Direct versus Representative Democracy: Reassessing the Evidence

Christopher R. Berry The University of Chicago Harris School of Public Policy [email protected]

This draft: December 2015

Abstract. One of the most robust empirical findings about direct democracy is that US states with the voter initiative tax and spend significantly less than states without the initiative, at least since the 1960s. The relationship between initiative status and fiscal policy has been interpreted as causal and as an indication that voters prefer smaller government than politicians. Yet, existing research is based on cross-sectional comparisons of states with and without the initiative observed decades after the institution was adopted. This paper makes two contributions. First, I establish that the fiscal differential between initiative and non-initiative states is concentrated in education funding. Next, drawing upon a newly created data set that traces state education funding from the late 1800s to today, I explore the origins of these fiscal differentials. I find that some of the differences between would-be initiative and non-initiative states are evident even before the initiative existed. Others emerged only many decades after the initiative had been adopted. Accounting for these historical differences causes contemporary estimates of the initiative effect to disappear. Tracing the causal pathway from the voter initiative to contemporary policy outcomes—if indeed there is one—is therefore a major challenge for scholarship in this area. I discuss implications of these results for contemporary scholarship on direct democracy.

1. Introduction Beginning in the mid-1990s, there has been a resurgent interest in the politics of the voter initiative and referendum. This burgeoning scholarship examines direct democracy as a window into more general aspects of politics. For example, scholars have looked at direct legislation to shed light on interest group influence, legislative responsiveness, voter competence, and the economic theory of government.1 Based on a comparison of policy outcomes between U.S. states with and without the voter initiative, a leading theme in this literature is that the availability of a direct legislation option tends move social (Gerber 1999) and economic (Matsusaka 1995) policies toward the preferred position of the median voter. In other words, representative democracy allows for slack between the will of the people and the policies enacted by politicians, which can be reduced or eliminated through the exercise of direct democracy. If it is true that policy differentials between initiative and non-initiative states reflect the will of the median voter, then examining differences in fiscal policy between states with and without direct democracy should provide insight into differences between voters and politicians over basic questions about the functions government ought to perform in society, how these functions ought to be prioritized, and which level of government ought to perform them. A major limitation of the existing literature, however, is that most of the states that were to adopt the voter initiative did so between 1900 and 1920. Yet nearly all prior studies have been based on cross-sectional comparisons between initiative and non-initiative states in the post-1970 period, a time long removed from the original adoption of the institutions of direct democracy.

1

Reviews of the literature include Garrett (2008), Matsusaka (2005), and Smith and Tolbert (2007). Also see Donovan and Bowler (1998).

1

In this paper, I exploit a new data set on education finance in the U.S. from the late 1880s until today, which allows me to analyze spending by state and local governments both before and after the adoption of the voter initiative. I find that wouldbe initiative and non-initiative states differed in their fiscal policies before any state had ever adopted the initiative. In particular, initiative states had lower state funding of education even before the initiative existed, and that difference has persisted over time. Local education funding, by contrast, was initially higher in initiative states and only became lower, relative to non-initiative states, late in the 20th century, decades after the initiative had been put into place. These findings suggest that the causal pathway connecting the voter initiative to state fiscal policy—if in fact there is one—is far more complex than existing theories would suggest. Attributing contemporary policy differences between initiative and non-initiative states directly to the institution itself, as most prior literature on the subject has done, is at best an incomplete explanation. The paper proceeds as follows. In the next section, I review prior literature and characterize the major theories and empirical results. In section 3, I reproduce the familiar cross-sectional finding that initiative states tax and spend less than non-initiative states today, and subsequently decompose this differential into its component parts, revealing that most of the difference comes from education finance. Section 4 introduces a new data set on historical education finance in the U.S. and uses it to analyze changes after the adoption of the voter initiative, in the short- and long-run. Section 5 discuses the implications of the historical analysis for contemporary scholarship on direct democracy.

