Some Simple Tests of Voting and Agenda Setting Sean Corcoran New York University Thomas Romer Princeton University Howard Rosenthal New York University Abstract* We analyze a set of election data that permit simple tests of rational voting and agendasetting. The rational voting test is an analysis of whether aggregate election results are consistent with single-peaked preferences. The prediction tested is that there is more opposition to the second of two budget proposals that are voted on simultaneously. Unlike the standard binary choice setting, not all voters have weakly undominated voting strategies. The game between the voters can nonetheless be solved simply by the iterative application of weak dominance. The agenda-setting prediction tested is that agenda setters should make one proposal rather than two when making a single proposal is an option. The work also discloses a regularity with regard to voter participation. There is typically less participation on the second proposal. The data for the tests come from Oregon school financial elections in the years 1980 through 1983. * We thank Alexander Ruder for very extensive and effective research assistance. We also thank seminar participants at Columbia University for comments. 1 Introduction We analyze a set of election data that permit simple tests of rational voting and agendasetting. For many years, school district operating budget referenda were an instance of democracy in about half of the states. Oregon, which had such referenda for many years, until the 1990s, held such elections and served as the quintessential case that motivated the RomerRosenthal (1978, 1979, 1982a, 1982b) model of agenda control by a budget-maximizing bureaucrat, a.k.a. the school board. For a brief time from 1980 to 1983, Oregon school elections had an institutional feature that generated a useful natural experiment. 1 In some cases voters were asked to vote simultaneously on two expenditure amounts that together comprised the locally funded portion of the school budget. We use data from these referenda to test whether aggregate election results are consistent with rational voter behavior under the assumption of single-peaked preferences. Much research in political science explores the situation where voters make a simple binary choice. Rationality in this instance is just voting for the preferred outcome. The “game” between voters is uninteresting. Often, however, as in the literature on split-ticket voting, voters are asked to make two or more choices. Testing rationality in such instances becomes complicated because of both institutional complexities and data limitations as in Mebane (2000). In contrast, we develop a prediction that has a straightforward, direct test. The 1980-83 Oregon data also allow for a new test for agenda-setting behavior. In the first two years of this period, school boards were required to split budget requests above a fixed value into two expenditure proposals. In the latter two years, boards could elect to propose a single expenditure amount. We assume the agenda setter is a budget maximizer (Niskanen, 1971; Romer and Rosenthal, 1978, 1979). We show that a budget-maximizing agenda setter should 2 always make a single proposal. We test this prediction by examining whether referenda were shifted to single proposals in the later period. The election data also disclose an interesting regularity in the number of people who cast votes on the two expenditure levels. We simply present the regularity. We briefly discuss interpretations in light of the voter information models developed by Feddersen and Pesendorfer (1997, 1999) and the non-voting from indifference model of Hinich and Ordeshook (1969). Theory We consider the following expenditure referendum. Voters are presented with a choice between a reversion level of spending Q and a higher level A. At the option of the agenda setter, there may also be a separate additional amount B to be voted on at the same time. Suppose both A and B proposals are present. If both A and B receive majorities, actual expenditure becomes Z ŁA+B. If only A receives a majority, the actual expenditure is just A. If neither receives a majority or if only B receives a majority, the expenditure is Q. Thus, the outcome set is {Q,A,Z}. Voter preferences are assumed to be single-peaked. We also assume for now that all voters vote on both the A and B ballot (i.e., no abstention). Finally, the reversion amount Q in future years is independent of current expenditure decisions. (This was the case in Oregon during the period of our data, though the rule changed in later years.) Thus, we can limit our consideration to a oneshot game. 2 Rational voting The pure strategy set of the voters can be described as {For A, For B},{For A, Against B}, {Against A, For B}, {Against A, Against B}. With single-peaked preferences, there are only four voter preference types (ignoring the knife-edge cases of indifference). Denoting the strict preference relation by , these are 3 ZAQ, AZQ, AQZ, and QAZ. Any strategy is a best response if a voter is not pivotal. We can study whether these types have weakly undominated