Relaxing electoral constraints in local education funding

Abstract

If voters do not like federal tax policy, they can elect new representatives. At the local level, though, voters can directly deny the tax increases their elected officials propose. All but three states have a limit on either the taxing or spending abilities of local governments and, most commonly, state laws require that increases in local taxes receive a public referendum before they are enacted (Mullins, 2010). In the November 2020 election, voters nationwide decided on the fate of $52.7 billion of proposed funding, down from $70 billion 4 years prior (Pierog, 2020).

Requirements that budgeting questions be put to a vote reduce the level of spending (Feld & Matsusaka, 2003; Funk & Gathmann, 2011). However, the rules that govern these votes vary in myriad ways. States differ in the types of taxes or spending the rules cover: Some only allow proposals for capital spending, while others also include current expenditure spending. States also often limit the total amount of tax revenue or the tax rate governments can propose, and vary in whether they adjust for inflation, for changes in population, or for growth in the property tax base. Certain election rules limit the timing of when governments can put proposals on the ballot, since local governments may use this timing flexibility strategically (Anzia, 2011; Kogan et al., 2018; Meredith, 2009). States also differ in the vote share required to approve the proposals, with many proposals requiring more than a simple majority. There is little empirical research, however, documenting how these differences in voting requirements may affect government budgets, local spending, and the provision of public goods and services. This is a significant gap in the literature given the millions of dollars in funding at stake in each local referendum and the billions at stake nationwide.

In this paper, we study a proposition in California that weakened the constraints on some local governments by lowering the vote share required to approve capital funding for schools and community colleges. We use a difference-in-differences design around this policy change and data for over 4,000 local elections across the state over 2 decades. We estimate the effects of this policy change on the proposals made by affected districts, their outcomes at the ballot box, and on the eventual funding outcomes.

We develop a theoretical model of the interaction between a school board and voters, building on the literature in local political economy (Barseghyan & Coate, 2014; Coate & Ma, 2017; Romer & Rosenthal, 1982). In our model, the school board makes a tax proposal that the voter can accept or reject. Thus, the school board has “agenda-setting” power to extract policies closer to its preferences than those of the voter. However, uncertainty in how residents will vote hinders the exercise of this agenda-setting power. We use this model to show how a change in the required vote share affects the size of the proposals and the vote outcomes, and to show how these effects depend on the divergence in preferences between voters and elected officials.

We study how the policy change affected the funding outcomes of local governments. We find that Proposition 39 led to an additional $57 per resident in approved local government bonds, a more than 100% increase. This is a large impact, and one that affects governments unequally. There were larger effects on funding outcomes in more racially and ethnically diverse districts. Similarly, we find larger effects in jurisdictions with moderate levels of poverty, moderate White population shares, and fewer older residents. These results shed light on the characteristics of places where funding is most constrained by referendum requirements. Our model shows that the extent of a community's responsiveness to a change in the vote share requirement depends on the extent of disagreement between voters and their elected representatives, as well as the level of uncertainty in referenda outcomes that the elected officials face. Thus, the empirical results show what types of communities have larger disagreements or more uncertain election outcomes.

There are two channels through which these effects could occur: the behavior of elected officials when proposing funding and of voters when voting on the proposals. We find that the policy change made treated districts no more likely to propose a bond relative to other jurisdictions. However, the size of the bond proposals increased substantially, by $48 per resident, or a 59% increase. Naturally, the larger bond proposals may result in lower vote shares. Conditional on proposing a bond, we find that the policy change resulted in bond proposals receiving a lower percentage of votes in favor. However, this decline was smaller than the 11.7 percentage point decrease in the vote share requirement. In other words, school boards and community college districts experienced declines in support for their proposals that were smaller than the full amount of the policy change. Thus, we observe no change in the probability that affected districts approved any kind of new funding, and large increases in the probability that they approved new bonds. Even though these districts submitted larger proposals that had less support from the electorate, they resulted in increases in the probability of success. These results—a positive effect on proposal size and a negative effect on vote shares—are consistent with the politician preferring higher spending than the voters, when interpreted through the lens of our model.

Our paper contributes to several strands of research. First, we contribute to empirical evidence on the fiscal and policy effects of tax limits. This literature builds on the work of Romer and Rosenthal (1978, 1979, 1982), which modeled the constraint on government in contexts where government officials have agenda setting power. Balsdon et al. (2003) applied the logic of these agenda-setting models to proposals of and voting on school bonds. The authors estimated a structural model of school boards and voters and found that school boards favor a higher level of spending than voters but are risk averse in their proposals. We extend this line of research by leveraging a policy change, and our finding that relaxing the voting threshold results in larger proposals is consistent with their result.