2

2. Background Direct democracy is lawmaking by voters. Although the United States does not allow direct legislation at the national level, it is widely practiced in state and local politics, in the form of the initiative and referendum. The referendum allows citizens to vote on laws proposed by the legislature; the initiative allows citizens to propose laws directly, and to have their proposals voted on by the electorate. The initiative is considered to be the purest form of direct democracy (Gerber 1999) in that it allows laws to be made entirely by citizens without the intermediation of the legislature. The standard form of the initiative is simple: a citizen, or interest group, wishing to propose a new law must collect a specified number of signatures from voters in order to have the proposal placed on the ballot. If the proposition gains the support of a majority of voters, it becomes law. The adoption of direct democracy swept through the United States at the turn of the century, concurrent with Populist and Progressive agitation to make legislatures more responsive to the broader “public interest” (Cronin, 1989).2 Figure 1 shows states with the initiative, and the year in which it was adopted. South Dakota started the trend in 1898, and within 20 years most of the states that were to adopt the initiative had done so. Indeed, only five states have adopted the initiative since 1918 – in order, Alaska, Wyoming, Illinois, Florida, and Mississippi – so that there are now a total of 24 initiative states. While the basic form of the initiative is common across states, the details vary, primarily in the signature requirements, as will be discussed below.

2

See Smith and Fridkin (2008) for a fascinating analysis of the politics of initiative adoption, focusing on the question of why a legislature would willingly delegate power back to citizens.

3

The most robust empirical finding about direct democracy is that states with the voter initiative have significantly lower taxes than states without the initiative, at least since the 1960s. The seminal study of the fiscal effects of the voter initiative is Matsusaka (1995), which showed that budgets in initiative states were roughly 4 percent smaller than in non-initiative states over the period 1960 to 1990. Subsequently, at least ten different studies have concluded that the presence of the voter initiative reduces state taxes and spending, all else equal.3 Closely related to these findings, initiative states are also more likely to enact tax and expenditure limitations and supermajority requirements for tax increases (Bowler and Donovan 1995). In a companion paper, Matsusaka (2000) finds that the voter initiative actually led to increases in public spending in the first half of the 20th century. Studying the period from 1902 to 1942, he finds that local, but not state, government spending was higher in initiative states. This paper contains one of the few difference-in-differences designs in the literature, and the results hold up to the within-state analysis. One interpretation of the different findings for the two time periods (Matsusaka 2004) is that voters wanted more spending than politicians in the early part of the century, but less spending than their politicians in the latter part of the century. The initiative, though moving spending in different directions in each case, consistently moved fiscal policy toward the median voter’s ideal point. There is also mounting evidence that initiative states enact more conservative social policies than non-initiative states, at least in the contemporary period. For example, Gerber (1999) argues that initiative states are more likely to have the death penalty and to

3

See Matsusaka (2004, Appendix 4) for a review. By contrast, Primo (2020) finds that the local initiative has a positive impact on city spending, at least where the signature requirements are not too stringent.

4

require parental notification for abortions. Arceneaux (2002) and Bowler and Donovan (2004) also find that initiative states adopt more restrictive abortion policies, while Hero and Tolbert (1996) find that initiative states are more likely to enact official English laws. Matsusaka (2007) examines laws on seven different social issues and finds that initiative states are about 20 percent more likely than non-initiative states to adopt a conservative policy. Importantly, none of these studies finds that citizens in initiative states are significantly more conservative than citizens in non-initiative states, so the policy differentials do not appear to be the product of ideological differences between the two groups of states. Matsusaka (1995, 2004) suggests three theoretical reasons to expect the voter initiative to produce different policy outcomes from pure representative decision making: (a) vote trading is not possible with direct legislation, meaning that logrolling should be reduced (e.g., Weingast, Shepsle, and Johnsen 1981; Weingast and Marshall 1988), (b) the initiative takes control of the agenda away from bureaucrats and politicians, removing their “setter’s advantage” (Romer and Rosenthal 1979), and (c) information asymmetries between legislators and citizens are avoided under direct voting. Besley and Coate (2008) offer a complementary view, arguing that the initiative improves congruence between citizen preferences and policy outcomes by allowing citizens to “unbundle” specific issues that would otherwise not be salient in general-purpose elections. If any one of these accounts is correct, then one implication is that policies in initiative states more closely reflect the will of the majority than in non-initiative states. That is, legislatures sometimes enact policies that are out of step with the preferences of the median voter, and the presence of the initiative causes policy to be brought back into

