Download The Impact of Tobacco Settlement Funds on State Spending on Tobacco Control Programs and more Lecture notes Literature in PDF only on Docsity!
Special Interest Groups and the Allocation of Public Funds
Monica Singhal
Harvard University and NBER
This version: January 2006
Abstract A long-standing puzzle in the Öscal federalism literature is the empirical non-equivalencein government spending from grants and other income. I propose a fully rational model in which violations of fungibility arise from dynamic interactions between politicians andinterest groups with the ability to raise funds for local government. The predictions of the model are tested by exploiting unique features of windfalls received by states under a settlement with the tobacco industry. Although windfalls are unrestricted, the medianstate increased spending on tobacco control programs from zero to $2.30 per capita upon receipt of funds. The marginal propensity to spend on such programs is 0.20 fromsettlement revenue and zero from overall income. States which were not involved in the settlement lawsuits spend less. The Öndings are consistent with the predictions of themodel when political partisanship is introduced: Republican governors spend less and factors which should lead to political convergence increase spending for Republicansand decrease spending for Democrats. These results cannot be explained by existing models in the literature.
E-mail: monica_singhal@harvard.edu. I am grateful to Alberto Alesina, Nava Ashraf, Keith Chen, Raj Chetty, David Cutler, Martin Feldstein, Amy Finkelstein, Ed Glaeser, Jens Hilscher, Caroline Hoxby, Larry Katz, Ben Olken,Ricardo Reis, Jonah Rocko§, Jesse Shapiro, Bryce Ward and seminar participants at Harvard University for numerous helpful conversations and comments. I thank the National Science Foundation and the National Institute on Aging fortheir Önancial support. Heather Langdon, Neil Mehta and Katherine Stanchik provided excellent research assistance.
1 Introduction
Standard models of Öscal federalism predict that grants received by local governments should be considered equivalent to increases in the income of the local constituency. Perhaps the most commonly studied violation of this fungibility principle is the áypaper anomaly: the empirical ob- servation that money "sticks where it hits." Local governments spend more from intergovernmental grants than from equivalent increases in constituent income, and grants for particular programs tend to increase spending on those programs far more than standard theory suggests. Numerous studies have documented the existence of áypaper e§ects, with estimates of the increase in local spending arising from a dollar grant ranging from 25 cents to one dollar (Hines and Thaler 1995).^1 I propose a new, rational model of government spending decisions that focuses on the potential role of special interest groups in ináuencing the allocation of public funds. In this model, special interest groups have the ability to raise funds for local governments by undertaking costly e§ort. In a dynamic setting, it is optimal for rational politicians to take the preferences of these interest groups into account when making spending decisions to ensure that groups have incentives to undertake the e§ort costs of raising funds in the future. I test the predictions of the model by examining the response of state spending on tobacco prevention and control programs to receipt of windfalls arising from state lawsuits against the tobacco industry. There are few theories in the existing literature that can explain observed violations of fungibil- ity. Filimon, Romer and Rosenthal (1982) propose a model in which agenda-setting bureaucrats are able to hide grants from voters. While this model predicts that money received by govern- ments will remain at the government level, it does little to explain why categorical grants should systematically increase expenditure in particular spending categories. Models focusing on rent (^1) Models illustrating the standard revenue equivalence proposition include Bradford and Oates (1971a, 1971b). See Gramlich (1977), Inman (1979) Fisher (1982) and Hamilton (1983) for a review of the earlier áMore recent work includes Baicker (2001), Gordon (2004), Lutz (2004), and Evans and Owens (2005).ypaper literature.
