dollars only slightly toward states aligned with the newly unified federal government. The ARPA's size relative to the 2020 packages is what induced larger absolute transfers to politically aligned states than did the previous packages (Tables 3 and 4 ). This is reminiscent of the mechanism through which the U.S. fiscal system can be at the same time both more progressive and less redistributive than the systems of other OECD countries: simply by being smaller (Slavov and Viard, 2016) . Analyses of how the budgetary pie is distributed will typically fail to detect this particular mechanism.

Our other main result is more in line with previous work in this area: representation, and in particular the overrepresentation of small states, matters quite meaningfully for the distribution of federal funds across states and localities. We find this small-state bias to be of substantial economic significance: having an additional Senator or Representative per million residents predicts an additional $670 dollars in aid per capita across the four relief packages combined. This is equivalent to 2% of U.S. income per capita (U.S. Census Bureau, 2019b). Across all four bills, we estimate that the small-state bias altered the allocation of around $30 billion in relief funds, which is equivalent to the funding this same legislation allocated to Pfizer, Moderna, the GAVI Vaccine Alliance, Regeron, Johnson & Johnson, AstraZeneca, GlaxoSmithKline, Eli Lily and Company, Merck, and Novavax for the development, manufacturing, and distribution of COVID-19 vaccines and therapeutics.

The impacts of representation on the distribution of funds can be viewed as deviations from the core purpose of fiscal relief, which is to stabilize state and local budgets in the face of macroeconomic shocks. In the U.S. context, automatic stabilizers for state and local government budgets flow primarily through the Medicaid and Unemployment Insurance programs. Such mechanisms have the benefit of reducing the need for ad hoc, and potentially politicized, policy making when negative shocks occur. At the same time, the design of automatic stabilizers involves choices that are less straightforward than their proponents sometimes imply, and that must also be designed through the political process. Whether automatic or ad hoc stabilizers better target the jurisdictions with greatest needs depends on the mix of economic and political factors that shape their design, as well as the nature of the shocks face. 12 Appendix Table A9 shows similar results when we weight observations equally rather than weighting according to state population. Panel B shows the relationship between state population and federal funds per state resident. Population is presented on a log scale. Federal funding is the sum of state and local fiscal assistance across the CARES Act, Families First Act, Recovery and Relief Act, and American Rescue Plan Act. Small states are defined as in Fig. 1 of Green and Loualiche (2020) , whose focus was on the relationship between population and allocations of general fiscal assistance through the CARES Act. These small states received the CARES Act's minimum allocation of $1.25 billion. (2020) Change in Personal Income Q42019 to Q42020

The percent change in real personal income between Q4 2019 and Q4 2020 in each state. Personal income is deflated by the personal consumption expenditures chained price index (PECEPI). (2021) Total State and Local Spending per Capita

The total direct expenditures of state and local governments in each state divided by the population in that state.

US Census Bureau (2019a, 2019b)

The total acreage of federal lands in each state divided by the population of that state.

Population Density The total state population divided by total state area in acres. Vincent et al. (2020) ; US Census Bureau (2020) 

Where Bill Funds ib is the total funding to state and local government in state i and bill b per capita. Adjusted Representatives Per Million i is the total number of US representatives and US senators divided by the population estimate for 2020 for state i in millions of people in Column 1. In Columns 2 through 6, the weight placed on US senators is increased, with Column 6 weighting the number of US senators 27 times more than the number of US representatives. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. k b represents bill fixed effects and e i is an error term. The ratio of senators to representatives used in each column is displayed as a ratio in the bottom row. Observations are weighted by state population and standard errors are clustered by state. Tstatistics are presented in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1 

Where Bill Funds b i is the total funding to state and local government in state i and bill b. Representatives Per Million: i is the total number of US representatives and US senators divided by the population estimate for 2020 for state I in millions of people. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. S i is a vector of state-level controls. These include the predicted tax shortfall for state and local governments from Whitaker (2020b) divided by the state population, the average number of unemployed persons each month in the fourth quarter of 2020 per capita, the percentage change in personal income between the fourth quarter of 2019 and the fourth quarter of 2020, the total direct expenditures from state and local governments per capita, the acres of federal lands per capita, and the log of population density for state i. e i is an error term. Standard errors are clustered by state. Columns 1 and 2 report results for the CARES Act, Columns 3 and 4 report results for the Families First Coronavirus Response Act, Columns 5 and 6 report results for the Recovery and Relief Act, and Columns 7 and 8 report results for the American Rescue Plan Act. *** p < 0.01, ** p < 0.05, * p < 0.1 