5

line more quickly than would be the case otherwise. Evidence supporting this interpretation is provided by Matsusaka (2004, 2007) and Gerber (1999), who show that, in the cases they examine, the policy changes wrought by the initiative are preferred by a majority of voters. Not all observers are so sanguine about the ability of the initiative to promote the popular will, however. To the contrary, many critics argue that the expense of waging an initiative campaign privileges moneyed interest groups rather than ordinary citizens (e.g., Broder 2000, Ellis 2002, Garrett 1999, Sabato, Ernst, and Larson 2001, Schrag 1998). According to this view, the initiative causes policy to be more favorable to interest groups than would be the case if the legislature acted on its own.4 Such concerns have led some scholars to challenge the view that the initiative makes policy more responsive to the will of the majority (Lascher, Hagen, and Rochlin 1996, Camobreco 1998).5 Indeed, scholars even called into question the populist credentials of the initiative-led “tax revolt” of the 1970s and 1980s, during which many states hobbled the government’s ability to raise revenues and expenditures (Smith 1998, 2004). Other scholars suggest that resistance in implementation by the legislature and bureaucracy often render even winning initiatives ineffectual in practice (Gerber et al. 2001). While I will not be able to provide a general answer to the controversy over whether the initiative benefits citizens or special interests, a detailed investigation of the fiscal differentials between initiative and non-initiative states, and their historical origins, will shed new light on the question. In fact, my results suggest that the fiscal differences 4

But see Gerber (1999), who finds that broad citizen interest groups are better able to use the initiative to change policy, while narrow economic interest groups are more successful in blocking initiatives to preserve the status quo. Meanwhile, Boehmke (2005) finds the initiative states have more interest groups and that they are more representative of the general population than in non-initiative states. 5 Matsusaka (2004, appendix 4) has challenged the methodology of these studies.

6

between initiative and non-initiative states were not actually caused by the initiative. In the remainder of the paper, I analyze itemized fiscal data for state and local governments to uncover the precise channels through which initiative-related spending differences are realized. Having done so, I then turn to a long-run historical analysis of the comparative fiscal policies of initiative and non-initiative states. 3. Contemporary Analysis Because historical public finance data from the 1800s are only available in the domain of education, as explained below, it is important to first assess what role, if any, education funding plays in the contemporary spending differential between initiative and non-initiative states. Thus, before proceeding to the historical analysis, I begin by replicating the main results from the prior literature and then decomposing the spending differential into its constituent parts to assess the relevance of education funding. For sake of comparability, I replicate the standard empirical strategies from the prior literature. I pool the observations across states and years and regress fiscal outcomes against an indicator for whether the state has a voter initiative, as well as a standard set of covariates used in prior studies. I emphasize that this approach does not estimate the effects of particular voter initiatives. Rather the models compare average fiscal policy outcomes in states with and without the initiative. This approach captures both the direct effect of the initiative and its indirect effect, or “threat effect,” as in Gerber (1999, chap. 7) and Matsusaka (1995, 2004). The logic of the threat effect is that the mere availability of the initiative leads legislators to adjust policy in anticipation of possible action by

7

voters. Thus, the initiative may have an effect on policy outcomes even if it is rarely, or never, used to pass legislation (Matsusaka and McCarty 2001).6 Because observations for the same state over time are clearly not independent, I cluster the standard errors in all models by state, which accounts for arbitrary forms of serial correlation, as well as heteroskedasticity (Arellano 1987, White 1984). I exclude Alaska, Hawaii, and Wyoming from the analysis.7 All dollar values are reported on a per capita basis, and adjusted to 2000 dollars using the consumer price index. All results are robust to the exclusion of California. Model (1) of Table 1 reproduces Matsusaka’s (1995, 2004) main finding: lower direct general expenditures8 for the combined state and local sector in initiative states during the era 1970 through 2000. The coefficient on the initiative dummy variable indicates that combined state-local expenditures are lower by about $162 per capita in states with the voter initiative. Given average direct general expenditures of $3,880, these point estimates suggest that the overall initiative effect is a reduction of about 4 percent in the state budget, roughly equivalent to Matsusaka’s results. The control variables are standard in the literature and perform as expected.