during the 1990s. Under the terms of the settlement, tobacco companies must pay states large annual sums (on the order of $7 billion per year) in perpetuity. I examine the response of state spending on tobacco prevention and control programs to receipt of settlement funds. Two key features of the settlement windfalls are advantageous for testing violations of fungibility in the allocation of funds. First, settlement money is unrestricted and use of funds is left entirely to the discretion of states. Settlement windfalls should therefore, in theory, be considered equivalent to increases in state income. Second, I demonstrate that the timing and magnitude of windfalls are plausibly exogenous to desired spending on tobacco control programs. Grants do not reáect underlying spending preferences and are truly lump-sum, without explicit or implicit matching provisions. The models proposed by Chernick and Knight are therefore not applicable in this case.^3 I Önd clear evidence of violations of fungibility in government spending decisions. Average per capita spending on tobacco control programs increased more than six-fold from the Öscal year before settlement revenues were received to the Öscal year after receipt. The marginal propensity to spend on such programs is 0.20 from settlement revenues and zero from other income. I Önd that states that did not Öle lawsuits prior to the settlement, where anti-tobacco interest groups presumably exerted less e§ort, spend signiÖcantly less on tobacco control programs after the settlement. Finally, I show that spending patterns conform closely to the predictions of the model in a world with political partisanship. Republican governors spend less than Democrats, and factors which should lead to political convergence, namely eligibility for re-election and facing an opposition controlled senate, result in increased spending by Republicans and decreased spending by Democrats. This empirical setting di§ers in at least two important ways from traditional empirical áypaper (^3) Concerns that apparent violations of fungibility may be driven by econometric misspeciÖcations, such as incorrect treatment of price e§ects arising from matching grants (Mo¢ tt 1984) or omitted variable bias (Hamilton 1983) arealso unlikely to be problematic.
studies. First, as mentioned, settlement revenues are unrestricted. Examining the response of spending on tobacco prevention and control programs to settlement windfalls therefore provides a test of fungibility but di§ers from classic áypaper since revenues were not speciÖcally labeled for such programs. Second, transfers in this case are from private industry to local government, rather than intergovernmental transfers. While these features have advantages in distinguishing among alternative models, they also raise potential caveats in generalizing the Öndings to other settings. I consider these issues in the concluding section of the paper. The remainder of the paper proceeds as follows. Section 2 presents the model. Section 3 provides background on the settlement agreement and payments. Section 4 describes the empirical methodology and data used, and Section 5 presents the results. Section 6 concludes.
2 Interest Groups and the Allocation of Funds
"Itís moral treason to me. We got all this money, then legislatures and governors who were not even in this Öght act like the money fell out of heaven and spend it onthe political whim of the day." ñMississippi Attorney General Michael Moore on state decisions to spend settlement funds on non-tobacco related programs (2001) New York Times,
2.1 Motivation
Grants-in-aid from the federal government to states are of two main types: mandatory "entitle- ment" grants, for which spending is determined by existing law, and discretionary grants, for which funding is allocated on an annual basis. In Öscal year 2003, almost 60% of federal dollars given in grants-in-aid to states, excluding Medicaid, were discretionary.^4 In this system, interest groups have the ability to ináuence grants-in-aid through contributions and lobbying e§orts. Interest (^4) Source: Center on Budget and Policy Priorities. The primary mandatory grants-in-aid from the federal govern- ment to states are through Medicaid, food stamp, and welfare programs. Most other grants-in-aid are discretionary.
2.2 Model
I begin with a simple stylized reputation model in which a long-run government player interacts with a number of short-run interest group players.^8 An interest group derives utility from spending on a particular good that it cares about: the "lobby good," z. It cannot produce z directly but can raise amount L for the local government by exerting e§ort. The government chooses spending on a variety of goods, including z, conditional on funds received from the interest group and other income Y. I do not assume political agency by the government in order to demonstrate that violations of fungibility are possible even in a framework equivalent to one in which decisions are made by a median voter. I consider the implications of the model in a world with political agency and partisanship in Section 5.5.4. I make the strong assumption that lobby groups are homogeneous and that all lobby goods enter the government utility function in the same way.^9 For simplicity, I also restrict the interest group e§ort choice to be binary. The government makes its decision simultaneously with its interest group opponent in each period. In the case in which the government receives no funds from the interest group, it solves the following problem: max z;x UG(z; x) subject to pz z + pxx Y; (1)
where z represents the lobby good (also a "good" in the government utility function) and x repre- sents other government and private voter goods. Prices in this case represent the cost of production of the various goods. Solving this problem gives the optimal choice of goods, which I denote as: (z^0 ; x^0 ). When an interest group chooses to raise funds for the government, it does so with an (^8) This model adapts standard models of reputation with a single long-run player; see Kreps and Wilson (1982), Milgrom and Roberts (1982), Fudenberg and Tirole (1998). (^9) Allowing heterogeneity in lobby goods does not alter the basic intuition of the model.
