1 Public Choice 95: , Kluwer Academic Publishers. Printed in the Netherlands. 335 The flypaper effect: Identifying areas for further research STEPHEN J. BAILEY & STEPHEN CONNOLLY Department of Economics, Glasgow Caledonian University, Glasgow G4 0BA, Scotland, U.K. Accepted 31 October 1997 Abstract. The flypaper effect literature dates back three decades. It is overwhelmingly neoclassical in approach, attempting to improve the median voter demand model and/or temper it with public choice supply-side perspectives. Whilst this approach provided valuable early substantial analytical insights, more recent papers seem to have contributed only marginally to an understanding of the flypaper effect, if indeed it exists at all. Hence, as well as providing a long-overdue comprehensive review of the literature on the flypaper effect, this paper identifies more productive avenues for further research. 1. Introduction The flypaper effect relates to the impact on exhaustive expenditures of a general lump-sum grant paid by central to local government. The traditional theories of the effects of grants-in-aid showed how, in terms of the stimulatory effect on local government spending, a general lump-sum grant has the same effect as an equivalent increase in the disposable incomes of the individuals in the community. In the traditional theories, the increase in spending, by local governments and by their local citizens, on goods and services reflects only the preferences and utility functions of the latter. However, studies have shown that a general lump-sum grant to a community has a far greater stimulatory effect on local government spending than the equivalent increase in individual incomes. This is puzzling since the traditional theory assumes that local government expenditures (both in total and in terms of the distribution of spending across the various services provided) are determined through democratic vote-counting processes. In effect, it assumes that the local voting system clearly and consistently reflects the utility functions of the local community of voters and that local politicians and bureaucrats act in accordance with voting outcomes. Hence, the additional public expenditures generated by an increase in local voters disposable incomes or, alternatively, by an increase in lump-sum intergovernmental grants of the same monetary amount received by their local government, should be identical. The fact that empirical research finds that these outcomes are
2 336 not identical casts doubt on the traditional theory which effectively assumes that grants income is returned to local taxpayers either directly via rebates or indirectly via reduced local taxes. Provided that the results of the empirical studies are accurate, the flypaper effect therefore either disproves the traditional theory of intergovernmental grants, or demonstrates that it is not fully developed enough to explain these seemingly contradictory results. Alternatively, the empirical results are themselves flawed because statistical and/or specification errors lead to an overestimation of the flypaper effect. Although fifty or so publications in journals and textbooks have attempted to find theoretical and/or empirical explanations of the flypaper effect, the search for a comprehensive and definitive explanation still proceeds apace. Many different avenues of research have been followed but the end result is always frustrating in that, at best, only partial explanations have been found and doubt remains about the size, and even existence, of this effect. The earlier theoretical and empirical research has been almost completely restricted within the confines of neoclassical economics, the attempt being to improve the median voter demand model and/or temper it with public-choice perspectives and to develop more robust empirical estimations of that model. Considering the substantial analytical and empirical insights provided by some of the earlier research, it seems that more recent research has provided rapidly diminishing marginal returns to intellectual and empirical effort. Hence, it would be opportune to provide an up-to-date review of all the literature in order to assess the achievements of the various avenues of research and identify the most promising avenues for future research. It is the contention of this article that much of the more recent research has lacked strategic direction, has too readily accepted the validity of the median voter demand model in testing the median voter hypothesis, and has often simply tried yet another variant of earlier econometric tests. Such apparently promising avenues of research have all too often ended prematurely in culs-de-sac. 2. A summary of the traditional theory of grants-in-aid There are three main conclusions arising from traditional grants-in-aid theory. First, general lump-sum and specific lump-sum grants have the same effects on grantee spending because they have only an income effect. Second, openended matching grants have a greater stimulatory effect on grantee spending than equivalent lump-sum grants because they have both income and substitution effects. Third, general lump-sum grants have similar (or the same) stimulatory effects on grantee spending as an equivalent rise in income in the community.