6

Gerber, Lupia, McCubbins, and Kiewet (2001) study the direct effects of winning initiatives on policy outcomes in California. They argue that legislatures, bureaucracies, and courts can thwart the implementation and enforcement of initiatives, and that most winning initiatives are implemented only partially. Their results suggest that the ultimate policy impact of an initiative’s direct effects vary widely from one initiative to another; however, they do not study the initiative’s threat effect. 7 Alaska is removed because of its extremely high level of mineral wealth relative to population, and Wyoming because it is an extreme observation on large number of variables. Hawaii has the only completely state-run school system. Including Hawaii in the analysis notably affects estimates of the allocation of spending between the state and local sectors. Matsusaka (1995, 2004) excludes Alaska and Wyoming from his analysis, but includes Hawaii. Following Matsusaka (1995), I do not count Illinois as an initiative state for this analysis. Although technically an initiative state, the subject matter of initiatives is restricted from directly addressing fiscal policy. 8 Direct general expenditures include all government expenditures except expenditures to other governments, utility, liquor store, employee retirement or other trust funds.

8

The next to columns of Table 1 decompose expenditures into education and noneducation spending, respectively. Model (2) shows that education spending is lower by $84 per capita in initiative states, while Model (3) shows that all other spending is lower by $78 per capita, although the non-education differential falls short of statistical significance at conventional levels (p = 0.13). In other words, education accounts for most (52%) of the spending differential between initiative and non-initiative states. Moreover, education spending is disproportionately affected by the initiative, in that education only accounts for 37% of direct general expenditures. Further analysis, reported in Appendix A, reveals the specific channel through which education spending is reduced in initiative states. Most of the difference in education spending comes through reduced state aid to local school districts, which is $74 per capita lower in initiative states. This figure represents 18 percent of mean state aid to school districts ($425 per capita). Local districts do not make up for the lost state aid by increasing revenue from their own sources—the initiative state differential is a statistically insignificant $16—resulting in a net decrease in education spending. The proportional differences are similar when the analysis is conducted in per pupil rather than per capita terms (see Appendix A). The preceding analysis yields novel results that paint a coherent picture of the fiscal differences between initiative and non-initiative states. Initiative states spend significantly less in aggregate than non-initiative states, which is well known. New is the finding that most of the difference in spending between initiative and non-initiative states is concentrated in education. Specifically, the education spending differential is due

9

primarily to reductions in state aid to school districts for elementary and secondary education. 4. Through the Lens of History That the contemporary initiative-related fiscal differential is specific to education is fortuitous, from a scholarly standpoint, because the federal government has been collecting state-level education finance statistics since 1870, beginning with the Biennial Survey of Education, now known as the Digest of Education Statistics.9 This deep reservoir of data allows me to explore the origins of the fiscal differentials related to the voter initiative. Most important, I am able to observe fiscal outcomes before any state had actually adopted the initiative.10 For this analysis, I constructed a state-by-year panel of education finance statistics from 1889 to 2010.11 Because the data were collected from print archives and entered manually, I was parsimonious in my selection of variables, compiling those most relevant to the present inquiry: total education funding per pupil, as well separate data for state government funding and local government funding. These variables speak to the main findings above, namely that initiative states today have lower total education spending, driven specifically by lower spending by their state governments. I converted the spending variables to constant 1996 dollars using the GDP deflator from Sutch (2006).12 4.1 Trends in Education Funding

9

The Biennial Survey of Education was the federal government’s first publication to systematically track statistics related to state and local education. The Biennial Survey began publication in 1869, changed title to become the Digest of Education Statistics in 1960, and has been published under that name to this day. These two publications are the source of data for all the figures presented in this section. 10 Recall that South Dakota was the first state to adopt the initiative, doing so in 1898. The majority of states that were to adopt the initiative had done so by 1918 (see Figure 1). 11 Some state education data are available as early as 1873, but reporting did not become consistent until the 1889-1890 school year. 12 Alaska and Hawaii are excluded from the analysis.

10

I begin by plotting the time-series of average education spending over time. For comparison, I group all the states that would ever adopt the initiative into one group and all those that have never adopted the initiative into another. I plot the average over time separately for each group irrespective of which states had actually adopted the initiative at any point in time. Doing so nullifies compositional effects and makes evident any differences between erstwhile initiative and non-initiative states that may have existed prior to the adoption of the initiative. Figure 2A shows the time-series of total education funding per pupil from 1889 to 2010. Because education spending has increased manifold over time, the scale of the yaxis is in log form to enhance readability.13 Figure 2B depicts the log difference in mean spending between would-be initiative and non-initiative states in each year from 18892010.14 There are obvious differences between would-be initiative and non-initiative states even before any state had ever adopted the initiative. Indeed, there appear to be three distinct phases in the relationship between state initiative status and education finance. From 1889 (at least) up until about 1930, initiative states on average spent more on education than non-initiative states. From the 1930s through the 1970s, there is no apparent difference between average spending in initiative versus non-initiative states. Beginning in the 1980s, initiative states have fallen slightly behind non-initiative states in education spending.