implicit understanding that the government will provide "payback" by spending the funds on the good the interest group cares about, z. Whether payback occurs depends on two factors: the type of the government and the action chosen by the government. This model assumes two types of governments: Committed and Strategic. The Committed government always chooses Reciprocate. The Strategic government can choose one of two strategies: Reciprocate or Renege. I deÖne these in the following way. Under Renege, the government breaks the implicit contract and treats the interest group funds as it would other income, maximizing UG(z; x) subject to the constraint pz z + pxx Y + L. Solving this problem leads to a choice of goods along the governmentís income expansion path: (z^0 ; x^0 ). Under Reciprocate, the government spends all the interest group funds on the lobby good, leading to the consumption choices (bz; xb). The government would prefer to allocate L across all goods and would therefore be better o§ by reneging. I assume that interest groups have utility functions such that UL(bz;e§ort) > UL(z^0 ;no e§ort) and UL(z^0 ;e§ ort) < UL(z^0 ;no e§ ort); that is, interest groups prefer to undertake e§ort and provide L if and only if the government pays them back. The above equations imply the following payo§ matrix:^10 GOVERNMENT
INTEREST GROUP
Reciprocate Renege E§ort (a; c) ( b; d) No E§ort (0; 0) (0; 0) The Nash equilibrium of the stage game is then (No E§ ort, Renege) yielding payo§s of (0; 0) even though (E§ ort, Reciprocate) results in higher payo§s (a; c) for both players. I now consider the implications of this model in a dynamic setting in which the government 1 0 (^) The zero payo§s in the second row arise from normalizing UG(z (^0) ; x (^0) ) and UL(z (^0) ;no e§ ort ) to zero for simplicity. The payo§s in the Örst row are then as follows: a = UL(z;be§ ort ), b = jUL(z^0 ; e§ ort )j, c = UG(bz; bx), and d = UG(z^0 ; x^0 ).
exceeds a threshold value: p^0 > (^) a+bb p. In the two period case, the Strategic government can Renege in period 1, revealing its type. The total payo§ to the government is then d + 0. The government can also Reciprocate in period 1 to build a reputation for commitment. If doing so causes the interest group to provide e§ort in period 2, the government gets a total payo§ of c + d. Solving yields the following necessary condition for the Strategic government to Reciprocate:
c d >^ (1^ ^ )^ (2)
The Strategic government is willing to Reciprocate in period 1 if doing so induces interest groups to provide funds in period 2. If the condition in equation (2) holds, the equilibrium depends on p^0 , the prior probability of a Committed government. If p^0 > p, the government Reciprocates in period 1 and interest groups provide e§ort in both periods. If p^0 < p, the Strategic government Reciprocates in period 1 with probability (^) (1 p^0 pa (^0) )b. Interest groups are indi§erent about providing e§ort in period 2 and provide e§ort in period 1 if p^0 >^ ^ a+bb^ ^2 = p^2 : Solving by induction to the N period case, the prior probability of a Committed government (p 0 ) required for the interest group to provide funds decreases in N geometrically at the rate^ ^ a+bb^ . For details of the solution, please see Appendix A.1. The standard revenue equivalence proposition states that government expenditure from grants and from other income should be the same: (^) @L@z = (^) @Y@z. Much of the empirical literature on categorical áypaper e§ects Önds instead that (^) @L@z > (^) @Y@z. This model implies a positive probability of the government reciprocating, thereby spending more on z when it receives a grant than if it followed the income expansion path, as long as the necessary condition given in equation (2) holds. The model thus predicts systematic violations of fungibility across spending categories consistent
with áypaper e§ects. Violations of fungibility are more likely when is high (more weight is given to future periods), holding the other parameters Öxed. Equation (2) also shows that for a given , the probability that the government Reciprocates is increasing in cd , the ratio of social welfare when the government Reciprocates to social welfare when the government Reneges. This implies:
prob
(^) @z @L >
@z @Y
= f^ ^ dc^ ; where f > 0 (3)
We should be more likely to observe violations of fungibility when there are low costs of misallocating toward the lobby good relative to pursuing the socially optimal spending path.