voting strategies by exhaustively considering their best responses in situations when they are pivotal. A voter can be pivotal in one of three ways. First, his vote is decisive in both the A and B elections. Second, the vote is decisive only in the A election. Third, the voter pivots only in the B election. The analysis is summarized in Figure 1. The situation for the low demand type, QAZ, is shown in the top table in the figure. When an agent of this type is pivotal in both elections, a best response must include voting against A since a vote against A leads to the most preferred outcome of Q. When the agent is pivotal only in the B election, a best response must include voting against B, to obtain Q if A is losing and A if A is winning. Voting against both A and B is a best response if the agent is pivotal only in the A election. Consequently, voting against both A and B is uniquely a weakly dominant strategy. The situation for the high demand type, ZAQ, shown in the second table in the figure, is simpler. When an agent of this type is pivotal in both elections, voting for A and for B is a unique best response. It can be readily checked that voting for A and B is a best response when the agent is pivotal in only one election. Hence, agents of this type have a weakly dominant strategy of voting “Yes” in both elections. The intermediate type AZQ, shown in the third table in Figure 1, has a unique best response of voting for A and against B when the agent is pivotal in both elections. It can be readily checked that this strategy is also a best response when the agent is pivotal in only one election. Consequently, voting for A and against B is a weakly dominant strategy for this type. 4 Finally, the type AQZ does not have a weakly dominant strategy. This intermediate type must also vote for A and against B when pivotal in both elections. This strategy is also a best response when an agent of this type is pivotal in only the B election. But it is not a best response when the agent is pivotal in only the A election. If, as shown in the bottom table of Figure 1, B wins the B election, the agent must vote against A to obtain Q rather than the least preferred outcome, Z. The agent “knows” however that if the other three types use their weakly dominant strategies, A will receive at least as many votes from these types as does B. Moreover, the agent also “knows” that, for other AQZ agents, voting for B is weakly dominated since a vote for B can only pivot to the least preferred outcome, Z. Thus, the type will never be in a position of being a pivotal voter for A but not for B. The only instance in which B wins is one in which the high demand types are a majority, and if the high demand types are pivotal for the B election they will also be pivotal for A. By iteration of weak dominance, therefore, voting for A and against B is a best response for the AQZ type when the type is pivotal. In summary, ZAQ types are predicted to vote “Yes” on both the A and B votes, QAZ types are predicted to vote “No” on both. The two intermediate types are predicted to vote “Yes” on A and “No” on B. Proposition 1. If there is at least one voter of either type A QZ or AZQ, then the “No” vote will be greater in the B election than in the A election. If there is no such voter, the “No” vote in the B election will equal that in the A election. Note that in standard one-election situations with two alternatives, weak dominance is used to argue that all voters should vote sincerely. From the oodles of Nash equilibria in a voting 5 “game”, weak dominance selects the natural one based on sincere voting. In the two-election case we model here, there is one type, AQZ that does not have a weakly dominant strategy. We can, however, use iterated weak dominance to select a unique Nash equilibrium to the voting game. 3 Note further that the strategies used by the voters do not depend on their knowledge of the number of voters of each type. Proposition 1 does depend on the assumption of single-peaked preference. In Figure 2 we show the situation for a preference type that does not have single-peaked preferences. This corresponds to individuals who prefer a high level of school expenditures but would rather not have public education if the schools are mediocre. (See Stiglitz, 1974.) When an agent of this type is pivotal in both elections, the agent must vote both for A and for B. If the agent is pivotal in only the A election and B is losing, the agent’s best response is to vote against A. So there is no weakly dominant best response. Moreover, the situation cannot be resolved by applying further iteration. When this agent votes against A, voting for B remains a best response, so A could get fewer votes than B, Agenda setting Suppose that the choices of the A and B amounts are made by a budget-maximizing agenda setter. The agenda-setter is constrained, by law, as to the maximum amount of expenditure that can be set in the A vote. Denote this as Amax. Anything greater requires a B vote. We assume further that if a majority of the voters would not approve any expenditure greater than Amax, the setter will make a