Where Bill Funds b i is the total funding to state and local government in state i and bill b per capita. Representatives Per Million: i is the total number of US representatives and US senators divided by the population estimate for 2020 for state i in millions of people. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. S i is a vector of state-level controls, which includes the predicted tax shortfall for state and local governments divided by the state population, the average number of unemployed persons each month in the fourth quarter of 2020 per capita, the percentage change in personal income between the fourth quarter of 2019 and the fourth quarter of 2020, the total direct expenditures from state and local governments per capita, the acres of federal lands per capita, and the log of population density for state i. k b represents bill fixed effects and e i is an error term. Observations are weighted by state population and standard errors are clustered by state. *** p < 0.01, ** p < 0.05, * p < 0.1 

Where Bill Funds b i is the total funding to state and local government in state i and bill b. Representatives Per Million: i is the total number of US representatives and US senators divided by the population estimate for 2020 for state i in millions of people. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. S i is a vector of state-level controls. These include the predicted tax shortfall for state and local governments from Whitaker (2020b) divided by the state population, the average number of unemployed persons each month in the fourth quarter of 2020 per capita, the percentage change in personal income between the fourth quarter of 2019 and the fourth quarter of 2020, the total direct expenditures from state and local governments per capita, the acres of federal lands per capita, and the log of population density for state i. k b represents bill fixed effects and e i is an error term.Observations are weighted by state population and standard errors are clustered by state. Columns 1 and 2 report results for the CARES Act, Columns 3 and 4 report results for the Families First Act, Columns 5 and 6 report results for the Recovery and Relief Act, and Columns 7 and 8 report results for the American Rescue Plan Act. *** p < 0.01, ** p < 0.05, * p < 0.1 

Where Bill Funds b i is the total funding to state and local government in state i and bill b per capita. Representatives Per Million: i is the total number of US representatives and US senators divided by the population estimate for 2020 for state i in millions of people. Dem:Deleg:Share i is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i. S i is a vector of state-level controls, which includes the predicted tax shortfall for state and local governments divided by the state population, the average number of unemployed persons each month in the fourth quarter of 2020 per capita, the percentage change in personal income between the fourth quarter of 2019 and the fourth quarter of 2020, the total direct expenditures from state and local governments per capita, the acres of federal lands per capita, and the log of population density for state i. k b represents bill fixed effects and e i is an error term. Observations are weighted by state population and standard errors are clustered by state. *** p < 0.01, ** p < 0.05, * p < 0.1 Where Outcome c ib is funding in category c in state i and bill b. Funds per capita for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies are the dependent variables in Columns 1 to 4, respectively. In Columns 5 to 8, Outcome c ib is Proportional Share Of Bill ib , which is the ratio of state i's share of funding in category c to state i's share of the US population in bill b. Columns 5, 6, 7, and 8 show this proportional share of funds for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies, respectively. Dem:Deleg:Share ib is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i when bill b was passed. Unified b is a dummy that takes a value of 1 when the Democratic Party assumes unified control of the US House, Senate, and Presidency in 2021. We interact this dummy variable with Dem:Deleg:Share ib . k b and k i represent state and bill fixed effects, respectively. e ib is an error term. Standard errors are clustered by state. *** p < 0.01, ** p < 0.05, * p < 0.1. (2021) Where Outcome c ib is funding in category c in state i and bill b. Funds per capita for total funds, funds to state governments, funds to counties and municipalities, and funds to educational agencies are the dependent variables in Columns 1 to 3, respectively. In Columns 4 to 6, Outcome c ib is Proportional Share Of Bill c ib , which is the ratio of state i's share of funding in category c to state i's share of the US population in bill b. Columns 4, 5, and 6 show this proportional share of funds for relief, funds to transit, and funds to education, respectively. Dem:Deleg:Share ib is the averaged share of state US representatives and US senators that are members of the Democratic Party in state i when bill b was passed. Unified b is a dummy that takes a value of 1 when the Democratic Party assumes unified control of the US House, Senate, and Presidency in 2021. We interact this dummy variable with Dem:Deleg:Share ib . k b and k i represent state and bill fixed effects, respectively. e ib is an error term. Standard errors are clustered by state. *** p < 0.01, ** p < 0.05, * p < 0.1