2.4 Alternative Models
One alternative theory is that governments spend lobby money on the lobby good because they fear voter punishment if they behave otherwise. Such a model, however, would require either behavioral preferences on the part of voters or a framework in which spending money on the lobby good provides a costly signal of some other characteristic voters care about. Another alternative is a bargaining model between interest groups and politicians. This type of model would need to explain why interest groups are more willing or able to punish the local government when the funds are for "their" good. I now test some of the predictions of the special interest group model by examining state responses to funds received under a settlement agreement with the tobacco industry.
following month. Settlement revenue is unrestricted and the allocation mechanism and use of funds are left entirely to the discretion of the states.
3.2 Payments
States receive three types of payments under the settlement: (1) initial payments, paid in Öve install- ments from 1999 to 2003; (2) annual payments, paid in perpetuity; and (3) Strategic Contribution Fund payments meant to compensate states for the costs incurred in state lawsuits, paid from 2008 to 2017. The two major adjustments made to annual settlement payments are an ináation adjustment and a volume adjustment. Annual payments increase by the CPI or 3%, whichever is higher. The volume adjustment is based on increases or decreases in the number of cigarettes shipped nationally relative to a base volume. The volume adjustment is not state-speciÖc. Initial payments are subject to the volume adjustment but not the ináation adjustment. At the time of the settlement, total unadjusted payments made to settling states under the agreement through 2025 were projected to be almost $206 billion (Table 1 ), or $120 billion in present value terms using a discount rate of 4%. Table 2 provides a summary of settlement disbursements to states in Öscal year 2002. The average amount of revenue a state receives is $100 million annually, which corresponds to $22 per capita and $100 per smoker. Initial and annual payments are distributed among the states according to Öxed state allocation percentages. Base allocation percentages are calculated using a formula that equally weights two factors: the stateís share of total direct medical costs related to smoking and the stateís share of smoking-attributable Medicaid expenditures (Modisett 1997). Total direct medical costs related to smoking represents smoking-related health costs incurred by all payment sources in a state in 1990. Smoking-attributable Medicaid expenditures represents
the amount of a stateís Medicaid expenditures directly attributable to smoking and to illnesses associated with smokeless tobacco use for individuals over 18 in 1993.^14 Two adjustments were made to direct medical costs: Ögures were multiplied by 1.28 to reáect ináation in medical costs between 1990 and 1993 and Medicaid costs were then subtracted to prevent double counting of these expenditures. The percent of the total settlement amount allocated to state i is then given by the following formula:
percenti = 0: 5
@ (^) PSM CDi i^ SM CDi
A
1993
@ (^) PAdjDM Ci i^ AdjDM Ci
A
1993
where SMCDi and AdjDMC (^) i are the smoking-related Medicaid costs and the adjusted direct med- ical costs for state i. Negotiations among states at the time of the settlement resulted in some small adjustments to these base percentages. Table 3 illustrates the allocation percentages as they would have been had the above formula been followed as well as the actual percentages under the settlement. Di§erences between the simulated and actual allocation percentages may not be completely random (it is unlikely to be a coincidence that California and New York receive exactly the same shares) but are generally very small. The coe¢ cient of correlation between the two is 0.99, and proxying for actual settlement revenues using the simulated allocation percentages does not a§ect the results. The size of a stateís windfall in a given year is then the aggregate annual payment, determined under the terms of the settlement, multiplied by its allocation percentage. Allocation percentages were Öxed at the time of the settlement agreement, so statesíspending decisions do not a§ect future 1 4 (^) The population of each state was categorized into non-smokers, current smokers, former smokers with less than 15 years exposure and former smokers with greater than 15 years exposure.smoking-related medical condition and then the level of expenditure was estimated as a function of smoking, medical The e§ect of type of exposure on each conditions and health status.paid for by all sources per year. The costs do not reáModels controlled for age, race/ethnicity, poverty status, marital status, education,ect lifetime medical care costs but rather medical care costs medical insurance, region, seat-belt use and obesity. See Modisett (1997) for further details on calculations.