single A proposal. If the setter makes an A proposal and a B proposal, budget maximization implies the A proposal will be Amax. Now suppose that a majority of the voters would approve expenditure at least equal to Amax. Relax the requirement that the setter must have a separate B vote on amounts greater than 6 Amax, and assume the agenda-setter is permitted to make a single proposal C to the voters. Assume further that there is at least one intermediate type voter who prefers Z (=Amax+B) to Q but prefers Amax to Z. Proposition 2. When given the option, the setter will make a single proposal C rather than make separate proposals A and Z. Corollary. C > (Amax+B) if a majority of voters support B. The proof of Proposition 2 is simple. We outline it for the case of symmetric preferences. First consider the vote for a single proposal C versus the reversion Q. With symmetric preferences, those with ideal points above the cutpoint (C+Q)/2 will vote for C and those below this cutpoint will vote for Q. Similarly, in the case of two separate elections, the cutpoint in a vote over the proposal Amax versus the reversion Q will be (Amax+Q)/2, and the cutpoint in a vote over the proposal Z versus Amax will be (Z+Amax)/2. In a jurisdiction where the majority of voters would support the higher proposal Z, voters would also support a single proposal C*=Z. (The cutpoint (Z+Q)/2 < (Z+Amax)/2 since Q < Amax). Indeed, the proposal C* will receive more votes than B receives in a B election. Consequently, for some C > C*, the setter can obtain a larger budget than C*. In a nutshell, in the ultimatum game between the setter and the voters, the setter’s threat is greater with one election than with two. 7 Empirical Context For many years, Oregon’s school finance system was characterized by school district “financial elections”. These elections were referenda in which school boards would propose a local budget that required approval by a majority of the voters. Through 1979, the proposals were simple all-or-none proposals to add additional expenditure to a local reversion expenditure (Q) that was fixed exogenously. The reversion expenditure varied by district, and could be zero or a strictly positive amount. The reversion could be sufficiently large that a board might not hold a referendum. In the event a proposal failed to achieve majority support, the board had opportunities to hold an additional election. If an additional election were held, the board could alter the proposed amount of expenditure. While in most districts the first election proposal was approved, in some districts multiple proposals (very rarely as many as six elections) were required. For the 1980-81 through 1983-84 school years, the traditional system was altered by property tax rebates that depended on characteristics of the taxpayer, characteristics of the district, and the amount of proposed expenditure (Hartman and Hwang, 1985). 4 For proposals at or below an exogenously fixed amount, which we denote Amax, a rebate was provided to some taxpayers in the school district. For expenditures above Amax, additional (B) expenditures received no rebate. Therefore, some taxpayers in the district were subject to two tax prices—a lower one for A expenditures and a higher one for B expenditures. The shift in tax prices justifies our theoretical assumption that expenditure no greater than Amax will be made through a single A election. 5 The property tax rebate system ended after the1983-84 school year. For 1980-81 and 1981-82, districts had to hold separate votes for A and B. A district could hold no election and operate at the reversion level, hold only an A election, or hold A and 8 B elections simultaneously. For 1982-83 and 1983-84, the school board could alternatively hold a “Combination” election in which there was a single proposal that included both A and B amounts. The board could, if it so chose, also hold separate A and B elections as in the two earlier school years. Elections were held in the calendar year of the start of the school year. Elections were held as early as the last week in March for the school year that would begin in September. Summary statistics on elections are provided in Table 1. In the table, we place all school districts in one of five mutually exclusive categories. 1. School districts that had at least one simultaneous A and B election. Some of these districts had multiple such elections in a given school year. Others could have later held an A election only after defeat of both an A and B proposal. They might also have held a B election only if an A proposal had passed and a B proposal failed. In the last two years, some districts in the first category could have held combination ballots that failed and then went to an A, B setup. 2. Districts that held only A elections. These were districts where the entire proposed operating budget qualified for the property tax rebate. 3. Districts that held only B elections. In these districts, few in number, any increase beyond Q would not have qualified for the rebate program. 4. Districts that held only combination elections. Combination elections were permitted only for the last two school years of this period. 