3 337 Traditional indifference curve theory maps the preferences of an individual between two alternatives (Wilde, 1968, 1971; and Bradford and Oates, 1971). Mapping the preferences of the decision-making body of the grantee authority is in principle the procedure likely to give the best predictions of the effects of grants (King, 1984: 89). Both Scott (1952) and Wilde (1968) assume that the decision-making body satisfactorily reflects the preferences of the community. Hence, the positive predictions of this model are similar to the predictions of both the median voter and the community indifference mappings, the latter being used by Oates (1972), Boadway (1979) and King (1984) because of their ability to make both positive and normative predictions. However, whilst Oates (1972) believes that the median voter is essentially the same as the community, Buchanan (1988) believes there is no comparison between the two. Scott (1952: 393) notes that the redistributive effects of a grant may result in the community indifference curves crossing and giving no clear indication of the desirability of contemplated changes. This problem is accepted by users of community curves and is normally dealt with by assuming that there are no redistributional effects. This is a major deficiency because many grants-in-aid programmes are intended to have redistributive effects. The median voter framework is used by Romer and Rosenthal (1980) and by Bradford and Oates (1971). The Bradford and Oates model shows how the traditional equivalence conclusions of the models of Wilde (1971) using the individual framework, can also be derived through a model of simple majority voting with fixed tax shares. While not directly referring to the median voter, the Bradford and Oates model is implicitly a median voter model. They refer to the well-known theorem of Duncan Black [which] informs us that the equilibrium level of provision of the collective good is the median peak, (i.e., the median of the members most preferred budgets) (Bradford and Oates, 1971a: 425), and apply this theorem to their model. The well-known traditional result of the median voter model can be demonstrated in Figure 1. A general lump-sum grant of CD received by the local government shifts the budget line from AC (pre-grant) to BD (postgrant), the equilibrium shifting from e 1 to e 2. An equivalent increase in the disposable income, equal to AB, of the median voter would have exactly the same result as a lump-sum general grant of the same amount (the equivalence theorem) since it would also produce the budget line BD with equilibrium again at e 2. This outcome will be the case as long as the goods are normal goods. This conclusion is reached in all of the traditional theories, irrespective of which indifference mapping is used because the neo-classical assumptions underlying all of the traditional models are the same. However, the empirical
4 338 Figure 1. Possible effects of varying general lump-sum grants and median voter income. evidence (summarized in Table 1) contradicts this theoretical result in finding a significant difference between the stimulatory effects of an unconditional general lump-sum grant and an equivalent increase in individuals incomes. Hence, the traditional model appears invalid. In Table 1, de/dg denotes the rate of change of expenditure relative to the rate of change of intergovernmental grant. Gramlich and Galper s (1973) measure of de/dg was Increases in individual s incomes yield the measure de/di (where I represents income) which lies between 0.05 and Hence de/dg is more than four times greater than de/di the flypaper effect. Whilst there is a wide disparity in the measures of de/dg, varying between 0.25 and 1.00, the empirical results overwhelmingly confirm the flypaper effect. This result is demonstrated in Figure 1. ICP represents the income consumption path, showing the locus of consumption choices arising from the increase in median voter income. However, GCP (the grant consumption path) represents the empirical result where receipt of a lump-sum grant has a greater stimulatory effect on local government spending than an equivalent increase in median voter income, X 3 being greater than X 2. A flypaper effect occurs as long as GCP lies below ICP.