tal evidence indicates that these groups felt that settlement dollars should be spent on tobacco prevention and control programs. The following quote is typical:
"A compassionate but naÔve person would expect the states to use their $246 billion [sic] windfall to try to prevent more people from su§ ering and dying from cancer, emphysemaor other smoking related illnesses. If this is blood money, why not try to stop the bleeding?the money ... Ií Ah, but the greedy deal makers in our state capitals have other plans form talking about construction projects. Paying bills, new non-medical programs ... Most of this spending would be Öne if it came out of state tax revenue, but... this money should not be poured into general funds. It should be used to help prevent and cure disease."1999) ñJudy Jarvis, radio host and lung cancer victim (New York Times,
I therefore focus my analysis on state spending on tobacco prevention and control programs. Figure 1 illustrates the number of states allocating substantial funds toward such programs over time. Although the settlement agreement was reached in 1998, states Örst received funds in the middle of the 2000 Öscal year. The number of states spending at least $0.50 per capita on tobacco control programs increased almost six-fold from six states in Öscal year 1999 to thirty-four states in Öscal year 2001. The Öve states with substantial programs prior to the settlement^18 funded their programs primarily through increases in excise taxes on cigarettes. The remaining states allocated virtually no state funds toward such programs prior to the settlement (Figure 2 ). Among these (non-prespending) states, mean per capita spending increased from only $0.04 in 1999 to $2. in the year after settlement funds were received. Despite displaying virtually no preference for spending on tobacco control programs through the mid-to-late 1990s (a period of substantial budget surpluses for most states), all but one of the non-prespending states had instituted such a program by Öscal year 2002.^19 States with pre-existing programs also responded to settlement revenues, increasing spending from an average of $4.15 per capita in Öscal year 1999 to $7.67 in Öscal year
1 8 1 9 (^) Arizona, California, Maine, Massachusetts and Oregon. The only state not allocating state funds toward tobacco prevention and control by Öscal year 2002 was Tennessee.
In the next sections, I test the predictions of the interest group model more formally.