5. Districts not holding elections. The board in these districts chose to operate with Q (or less). In each of the four years, Oregon had just over 300 school districts. 9 Table 2 provides additional details for districts holding A and B, and/or Combination elections. The categories in this case are no longer mutually exclusive. The first row, carried over from Table 1, shows the total number of school districts holding at least one election with both A and B ballots. The second row shows the number of districts that held at least one Combination election. (Some of these districts also held simultaneous A and B elections at a later date, so the numbers in row 2 differ from the numbers in row 4 of Table 1.) The third row shows the total number of instances where A and B votes were held on the same day. The numbers in this row exceed those in the first row because of multiple elections held by some school districts. Some districts held two or three simultaneous A and B elections in the same year. 6 School districts were able to split the B vote into two (typical) or more (rare) votes, presumably to offer a menu of add-ons, such as kindergarten. These votes are not covered by our formal model. Therefore in the fourth row of Table 2 we show the number of observations with an A election and a single B election on the same day. These elections constitute our population of interest. As a comparison of the third and fourth rows indicates, multiple B elections were infrequent. Our main results are presented in the remaining rows of the table. Results Proposition 1 We focus on differences in the percentage voting “No” on the B and A votes because there are variations in the number of voters casting A ballots and B ballots. These variations are relatively small. We report on abstention in a later section. The fifth row in Table 1 shows the number of observations with an increase in the percentage voting “No” on the B as compared to the A ballot. The sixth row reports observations with an increase in the subset of observations with a single B election. No change in the votes against is also consistent with the model. This 10 could arise if there were no voters of preference type AQZ or AZQ. There are one or two observations of this type in each of the first three years. There are none in the fourth year. The seventh and eighth rows report the number of cases in which there was no change in the votes against. The errors to the prediction of Proposition 1 are shown in rows nine and ten. If we focus on districts with single B elections, there was only one violation of Proposition 1 in 1980. The district with a decrease in the “No” vote had a decrease of 1.64 percentage points, quite small in comparison to the average increase of 9.74 percentage points in that year. There was also a single, somewhat uncertain, violation in 1981. 7 There were no violations in either 1982 or 1983. In summary, the evidence in support of Proposition 1 is overwhelming. In a total of 295 observations, there were only two violations, with one of the two being suspect.8 It is delightful to find a result where reporting a statistical test would be absurd. In those districts where there was an increase in the percent No, the increase was substantial, as reported in the eleventh row. The average increase ranged from 6.75 percentage points in the 1981 elections to 9.74 in the 1980 elections Proposition 2 Proposition 2 predicts that school boards should prefer to conduct Combination ballot elections rather than separate A and B elections. Thus, when the state permitted Combination ballots, there should be a substantial shift away from A and B elections to Combination ballots. The shift is supported by the data. Table 1 shows that the number of school districts holding A and B votes decreased from 116 in 1981 (and 77 in 1980) to only 23 in 1982, before increasing back to 35 in 1983. Moreover, of the 23 districts that held A and B votes in 1982, 12 had previously tried a Combination ballot in the same year; in 1983, 27 of 35 districts tried a 11 Combination ballot first. Districts holding at least one combination ballot numbered 77 in 1982 and 78 in 1983. Proposition 2 is not supported as strongly as Proposition 1. A few districts continued to go directly to A and B ballots. Somewhat more reverted to A and B after Combination ballots failed. Romer and Rosenthal (1979) argue that, were Combination ballots available, budgetmaximizing agenda setters would make a relatively high proposal initially and then cut it after a failure. Perhaps switching from Combination to A and B represents a concession expected by voters in the new institutional framework established by the property tax rebate system. 