We also contribute to an empirical literature on the effects of tax limits on the quality of public services (Dye & McGuire, 1997; Figlio & Rueben, 2001; Poterba & Rueben, 1995; Rose, 2010). These papers study the effect of the existence of local tax limits. By contrast, in this paper we study a change in a particular characteristic of the limit itself. Romer et al. (1992) shared that aim and considered the effects of matching aid and supermajority requirements in the context of their structural model. However, their data did not contain any heterogeneity in the vote requirements. We show that the vote share required to pass public spending has large effects on the amount of spending both proposed and implemented.

While this paper is the first to our knowledge to study the effects of the change in the voting threshold, two papers have studied voting on the constitutional amendment that resulted in this policy change. Brunner and Ross (2010) developed a model of voter support for supermajority rules and showed that voter support for the supermajority rules is related to the income distribution in their school district. Balsdon et al. (2005) showed that voters in metropolitan areas with a more fragmented set of school districts were more likely to vote for the reduction in the threshold. An existing literature has studied supermajority voting rules in the context of legislative voting (Crain & Miller, 1989; Messner & Polborn, 2004). Most related to our work is Knight (2000), which instrumented for supermajority requirements with the ease of amending state constitutions and found that supermajority requirements lead to lower taxes. This paper complements these analyses by examining the effect of a change in the severity of a supermajority requirement in the context of public referenda.

The environment of voting on school financing and the effect of the resulting funding has received significant study. Cerdán and Rueben (2003) detailed the history of school funding referenda in California. While we study the process of the school votes themselves, Cellini et al. (2010) studied the effects of the resulting capital investments on house prices and Cellini (2009) examined their effects on higher education. These papers are part of a broader literature that studies the effects of school construction; the evidence on its effect on student outcomes is mixed. Large school construction projects may lead to improved student outcomes (Aaronson & Mazumder, 2011; Conlin & Thompson, 2017; Duflo, 2001; Neilson & Zimmerman, 2014). However, studies using regression discontinuity approaches to study school capital improvement bond referenda, like those we study, have found limited evidence of positive effects on student outcomes (Cellini et al., 2010; Choi, 2019; Martorell et al., 2016). An exception is Hong and Zimmer (2016), which found positive long-run effects on test scores in Michigan. Biasi et al. (2024) found generally positive effects of capital improvements on test scores, and Jackson and Mackevicius (2024) found generally positive effects of spending on student outcomes, as have studies that focused on particular targets of spending like air conditioning (Park et al., 2020). These studies have consistently found positive, sometimes large, effects on house prices, suggesting that households care a great deal about school facilities and school quality.

Lastly, we contribute to a literature on the support for public goods in diverse communities. We show how the effects of the loosened electoral constraint differ by the racial makeup of the jurisdiction. A broad literature has found that diversity is related to decreased support for government (Alesina et al., 1999; Dahlberg et al., 2012). However, some of these findings are not robust to alternative specifications (Boustan et al., 2013). Closely related to our work is Rugh and Trounstine (2011), which showed that more diverse cities propose fewer, larger municipal bonds than less diverse cities but end up authorizing similar levels of debt. The strategic proposals they documented could drive the differential response to the policy change we study. Recent work has shown that increased diversity among local elected officials results in less spending on public goods (Beach & Jones, 2017). Our results suggest that in more diverse places, elected officials would prefer a higher level of spending than the voters.

The remainder of the paper is organized as follows. In the following section, we provide further detail about local government funding and the policy change we study; in “Conceptual Framework” we describe our model; in “Data” we describe the data we use; in “Empirical Approach” we explain the empirical method; in “Results” we describe our results; and in “Conclusion” we conclude.

Our study focuses on funding for capital investments in local governments in California. Local governments with the power to tax include cities, school districts, community college districts, counties, and special districts that provide a particular service, such as airports, parks, water, and transit. In this section, we outline the options for these governments in generating revenue for investments.

Before describing the quantitative estimates of the effect of Proposition 39, we turn to a disagreement voting model to explore the implications of the policy change. The model shows how a change in the vote share required to pass a funding referendum can change the funding amounts that local politicians will propose, the probability that those referenda will prevail, and the resulting funding. In this section, we outline the model and describe the main results. Derivations and further details of the model are in Appendix Section A2.

Our model builds on the framework developed by Romer and Rosenthal (1979, 1978, 1982), who studied the interaction between a bureaucrat with agenda setting powers and voters who must approve their decisions in referenda. We extend this model by studying the role of the voting thresholds that referenda are required to achieve. In this, our model is similar to Brunner and Ross (2010), who studied voting behavior on Proposition 39 itself.10

Our model considers the interaction between voters and a representative politician in setting a local government's expenditure on a public good. There is a status quo level of funding, denoted g. The politician has the option to make a proposal for additional spending on the public good. If the politician makes no proposal or if the proposal fails to receive the required support among voters, the status quo level of funding is implemented.