4.2 Econometric SpeciÖcation
The empirical strategy I employ to test for violations of fungibility is a variation on a traditional Öxed e§ects speciÖcation. By exploiting both the time series and cross-sectional variation in settlement revenue receipt, I test for violations of fungibility in two ways. Consider the following regression framework:
T obacco Controlit = 0 + 1 (Settlement revenue)it + 2 (Income)it + (^) t + i + Xit + "it (5)
where (^) t is a set of year dummies, i is a set of state dummies and Xit is a set of time-varying state controls. In a standard Öxed e§ects setting, the key parameter of interest is 1 , which would be interpreted as measuring the e§ect of settlement revenue receipt on tobacco control spending. (^) t would be included primarily as a control to pick up underlying trends in spending over time. The tobacco control experiment is unusual in that the pre-trend in tobacco control spending is essentially áat and close to zero. A large, discontinuous increase in spending occurs when settlement revenues are received (Figure 1 ). Thus, both 1 and the (^) tís have causal meaning and can be used to test for violations of fungibility. Coe¢ cients on the time dummies pick up changes in tobacco control spending within a state over time; the Örst test is whether there exists a discontinuity in spending at the time of settlement revenue receipt. The second test is whether the marginal propensity to spend on tobacco control from settlement revenues is higher than the marginal propensity to spend from state income. The relevant test is 1 > 2 (rather than 1 > 0 ) to distinguish the income e§ect component of settlement revenue receipt from a true fungibility e§ect. The (^) t coe¢ cients indicate whether states spent on the lobby good when they received lobby funds and 1 indicates whether states that received more lobby money spent more on the
They do not include funds directed toward tobacco research, health services, tobacco farmers or tobacco dependent communities (CDC 2001). In a few cases, appropriations were made for mul- tiple Öscal years at once or revenues were set aside in trust funds. The CDC includes the full appropriation amount in the year in which it was allocated. A secondary data source is information on allocation of tobacco settlement revenue compiled by the National Conference of State Legislatures (NCSL). Reported state allocations include funds for community and school-based tobacco-use prevention programs, media campaigns, tobacco control measures and tobacco cessation treatment (NCSL 2002). The major advantage of the NCSL data is that they contain allocations for Öscal years 2003 and 2004, whereas CDC data are currently limited to Öscal years 2001 and 2002. The main drawback is that NCSL data include only tobacco control spending from settlement revenues. Data from the two sources are close in most cases, but NCSL data underreport spending in states where settlement revenues were not the only funding source for tobacco control programs. In addition, NCSL data do not include money set aside in endowment funds. Both data sources reáect appropriations for spending related to tobacco control at the beginning of the Öscal year and may di§er from actual expenditures. I use CDC data whenever possible and supplement the analysis with NCSL data as a speciÖcation check and also in cases in which adding additional years of data is especially useful. The two data sources produce almost identical results. Settlement revenues received by states were tabulated by the National Association of Attorneys General (NAAG) and reáect the amount disbursed to each state in a given Öscal year.^21 Sources on the remaining variables are given in Appendix A.2. 2 1 (^) Arkansas and Missouri did not immediately achieve state-speciÖc Önality. Their settlement disbursements for Öscal 2001 therefore reáanalyses. ect both Öscal 2000 and 2001 payments. I exclude 2001 data for these two states in all
5 Results
5.1 Main Findings
Table 5 provides summary statistics on per capita tobacco control program allocations for Öscal years 1998, 1999, 2001 and 2002.^22 The mean amount allocated toward tobacco control after receipt of settlement funds is a little more than $3.00 per capita. There is substantial variation in allocation amounts across states. I test for violations of fungibility by estimating the following equation:
T obacco Controlit = 0 + 1 (Srevit Srevt) + 2 (Incit Inct) + (Af ter) + i (6)
Srevit and Incit are per capita settlement revenue and income for state i in year t, Af ter is an indicator which is equal to 1 for Öscal years 2001 and 2002 and i is a state Öxed e§ect. The two tests are: 1 > 2 and > 0. Settlement revenue and income are measured in deviations from the year mean so that can be interpreted as the change in spending at the average levels of these variables. Per capita settlement revenue and income are measured in real 2002 dollars. I Önd strong evidence that states violate fungibility in spending decisions as predicted by the interest group model (Table 6 ). Column 1 gives the results when state Öxed e§ects are not included.
1 , the propensity to spend from settlement revenue, is 0.18 and signiÖcant at the 1% level;^2 , the propensity to spend out of income, is essentially zero and insigniÖcant. The average increase in spending upon receipt of settlement funds, ; is 2.93 and also signiÖcant at the 1% level. This represents an almost six-fold increase in spending. Adding state Öxed e§ects does not a§ect the coe¢ cient estimates, as shown in Column 2. 1 increases to 0.21 and to 3.07. In columns 3 and 4, I replace the indicator for Af ter with a full set of time dummies. It is clear that the e§ect 2 2 (^) Data are not available for Arizona and Massachusetts for Öscal year 2002.