9 In any event, even though Combination ballots were not adopted uniformly, there was a strong tendency to substitute Combination ballots for A and B ballots. This tendency was especially strong on the first annual election in the district. Proposition 2 is consistent with the drop in A and B elections between the first two years and the second two years. Abstention when A and B proposals are both on the ballot As we have noted, voters did not always vote on both the A proposal and the B proposal when these were both on the ballot. Although the differences in votes cast in the two elections are small, the differences are systematic. The differences do not appear to have an explanation in the purely random turnout model of Romer and Rosenthal (1979). The voters have taken the trouble to vote; they should be no more likely to “forget” to vote on one of the two proposals than the other or to make a “mistake” and omit voting. Consequently, differences in turnout should not be systematic. The number voting on the A amount should exceed those voting on the B amount about half the time, and vice versa. The differences would also seem unrelated to the strategic turnout models based on 12 the cost of voting (Palfrey and Rosenthal, 1985). The voters, by getting a ballot, have paid the cost of voting. What might be relevant to the differences is informational uncertainty about preferences (Feddersen and Pesendorfer, 1997, 1999). A voter might not be sufficiently informed, say, to know whether he has preference ZAQ or AZQ. This voter knows he should vote for A, but is unsure about whether to vote for B. Such a voter might abstain strategically and leave the decision to informed voters. Alternatively, the voter might be close to indifferent between A and Z. In such a case, the voter might abstain due to indifference (Hinich and Ordeshook, 1969). What we have in mind here is not “I’m close to indifferent, so I do not want to pay the cost of voting”, as in Riker and Ordeshook’s seminal 1968 paper. It is more “I’m close to indifferent, and I can’t make up my mind, so I’ll just pass on this one”. A strong empirical relationship with regard to votes cast is shown in Table 3. In all four years, the number of districts where there are fewer votes on the B proposal than on the A far exceeds the number of districts where the opposite is true. The last row of the Table presents a simple binomial test of the null hypothesis that votes in the B election are fewer than those in A elections with probability 0.5. That is, fewer votes is a “success” and more votes or no change is a “failure.” The test is run for single B election districts only. For each year, the null is rejected with p-values lower than 0.001. Conclusion We have presented evidence that voters, at least in the aggregate, respond rationally when presented with a budget proposal and an “add on” second proposal. Under the assumption that voters have single peaked-preferences over spending, we should systematically find more votes against the second proposal than against the first. This proposition is verified. Another 13 observation pointed to systematic behavior by budget-maximizing agenda setters. When given the option of not having two proposals and proposing just a single budget, the setters should opt for the single proposal. Setters did opt for a single proposal, with relatively few exceptions. Voters do vary in their participation when voting on two proposals. Second proposals typically have fewer voters than first proposals, at high levels of statistical significance. The results in this paper are in the spirit of Romer and Rosenthal (1979), who presented a theoretical model of budget-maximizing agenda setting and then presented qualitative evidence that was consistent with the model. Their model also made quantitative predictions about the amounts of the proposals and about the percentage of voters that should support a proposal. These predictions were then tested in a series of later papers (Romer and Rosenthal, 1982a, 1982b; Filimon, Romer, and Rosenthal, 1982). The two-proposal scenario will similarly generate a set of quantitative predictions about budget proposals and voting. These predictions will have to account for the endogeneity of tax price (see Romer, Rosenthal, and Munley, 1992) induced by the A and B setup. The various predictions generated by the A and B setup can be pursued in further research. 14 Table 1. Summary Statistics, Oregon School Referenda 1980-83 Election Law Election Law A, B Mandatory Combination O.K. Districts with: 1980 1981 1982 1983 1. At least one A, B election 77 116 23 35 2. A election only 133 59 80 58 3. B election only 4 2 5 20 4. Combination only n.a. n.a. 65 53 5. No election 93 129 134 144 307 306 307 310 Total School Districts 15 Table 2: Oregon school budget elections by type, 1980 - 1983 1980 1981 1982 1983 1. Unique districts holding A & B elections 77 116 23 35 2. Unique districts holding Combination elections n.a. n.a. 77 78 3. Observations with A & B votes on same day. 92 163 32 39 4. Observations with A & single B election on same day 84 159 23 39 5. Observations with increase in %No vote 87 161 31 39 6. Observations with increase in %No, A & single B election 81 157 22 39 7. Observations with no change in %No vote 2 1 1 0 8. Observations with no change in %No, A & single B election 2 1 1 0 9. Observations with decrease in % No vote 3 1 0 0 10. Observations with decrease in % No, A & single B election. 