The voters desire a certain level of funding, denoted θ. The voter utility reflects their opinion on the value of the public good and the cost of the taxes that would be necessary to fund it. We assume voter utility declines symmetrically as the funding level diverges from the ideal point, with quadratic distance policy preferences given by −(g − θ)².

We assume the politician's ideal funding is at least as much as what the voters desire. The politician's ideal level is θ + b, where b represents the disagreement between the politician and the voters. We assume that b is non-negative, which is consistent with the common justification of referendum requirements, that limiting the politician's authority will lower spending. When b is equal to zero there is no disagreement. As with the voters, the politician has preferences, −(g − θ − b)^2, that are symmetric in the distance of the funding level from their ideal level of funding.

The sequence of events in the model is as follows. First, the politician chooses whether to propose a referendum to adopt a public good level, g’, rather than the reversion level, g. Should they decide against proposing a referendum, the reversion level is adopted. Conversely, if they propose a referendum, the voters then vote yes or no. If at least v share of voters vote in favor, then g’ is adopted, where v is the vote share requirement, the focus of this paper. Otherwise, the status quo g is adopted. We assume that v is at least one half. Further, we assume that the politician's preferred level of the public good, θ + b, exceeds g. In order to introduce uncertainty in the outcomes of proposals, and hence allow the model to rationalize failed proposals, we assume that there are shocks to the voter preferences that the politician does not know at the time they make proposals. The probability that proposal g’ will prevail given the vote share required is p(g’; v). Since politicians are uncertain whether any proposal they make would pass, they will choose their proposal so as to maximize their expected utility over the possible outcomes.

The policy change we study empirically is equivalent to a change in the vote requirement, v. We are interested in how changing v alters the proposed level of investment (g’ − g), its likelihood of success (p(g’; v)), and the resulting public good levels, which we describe as the expected level of investment, p(g’;v)(g’ − g).11

Table 1 describes the effect of a decrease in the vote requirement on each outcome when there is and is not disagreement between voters and the politician on the optimal level of provision. When the politician and voters prefer the same level of spending, a decrease in the threshold does not change the proposal or the fraction of votes in favor. Nevertheless, the probability of the proposal prevailing will increase and, as a result, the expected level of investment will also increase. When the politician prefers higher spending than the voter, a decrease in the threshold will result in larger proposals and, consequently, lower vote shares in favor. However, since the threshold is lower, the probability that the proposal prevails increases, leading to an increase in investment.

In sum, the model presents a framework for understanding the effect of a change in the vote requirement on the behavior of local politicians and voters. It shows how the policy change that we study affects the proposals made by politicians and how voters will vote on them. It has clear predictions for the importance of disagreement between voters and elected politicians.

We combine various sources of administrative and publicly available data on all public schools in California over the past 2 decades. Our main source of data is the set of all election results for all local measures in California between 1995 and 2016. These data, similar to what Cellini et al. (2010) and others have used, come from the California Election Data Archive (CEDA), a project of the Center for California Studies at California State University, Sacramento. We include elections from counties, municipalities, community college districts, and K–12 school districts. Our set of measures includes all those that would have authorized new, increased, or renewed taxes. Our main estimates are limited to only general obligation bonds, but we include other funding measures in robustness checks.12

For each measure in the CEDA dataset, we observe the full text of the ballot question, which includes the proposed dollar amounts for general obligation bonds. We also observe whether the measure passed and the number of votes for and against, from which we calculate the share of voters who voted for passage. During the period we study there were 4,520 tax-related measures. There are 10 different types of measures included among these: GO bonds, other bonds, business taxes, overrides of the Gann limit, Mello/Roos bonds, parcel taxes, sales taxes, transient occupancy (hotel) taxes, and utility taxes. Of all the measures, 2,075 (46%) were for GO bonds, which are the main measure we focus on in the primary analyses.

We complement the CEDA dataset with other sources of publicly available information. We use school- and district-level information on student demographics and proficiency on standardized tests from the Common Core of Data. We use Decennial Census information from 2000 for population counts, demographics, and socioeconomic characteristics of each local jurisdiction. Counts from the census are readily available for counties, municipalities, and school districts. Census tabulations are not available for community college districts, however. To produce counts of the number of residents in a community college district, we overlaid their current boundaries, available from the Foundation for California Community Colleges, with a map of census tracts. We then estimated the proportional overlap of tract-level population with the college districts.13

In sum, we create two analysis datasets. The first is a “jurisdiction-level” panel dataset. This panel consists of 1,589 jurisdiction-year observations from 1995 to 2016, comprising 977 K–12 districts, 482 cities, 72 community college districts, and 58 counties. For each observation, we observe the number of relevant elections held and passed, the number of GO bonds proposed and approved, and the amount of GO funding per capita proposed and approved.