5 339 Table 1. Some empirical studies of the effects of general lump-sum grants Author Year Sample de/dg Gramlich and Galper 1973 Time series data of federal grants to 0.43 local and state governments Gramlich and Galper 1973 Cross-section data of federal and 0.25 state aid to 10 large urban governments Feldstein 1975 State block grants to Massachusetts 0.60 towns Bowman 1974 West Virginia untied aid to independent 0.50 school districts Weicher 1972 State school aid to fiscally independent 0.59 school districts Inman 1971 Panel study of 41 city budgets 1.00 Weicher 1972 Untied aid to 106 municipal governments 0.90 Nathan et al Monitoring of revenue sharing in state or local governments Olmstead, Denzan, and Roberts 1993 Missouri state aid to local school 0.58 districts Case, Hines, and Rosen 1993 Federal grants to 48 states Sources: Gramlich (1977), Hines and Thaler (1995), Becker (1996). Notes 1. Dollery and Worthington (1996) provides a summary of the major studies of the flypaper effect undertaken during the 1980s and 1990s. Although no estimates of de/dg are given, there is overwhelming evidence in support of the flypaper effect. 2. Becker (1996) provides results for logarithmic and linear estimates 3. Questioning the empirical evidence If errors, of whatever form, exist in the empirical studies, then doubt may be cast on the existence or size of the flypaper effect. The wide variance in the level of the flypaper effect (Table 1) suggests that a number of the studies overestimate the size of the flypaper effect. There are three main types or errors Mis-specification of the type of grant King (1984) emphasized the importance of the type of grant to the empirical result, an open-ended matching grant having a greater stimulatory effect than a lump-sum grant of the same amount. This is because, whilst both types of
6 340 grant have an income effect, only the matching grant has a substitution effect. Chernick (1979) observed that a non-matching grant may effectively become an implicit matching grant if the grant administrators award the grants to the recipient who is willing to commit a substantial amount of their own funds to the project. Chernick (1979) found there to be a positive and significant relationship between the cost contribution of the recipient and the size of the grant awarded, ceteris paribus (Fisher, 1982: 338). Chernick s results applied to HUD Basic Water and Sewer Facilities grants in the USA between 1966 and 1972 and have been criticized for their inapplicability to general revenuesharing grants which are truly unconditional (Hines and Thaler, 1995). Nonetheless, the model does highlight a problem for empirical studies. If this process of negotiation occurs for other grant programs, then the flow of causation assumed in many of the models is questionable. The assumed flow from the amount of the grant to its stimulatory effect on the level of expenditure, can occur the opposite way, with the amount of the grant awarded dependent upon the grantee s level of expenditure on the project. This mutual dependency destroys the distinction between the dependent and independent variables and therefore invalidates attempts to model the flypaper effect. Barnett (1993) notes that joint use of lump-sum and matching grants within a programme of grants-in-aid is not uncommon and characterized the system of intergovernmental grants in England and Wales prior to Moffitt (1984) proved that treating such project grants differently in econometric terms could eliminate the flypaper effect. Where the expenditure level of the local government and the matching rate they received (and thus the marginal price) are determined simultaneously, use of simultaneous equations avoids creating an upward bias in the estimates. Moffit s study of the AFDC (aid for dependent children) programme in the USA shows that when the AFDC grants are treated in this way, any flypaper effect disappears. Megdal (1987) also found upward biases using ordinary least-squares (OLS) models. However, after comparing an OLS model with one in which the bias is eliminated through the freeing of the correlation between price, income and the error term, Wyckoff (1991: 316) concluded that there is no doubt that Moffitt and Megdal have pointed out the theoretically correct way to estimate demand functions in the face of closed-ended matching grants, but their correction does not explain the flypaper effect Use of an inappropriate functional form Overestimation of the flypaper effect may also occur because most studies use a linear, rather than logarithmic, functional form because of its ease of use. For example, Becker (1996: 97) finds that linear equation produces