13

The scale of education funding has increased dramatically over time in real terms. In inflation-adjusted 1996 dollars, the average state spent $171 per pupil in 1889 and $9,313 per pupil in 2010. (In nominal terms, the average state spent $11 per student in 1889 and $12,293 in 2010.) Given the difficulty of displaying the data on a common scale, I do not show confidence bounds, but rather provide an accompanying graph, Figure 2B, depicting the mean difference and confidence interval over time. 14 Figures 2B, 3B, and 4B are produced from annual regressions of log spending on the initiative indicator, estimated one year at time from 1889 through 2010. States are always classified according to their ultimate initiative status. The coefficients form these regressions are equivalent to the mean difference in log spending between initiative and non-initiative states.

11

Figure 3A and 3B show the time-series of local government education funding for would-be initiative and non-initiative states. Pre-treatment differences between initiative and non-initiative states can be seen again, with initiative states providing higher levels of local education funding. Over time, spending levels of the two groups of states converged, and from the mid-1970s initiative states have spent less on average than noninitiative states. Figures 4A and 4B show the evolution of average education funding by state governments over the same time span. In the earliest years, the pattern is somewhat noisy, in part because state funding was more volatile at the time, occasionally showing large year-to-year fluctuations within a state. Figure 4C depicts median (rather than mean) spending for the two groups of states and reveals a somewhat clearer picture: state education funding was lower in would-be initiative states even before the initiative had been adopted. Moreover, the difference is persistent. Except for the period between 1910 and 1925, initiative states have regularly delivered lower state education funding than non-initiative states. To put into context the scale of the pre-existing differences between wouldinitiative and non-initiative states, Figure 5 shows average education funding in nominal terms for the period 1889 to 1940. Differences between the two groups of states are evident before and after any had adopted the initiative and appear relatively stable throughout the entire period. Table 2 summarizes funding for would-be-initiative and non-initiative states from 1889 to 1898, the last year before any state had adopted the initiative. Average annual total education funding was $13.13 per pupil in would-be initiative states and $9.84 in would-be non-initiative states and the difference is

12

significant at p = 0.03. Corresponding figures for local education expenditures are $10.98 and $7.69 (p = 0.03), and for state funding $1.97 and $2.19 (p = 0.36), for would-be initiative and would-be non-initiative states, respectively. 4.2 Other Differences between Initiative and Non-Initiative States Given the extremely long time period under study and the relative dearth of statelevel data available for the earliest years, I was not able compile as rich a set of control variables as was used for the analysis in Part I of the paper. However, I have assembled a set of key variables that span the entire 120-year time period. In this section, I introduce those variables and describe trends over time between the two groups of states. State-level income data come from the Bureau of Economic Analysis for years 1929 onward and from Turner et al. (2006) for earlier years. The variable is defined as income per private sector worker (see Turner et al. (2006) for details). As Figure 6 demonstrates, initiative states were more prosperous than non-initiative states around the turn of the 20th century, incomes gradually converged, and by the turn of the 21st century initiative states were less affluent. Table 3 shows that the difference between would-be initiative and non-initiative states was significant in the decade prior to the introduction of the initiative (p = 0.05). Figure 7 plots income and education funding together. For each year, Figure 7 shows the ratio of income in the average initiative state relative to the average non-initiative state, as well as the ratio of per pupil education funding. Both series reveal a pattern wherein would-be initiative states begin higher than non-initiative states up, convergence to rough equality by 1925, and then show a reversal of fortunes by 1980 or so. Thus, income suggests itself as one possible factor in explaining the observed trends in spending between the two groups of states. For instance, in 1900 the average