The second dataset is an “election-level” dataset with the full set of 4,520 elections between 1995 and 2016. For these, we observe the jurisdiction, purpose, and vote share. Notably, this dataset includes GO bonds as well as the nine other types of measures.

Table 2 shows summary statistics of the jurisdiction-level panel, by jurisdiction type, prior to the passage of Proposition 39. Between 1995 and 2000, almost half of the school districts proposed a GO bond, as did one fifth of community college districts. The other jurisdictions in the sample—cities and counties—were much less likely to propose this type of funding. This makes sense given that these jurisdictions have a wider set of fundraising tools than school and community college districts. On the other hand, cities and counties were much more likely to put other types of funding proposals, primarily changes in parcel taxes, on the ballot. Passage rates of GO bonds and other elections did not vary across jurisdiction type. Education-related GO bonds tended to be much larger than the GO bonds proposed by counties and cities.

In this section, we describe our strategy to empirically investigate the effect of the change in vote requirements on the spending proposals made by the governments, the support the spending proposals received from voters, and the resulting capital spending of local governments. We use a difference-in-differences strategy to identify all effects. However, they require two different types of datasets. We first describe the unconditional approach, which we use to answer the questions regarding government proposals and funding outcomes. We then describe the approach we use to answer questions regarding voting, which conditions on local governments that proposed a bond.

We organize the results into three sections. First, we examine whether Proposition 39 changed the proposal behavior of school boards and community college districts in terms of their likelihood of proposing a GO bond, and the size of the proposals. This uses the methods described in “Estimating Effects on Government Behavior and Funding Outcomes.” Second, we ask whether the performance of GO bonds from educational jurisdictions changed as a result of Proposition 39. This uses the methods described in “Estimating Effects on Election Outcomes.” Third, we study the overall effects of Proposition 39 on funding outcomes, which combines the effect on government behavior and the effect on voting outcomes. This again uses the methods described in “Estimating Effects on Government Behavior and Funding Outcomes.” After discussing the main results we then move to heterogeneity analysis across jurisdiction types, and robustness checks.

In this paper, we analyze how a constitutional change to the voting threshold required for passing a school bond in California affected the share of voters who supported such bonds. We find that schools more than doubled their bond funding per resident due to the change. In addition, we show that governments were no more likely to propose a measure but are more likely to pass a general obligation bond due to the policy change. School bonds saw a drop of 6 percentage points in voter support following a decrease in the voting threshold by 11 percentage points from two thirds to 55%.

We interpret these results in the context of a political economy model of the interaction between a voter and an elected official. With no disagreement between voters and elected officials, a lowering of the vote share would not affect the proposals made. It would, however, increase the probability that they prevail. Yet, we do find an increase in both the proposals made and the probability that they prevail. This suggests that there is disagreement between the voters and the elected officials. The elected officials use their expanded flexibility to both request more spending and ensure that it passes with a higher probability. Together these result in a large increase in the amount of funding approved.

We note that our results should be interpreted with some deference to potential spillover effects. School districts share a common property tax base with the overlapping jurisdictions that were not subject to the policy change. Thus, changes in the tax rate imposed by school districts could affect voting behavior in these overlapping jurisdictions. For example, additional capital funding in a school district might make it less likely that voters would support higher tax rates for the city or county in which they live. If that were the case, the estimated treatment effects presented in the paper are the combination of these two effects, not simply the effect on the school districts.

Policies that limit the unilateral power of local elected officials and require tax policies to be approved via referendum are common throughout the United States. We study an incremental change in this requirement and show that it had large effects. Lowering the vote threshold worked as intended and more than doubled funding from GO bonds. However, the policy change did not expand the set of school districts that proposed a GO bond. In that sense, Proposition 39 likely increased funding among districts that would have proposed bonds even without the change. This occurred both because districts whose proposals may have failed saw them succeed and because districts proposed larger bonds.

The policy change did not have equal effects across all jurisdictions. Its effects were concentrated in jurisdictions that were not racially and ethnically homogenous. This suggests that relaxing or removing constraints on local elected leaders increases spending more in places with more diverse populations. Our analysis points to two possible reasons for why this may be. Elected officials in those places may have larger disagreements with referendum voters over their preferred level of spending or have less uncertainty over referendum outcomes.

These results have implications for policy discussions that involve changes to local tax limits. Some states frequently modify the terms of these limits. States change the latitude with which local officials can set policy without a voter referendum, in some cases through annual adjustments to allowable budget increases. States also change the rules for referenda, as in the case studied. Our results suggest that the details of how these policies are designed matter. It can affect the amount of funding that occurs, and hence taxes that residents must pay. Since it deferentially affects jurisdictions with different characteristics, it has important distributional consequences.