7 341 estimates of the spending responses to grants that are inflated by a factor of nearly six, while the use of a logarithmic form with identical data and explanatory variables provides no evidence of the flypaper effect. In a similar vein she finds that not taking into account the biases created by endogeneity can result in the flypaper effect being inflated by a factor of almost ten. In effect, this is a measure of the impact of the inter-dependency between the amount of grant awarded and the recipient s expenditure promises to grant administrators identified by Chernick (1979) Use of inappropriate variables or omission of appropriate variables A flypaper effect may also be created by exclusion of appropriate variables from (or inclusion of inappropriate variables in) the model. Hamilton (1983) develops an idea of Bradford, Malt, and Oates (1969), expanded upon by Oates (1977, 1981), that the quality of the residents of a jurisdiction is an important input in the provision of local public services (Hamilton, 1983: 347). The suggestion is that a given level of service can be achieved using less resources for some types of populations than for others. For example, according to educational research, children from higher income and higher socioeconomic-status communities require less public expenditure to attain a particular standard than those on lower income, and therefore with lower socioeconomic-status. Similarly, those higher-income communities have a lower level of street crime and require therefore less policing, again reducing the level of public expenditure required. Hamilton (1983) is in effect claiming that other models have omitted relevant variables (i.e., socioeconomic status) and are therefore incomplete. In its use as a proxy for socioeconomic characteristics, Hamilton uses an income as an input hypothesis in the production of education to explain a significant part of the flypaper effect. He finds that, based upon obtaining reasonable parameters for the relevant variables, technology for producing local goods, and price elasticity of output demand, a significant portion (but not all) of the flypaper effect can be explained in this way. However, Wyckoff (1991) contradicts Hamilton (1983). His model included many socioeconomic variables mentioned by Hamilton, and some more besides. While he admits that he may have omitted some variables, he included the crucial variable of the educational level of the community. Using a non-linear model in the form suggested by Moffitt (1984) and Megdal (1987), Wyckoff (1991) compared a model with the extra variables to one without. He discovered that, while the extra socioeconomic variables helped to explain expenditure levels,they do not help to explain the flypaper effect in terms of increases in expenditure subsequent to receiving lump-sum grants. This seems a reasonable conclusion since, whilst socio-economic status is
8 342 positively related to educational achievement, it is by no means clear whether higher or lower status would have a greater stimulatory effect on expenditure. Wyckoff (1991: 320) concludes that... the flypaper effect is unlikely to fade away with the inclusion of these omitted variables. A second potentially relevant variable that has been omitted is the local government s savings ratio. Econometric studies take account of median voter savings by default since they consider median voter expenditures. However, to the extent that they use local government budgets inclusive of financial balances, they overestimate municipal expenditures. This is clearly invalid because balances are used to finance future, not current, expenditures. If account was taken of municipal balances, and if savings ratios were the same for the median voter and his or her local government, the flypaper effect would disappear. Even if it is not the same as the median voter, any positive marginal propensity to save on the part of local government would mean that such econometric studies overestimate the flypaper effect. A third potentially relevant variable that has been omitted is expenditure on private sector alternatives. Local government expenditure figures necessarily exclude voter expenditures on private sector alternatives (e.g., for school education and leisure and recreation). Substantial alternative private sector provision could lead to median voter preferences for lower levels of public sector provision by the local public sector, preferences to which local governments may be reluctant or slow to respond, in the latter case because of delays inherent in adjusting service capacity to rapidly fluctuating (rising or falling) demand. 4. Theoretical explanations of the flypaper effect 4.1. Deadweight loss Hamilton (1986) argues that the deadweight loss of welfare created by local government taxation (e.g., because of disincentive to work effects) creates added costs for locally-raised tax revenues. In comparison, grants received from central government are free from these additional costs and therefore it is not as costly in terms of lost economic potential for the community to spend grant money as it is to spend taxation revenues raised locally. However, this explanation has been rejected since the marginal deadweight losses from taxation are typically far too small to reconcile the large differences between propensities to spend out of changes in grants and changes in private incomes (Hines and Thaler, 1995: 221).