Partisan Representation in Congress and the Geographic Distribution of federal funds

States Dole Out Cash to College Kids

Divided Government, Fiscal Institutions, and Budget Deficits: Evidence from the States

Equal Votes, Equal Money: Courtordered Redistricting and Public Expenditures in the American States

Slicing the Federal Government Net Spending Pie: Who Wins, Who Loses, and Why

Fiscal Effects of COVID-19

The President and the Distribution of Federal Spending

Cardinals or Clerics? Congressional Committees and the Distribution of Pork

Youth Enfranchisement, Political Responsiveness, and Education Expenditure: Evidence from the US

Political Institutions and Policy Choices: Evidence from the United States

The Electoral Dynamics of the Federal Pork Barrel

The Fiscal Effects of the COVID-19 Pandemic on Cities: An Initial Assessment

Potential Impact of Additional Federal Funds for Medicaid HCBS for Seniors and People with Disabilities, Kaiser Family Foundation

US Fiscal Federalism During the COVID-19 Pandemic

Forthcoming) Medicaid and Fiscal Federalism During the COVID-19 Pandemic

Implications of the Covid-19 Pandemic for State Government Tax Revenues

The COVID-19 Pandemic and the Revenues of State and Local Governments: An Update, AEI Economic Perspectives

Committee for a Responsible Federal Budget (2021) COVID Money Tracker

Noisy Retrospection: The Effect of Party Control on Policy Outcomes

Do Political Parties Matter? Evidence from

State and Local Government Finances in the COVID-19 Era

Federal Aid to School Districts during the COVID-19 Recession

State and Local Government Employment in the COVID-19 Crisis

Small States, Big Pork. Quart

Regional Favoritism

States in Fiscal Distress

Endogenous Federal Grants and Crowd-Out of State Government Spending: Theory and Evidence from the Federal Highway Aid Program

Legislative Representation, Bargaining Power and the Distribution of Federal Funds: Evidence from the US Congress

Representation and Public Policy: The Consequences of Senate Apportionment for the Geographic Distribution of Federal Funds

Political Parties and the Distribution of Federal Outlays

The Impact of Federal Spending on House Election Outcomes

Congressional Distributive Politics and State Economic Performance

Vote View. University of California

19-trillion-what Medicaid and Chip Payment Access Commission (2021) Annual Analysis of Disproportionate Share Hospital Allotments to States, US Department of Health and Human Services

States and the Fiscal Experiment Flowing from Washington

State Responses to Fiscal Crises: The Effects of Budgetary Institutions and Politics

Democrats, Republicans, and Taxes: Evidence that Political Parties Matter

Strength in Numbers? Representation and Redistribution in the European Union

Taxes, Transfers, Progressivity, And Redistribution: Part 2, Tax Notes

Local Area Unemployment Statistics: Unemployment, Seasonally Adjusted

Fiscal Year 2021 American Rescue Plan Act Supplemental Public Transportation Apportionments and Allocations. US Department of Transportation

Emergency Assistance for Non-Public Schools. US Department of Education

Federal Land Ownership: Overview and Data

For Meaningful Federal Aid to States, Why Not Reimburse Unemployment Compensation Payments? Tax Foundation

Estimates of State and Local Government Revenue Losses from Pandemic Mitigation. Cleveland Fed District Data Brief

How Much Help Do State and Local Governments Need? Updated Estimates of Revenue Losses from Pandemic Mitigation. Cleveland Fed District Data Brief

The Political Economy of New Deal Spending: An Econometric Analysis