1 1 0 0 9.74 6.75 9.54 7.06 11. Average increase in %No vote if increase > 0 Notes to Table 2: 1. Lake Oswego 7J in Clackamas County was the only observation in 1980 that had a single B election and had a percentage decrease in the percentage No. The decrease was 1.64%. 2. The two observations with no change in the No vote in 1980 were both very small districts. Ash Valley 125 in Douglas County had a total of 52 voters and passed both ballots by a vote of 43-9. Pine Creek in Harney County had a total of 17 voters and passed both ballots by a vote of 16-1. They both would appear to be homogeneous rural communities where the setter model is inappropriate. 3. Tygh Valley 40 in Wasco County is the only observation in 1981 that had a percentage decrease in the percentage no. The decrease was 8.81%. This observation may not be a model error. The B election was reported as an A election in the official document. It may have been that the A amounts were broken up into a menu of two pieces. In this case, the model does not apply. 4. Carlton 11 in Yamhill County is the only observation in 1981 with no change in the No vote. Both A and B passed by a margin of 187-157. In this case, the A amount was very small, only $11,213, while the more important amount, $126,291, was in the B election. 5. The number of simultaneous A, B elections drops in 1982 as a result of the use of Combination ballots. In fact, of the 23 school districts that held at least one simultaneous A, B election in 1982, 12 had first tried a Combination ballot. 6. The single observation in 1982 with no change in the No vote between the A and B ballots was Alsea 7J in Benton country. The vote was 148-277 on both ballots. The A ballot was for $519,268 as against only $25,629 on the B ballot. 16 Table 3: Oregon school budget elections by type, 1980 - 1983 1980 1981 1982 1983 1. Unique districts holding A & B votes 77 116 23 35 2. Observations with A & B votes on same day. 92 163 32 39 3. Observations with #voting on B < #voting on A 64 103 24 39 4. Observations with #voting on B = #voting on A 7 8 2 1 5. Observations with #voting on B > #voting on A 21 52 6 8 6. Observations with A & single B election on same day 84 159 23 39 7. Observations with #voting on B < #voting on A 61 103 20 30 8. Observations with #voting on B = #voting on A 7 8 1 1 9. Observations with #voting on B > #voting on A 16 48 3 8 0.00009 0.00016 0.00077 0.00053 Districts with A and single B p-level, Null: Prob(increase) = 0.5, single B elections 17 Figure 1. Best Responses of the Voter Types (Single-Peaked Preferences) Q A Z (Low demand) When Pivotal Pivotal for A Pivotal for B Non-pivotal outcome B wins B loses A wins A loses Pivotal for both Best Response for Agent Against A Against A Against B Against B Against both A and B Z A Q (High demand) When Pivotal Pivotal for A Pivotal for B Non-pivotal outcome B wins B loses A wins A loses Pivotal for both Best Response for Agent For A For A For B For B For A and B A Z Q (Intermediate) When Pivotal Pivotal for A Pivotal for B Non-pivotal outcome B wins B loses A wins A loses Pivotal for both Best Response for Agent For A For A Against B Against B For A, Against B A Q Z (Intermediate) When Pivotal Pivotal for A Pivotal for B Pivotal for both 18 Non-pivotal outcome B wins B loses A wins A loses Best Response for Agent Against A For A Against B Against B For A, Against B Figure 2. Best Responses of a Voter with Non-Single-Peaked Preference Z Q A When Pivotal Pivotal for A Pivotal for B Pivotal for both 19 Non-pivotal outcome B wins B loses A wins A loses Best Response for Agent For A Against A For B For B For A, For B References Feddersen, Timothy, and Wolfgang Pesendorfer. 1997. Voting Behavior and Information Aggregation in Elections with Private Information. Econometrica 65: 1029-1058. Feddersen, Timothy, and Wolfgang Pesendorfer. 1999. Abstention in Elections with Asymmetric Information and Diverse Preferences. American Political Science Review 93: 381-398. Filimon, Radu, Thomas Romer, and Howard Rosenthal. 1982. Asymmetric Information and Agenda Control: The Bases of Monopoly Power in Public Spending. Journal of Public Economics 17: 51-70. Fudenberg, Drew, and Jean Tirole. 1991. Game Theory. Cambridge, Mass.: MIT Press. Gramlich, Edward, and Daniel Rubinfeld. 1982. Micro Estimates of Public Spending: Demand Functions and Tests of the Tiebout and Median-voter Hypotheses. Journal of Political Economy 90: 536-60. Hartman, William T., and C.S. Hwang. 1985. Effectiveness of Property Tax Relief in Oregon. Journal of Education Finance 10: 339-359 Hinich, Melvin J., and Peter C. Ordeshook. 1969. Abstention and Equilibrium in the Electoral Process. Public Choice 7: 81-106. Mebane, Walter R., Jr. 2000. Coordination, Moderation, and Institutional Balancing in American Presidential and House Elections. American Political Science Review 94: 37-57 Niskanen, William. 1971. Bureaucracy and Representative Government. Chicago: AldineAtherton. 20 Palfrey, Thomas R., and Howard Rosenthal. 1985. Voter Participation and Strategic Uncertainty. American Political Science Review 79: 62-78. Riker, William, and Peter C. Ordeshook. 