9 Transaction costs Quigley and Smolensky (1992) analyse the impact of transaction costs in terms of the costliness of changes in tax rates. Local government tax rates may be determined only once each year (i.e., when budgets are set) such that a mid-year change in the disposable income of the median voter would require the full scale redetermination of the budget if tax rates were to be altered. This would be very costly in terms of both democratic decision-making and administration. Hence, the impact on local government expenditure of an increase in mid-year income will be small. Similarly, transaction costs will be considerable if additional grants are received during the financial year. Hence, high transaction costs lead to the most economically efficient situation being to increase local government expenditure by the full amount of the grant rather than returning a proportion to the voters in the form of a tax cut. Under certain scenarios of modest adjustment costs, the median voter is absolutely better off by not reoptimizing in response to a block grant but instead spending the entire grant on public production. Reoptimization will not be necessary if grant values are announced in advance of local budgeting such that Quigley and Smolensky s model would become inapplicable A low-income constraint Here, the median voter s utility is maximized subject to the constraint that the tax rate set must not leave the authority s poorest citizens with a net income below some level specified by the median voter (King, 1984: 114). Higher levels of output of local government services require higher local tax rates, assuming intergovernmental grants are fixed. At some point, increasing the tax rates set by the local authority would result in the poorest citizens having disposable incomes below the level specified by the median voter, assuming that all income groups are liable to pay the local tax. It is then possible to demonstrate the flypaper effect in diagrammatic terms (King, 1984: ). However, the assumption that all income groups are liable to pay local taxes is invalid. By definition, the poor do not pay local income tax and their payment of a local sales tax or a local property tax will be reduced by sales tax exemptions (e.g., food and fuel) and by income-related rebates for the local property tax Failure by institutional structure Failure by institutional structure negates the assumption that the political process is highly responsive to the preferences of individuals, either because of the restrictions placed upon the lower level of government by higher authority, or by the rules of the local government itself (Bradford and Oates,
10 ). It can be failure by neglect rather than by design. A flypaper effect could occur under certain conditions Failure by learning or habit Failure by learning or habit occurs when citizens don t know, and can t correctly imagine the consequences of some proposed collective action. If those preferences are confused then government failure occurs (Bradford and Oates, 1971). This could create a flypaper effect Tax capitalization Turnbull and Niho (1986) argue that the flypaper effect is due to tax capitalization arising from the payment of grants-in-aid, the resultant increase in the local property tax base leading to greater local tax receipts (for a given tax rate) and thus higher expenditures. However, it could be expected that, in subsequent years, voters would seek lower tax rates if their demand elasticities remain unchanged. This emphasizes the need for a dynamic model (see below) Disharmony of interests The traditional model s assumption is one of a harmony of interests between voters and politicians/bureaucrats. Wilde (1968) argued that disharmony could occur because politicians and bureaucrats may act in their own interests rather than those of local citizens and Gramlich (1977) explicitly pointed towards disharmony as the primary reason for the flypaper effect. There are three models of disharmony Niskanen s model of bureaucratic behaviour Niskanen (1968) assumes that bureaucrats are only interested in maximizing their own welfare, not that of the community (i.e., voters and politicians) for which they provide services. The bureaucratic welfare function contains many variables, for example, salary, perquisites of the office, public reputation, power, patronage, ease of managing the bureau, and ease of making changes. Niskanen argues that all of these are a positive monotonic function of the budget, thus making budget maximization the goal for bureaucrats seeking to maximize their own utilities. This explains the flypaper effect because an equivalent rise in the incomes of voters would have increased the budget of the bureau by a lesser amount. Schneider and Ji (1987) find that whilst competition does not consistently limit the flypaper effect, nonetheless the negative relationship between the
11 345 degree of competition and the size of the flypaper effect is consistent with Niskanen s model of bureaucratic behaviour as moderated by the Tiebout (1956) model of the local market for municipal outputs. The greater the degree of competition between municipalities attempting to attract residents by means of alternative tax/expenditure packages, the more constrained are bureaucrat s attempts to maximize their own welfares by maximizing their budgets. Hence the Niskanen model is limited in its usefulness to explain the existence of the flypaper effect because it assumes that the bureau has some degree of monopoly power, presenting governing politicians with an all-or-nothing choice The setter model Romer and Rosenthal s (1980) model is in similar vein to the Niskanen model. It assumes use of referendum voting to decide upon the level of expenditure on the single public output proposed by a legally-designated agent which knows the preferences of the voters it represents and which seeks to maximize the budget. If the proposed level of expenditure is defeated in the referendum, the level of public provision is established by an exogenous, legally-designated reversion level. The agent proposes an expenditure level where the median voter is indifferent between the proposed level (such as X 3 in Figure 1) and the reversion level (which is therefore X r ), or is marginally better off under the proposed level (i.e., an expenditure level just below X 3 ), thus ensuring maximization (or nearly so) of expenditure. In contrast to the traditional model, the rise in incomes would have no effect on the expenditure level in the setter model, so long as the rise does not result in the point of tangency between the indifference curve and the new budget line rising beyond the agent s proposed level of expenditure. Therefore, since it is assumed that the lump-sum grant has to be spent in addition to the pre-grant reversion or proposed levels, the flypaper effect occurs. However, the Romer and Rosenthal (1980) model s results are affected by different reversion levels and the type of good (i.e., normal or inferior), such that the effects of a grant can be less than an equivalent increase in income. Moreover, the assumptions of the setter model can be criticized as unrepresentative of the way local government operates, usually without referenda. Romer and Rosenthal developed their model to analyze U.S. school boards and, whilst their model can be generalized (King, 1984), their assumptions cannot. It is not clear why local voters would agree to a suboptimal level of output, e.g., any level of output other than X 2 in Figure 1. They could be forced to acquiesce to a suboptimal level of output if the legally-designated authority (agent, setter) has powers to override local voters preferences.
12 346 Whilst this may not be unreasonable in the case of U.S. school boards it would abrogate the underlying democratic rationale of local government The greedy politicians model McGuire (1975) assumes that politicians spend all the money they can get their hands on, as long as individuals are not made any worse off, in order to maximize their chances of political survival and hence their own welfares. Again, this is similar, in essence, to Niskanen s model. That both politicians and bureaucrats may not act in the best interests of the individuals they represent is a more comprehensive explanation of the flypaper effect Fiscal illusion Fiscal illusion models have been developed by Oates (1979), Courant, Gramlich, and Rubinfeld (1979) and Logan (1986). In Oates model, lump-sum grants also have substitution effects, arising because of fiscal illusion on the part of voters. Local government officials are assumed to be output maximizers and to use the grant to deceive local voters into thinking that the cost of service is less than it actually is. Reference is made to the reduced average tax price. Since, however, the marginal cost to the electorate is unaffected, the level of output is excessive. The stimulatory effects of a lump-sum grant are not equivalent to an increase in individual s incomes, because although they may generate the same true budget constraint, they do not result in the same perceived budget constraint (Oates, 1979: 29). Dollery and Worthington (1995) find some support for the fiscal illusion explanation in Australia, and Heyndels and Smolders (1994) find important illusionary effects in Flemish municipalities. Borge (1995) demonstrates that fiscal illusion unambiguously predicts a flypaper effect. However, Oates (1979) theory and the public-choice theories of Filimon, Romer and Rosenthal (1982) were rejected by Wyckoff (1991) because they were rejected statistically by his study. Moreover, both the Oates (1979) and Courant, Gramlich, and Rubinfeld (1979) models admit that while voters using average instead of marginal prices can create an illusion that may lead to a flypaper effect, this may only explain part of that effect Uncertainty and risk aversion Fossett (1990) and Turnbull (1992) develop models incorporating uncertainty and instability of grant revenue. Fossett s model also incorporates risk-averse behaviour of local officials. He believes that the uncertainty about grantsin-aid, both in terms of the level of aid and the type of aid, results in local