1968. A Theory of the Calculus of Voting. American Political Science Review 62: 25–42. Romer, Thomas, and Howard Rosenthal. 1978. Political Resource Allocation, Controlled Agendas, and the Status Quo. Public Choice 33: 27-43. Romer, Thomas, and Howard Rosenthal. 1979. Bureaucrats vs. Voters. Quarterly Journal of Economics 93:563-87. Romer, Thomas, and Howard Rosenthal. 1982a. Median Voters or Budget Maximizers: Evidence from School Expenditure Referenda. Economic Inquiry 20: 556-578. Romer, Thomas, and Howard Rosenthal. 1982b. An Exploration in the Politics and Economics of Local Public Services. In Public Production, ed. Dieter Bös, Richard A. Musgrave, and Jack Wiseman. New York: Springer, 105-126. Romer, Thomas, Howard Rosenthal, and Vincent Munley. 1992. Economic Incentives and Political Institutions: Spending and Voting in School Budget Referenda. Journal of Public Economics 49: 1-33. Rosenthal, Howard. 1990. The Setter Model. In Advances in the Spatial Theory of Voting, ed. James Enelow and Melvin Hinich. New York: Cambridge University Press. Stiglitz, Joseph E. 1974. The Demand for Education in Public and Private School Systems. Journal of Public Economics 3: 349-395. 21 Endnotes 1 All data on elections come from Oregon Department of Education, “1980 Summary of School District Financial Elections”, Jan. 1981; “1981 Summary of School District Financial Elections”, Dec. 1981; “1982 Summary of School District Financial Elections”, Mar. 1983; “1983 Summary of School District Financial Elections”, May 1983. Copies were obtained from the Oregon State Library. A pdf that contains the documents is available through Google Docs at: https://docs.google.com/open?id=0B18lDjY36mjqZjY2NWZhNDUtNTIxOS00OWNlLWJlZTgt N2JkMWRiOWRlYzIy 2 Our formal model is one of complete information. We therefore ignore the strategic implications of the fact that, in practice, the setter could hold additional elections if a given A or B proposal failed. Strategic behavior by the setter was explored theoretically in Romer and Rosenthal (1979). Strategic behavior by the voters, discussed in Rosenthal (1990) is an open research question. 3 The concept of iterated weak dominance is presented in Fudenberg and Tirole (1991), chapter 11. 4 The Oregon Education Association’s summary states, “The Property Tax Relief Program limited growth of assessed value of all property in the state to no more than five percent per year and provided that the state General Fund pay up to 30% of local residential property taxes – capped at $800 per low-income homeowner and $400 per low-income renter.” (http://www.oregoned.org/site/pp.asp?c=9dKKKYMDH&b=4955821, accessed Oct. 14, 2011.) The $800 amount for 1980 was reduced to $425 in 1981, $192 in 1982, $170 in 1983 and 1984, and $100 in 1985 before the program was ended. No A, B elections took place for the 1984-85 and 1985-86 school years. The description of the dollar amounts is contained in “Appendix B: A 22 Recent History of Oregon Property Taxation,” downloadable at http://www.oregon.gov/DOR/STATS/303-405-01-toc.shtml, accessed Oct. 15, 2011 5 It is always true that for any expenditure less than or equal to Amax, taxes for a strictly A expenditure are weakly less for all voters than taxes for a mixture of A and B expenditure of the same amount. Hence, raising an amount no greater than Amax through an A election is weakly Pareto superior for both voters and the setter compared to raising the same amount through a mixture of A and B. 6 There was one case, Prairie City in Grant County, where four A and B elections were held from May through November 1981. Both A and B ballots failed all four times. The A ballot was for $31,600. The B ballot was cut from an initial $402,167 on the first try to $68,300 on the last two. A single A ballot election had failed in 1980. Prairie City was unable to pass a school budget until September 1982. 7 The violation occurred in Tygh Valley in Wasco County with a decrease of 8.81 percent. The observation may not have been a violation. The official election document reported that there were two A elections rather than an A and a B. We have conservatively treated the second election as a misreported B election. It may have been, though, that the district split the A ballot into two pieces. If that were the case, Proposition 1 would not apply. 8 Of course, not all the observations are independent, since some school districts appear in more than one year and others hold more than one A and B elections. Gervais UH1 also proposed two A ballots in 1982. 9 Prairie City (see note 6) made such a concession. After failing a Combination ballot in May, it succeeded with A and B ballots on September 21, 1982. The total budget was the same in both May and September. Romer and Rosenthal (1979) report that budgets not infrequently had zero 23 or very small cuts in later elections. The result is consistent with work by Gramlich and Rubinfeld (1982) who report that participation changes as the impact of losing the custodial value of the schools becomes apparent to parents. In this case, the May election took place on a general primary ballot. The September election was a special school election. 24