13

would-be initiative state had 1.21 times the income per worker of the average would-be non-initiative state, and provided 1.27 times the funding per pupil for education. A century later, the average initiative state had 0.94 times the income per worker of the average non-initiative state and provided 0.88 times as much education funding per pupil. In the intervening years, the income and funding ratios tracked one another relatively well, although not perfectly. Another striking difference between initiative and non-initiative states is that the former have historically had a significantly larger share of males in the population, as shown in in Figure 8. The data are from the decennial census. Perhaps reflecting the frontier status of the West at the time, the average would-be initiative state was 56 percent male at the turn of the 20th century; by comparison, the average non-initiative state was 50.75 percent male at the time.15 The difference is highly significant statistically, as shown in Table 3. Over time, the gender composition of the two groups of states converged, although initiative states remain more male to this day (the contemporary difference is not significant, however). There are relatively small differences in the rate of urbanization between initiative and non-initiative states. As shown in figure 9, would-be initiative states were less urban than non-initiative states at the turn of the 20th century and slightly more urban by the turn of the 21st century. However, the differences in urbanization between the two groups of states have never been significant. 4.3 Did the Initiative Increase Education Spending in the First Half of the 20th Century?

15

Note that these figures reflect the average state, not necessarily the national population.

14

Clearly there were significant differences in fiscal, economic, and demographic outcomes between would-be initiative and non-initiative states that predate the adoption of the institution. These “pre-treatment” differences complicate estimation of the effect of the initiative on fiscal outcomes. In this section, I present a series of OLS and fixed effects models that attempt to uncover the all-else-equal relationship between the voter initiative and education funding. I begin by focusing on the period pre-WWII, when previous scholarship suggests that the initiative led to increased spending (Matsusaka 2000). In each model, I regress education spending against a standard set of control variables: income per worker, enrollment, population, urbanization, and percent male. I also control for the average first-dimension DW-NOMINATE score of the state’s two U.S. senators, a measure of political ideology (Poole and Rosenthal 1991). All models include year fixed effects, which account for any secular trends. The dependent variables, as well as income, enrollment, and population, are log transformed. In Table 4, the dependent variable is combined state and local education funding per pupil. The initiative dummy is set to 1 for those years after a state had adopted the initiative, zero before and for states that never adopted the initiative. Model 1 presents the simple bivariate regression, indicating that states with the initiative spend roughly 30 percent more. Model (2) introduces control variables and the coefficient drops by half but remains significant. Model (3) adds state fixed effects to implement a difference-indifferences design. The coefficient on the initiative dummy changes signs and becomes smaller in magnitude and statistically insignificant. The change in the initiative coefficient between models (2) and (3) indicates that much of the difference in spending

15

between initiative and non-initiative states can be attributed to time-invariant omitted variables that differ between the two groups of states. Within states over time, education spending did not appear to change significantly in the years after the adoption of the initiative. The identifying assumption for the difference-in-differences analysis is parallel trends, meaning that initiative and non-initiative states would have experienced identical changes over time but only for the adoption of the initiative. That assumption appears to be problematic in light of Figures 6 to 9, for example, which show obvious differences in the trajectory of key variables over time between the two groups of states. As a partial remedy, models (4) and (5) introduce region- and state-specific linear time trends, respectively. Identification in these models comes from within-state changes over time in the deviation from regional or state-specific linear time trends. In model (4), the initiative coefficient is essentially unchanged relative to the baseline difference-in-differences estimate. In model (5), the initiative coefficient is once again positive, but not significant. Tables 5 and 6 repeat the same analyses separately for local and state government education funding, respectively. The simple bivariate model shows that local government funding was roughly 40 percent higher in states with the voter initiative (see model (1) of Table 5). The coefficient drops to 13 percent and becomes insignificant when covariates are added in model (2). Introduction of fixed effects in model (3) causes the initiative coefficient to change signs, though it remains insignificant. The initiative coefficient is smaller and remains insignificant in the models with regional and state time trends. The initiative coefficient is always insignificant in the models of state government funding for education (Table 6). The bivariate model shows that state funding is lower in

16

states with the initiative. The coefficient changes signs when covariates are added, drops effectively to zero in the difference-in-differences analysis, then remains positive at around 9 percent in the models with time trends. In sum, the point estimates are volatile and never significant. An important identification strategy in the existing literature, originating with Matsusaka (1995), is to interact the initiative dummy with signature requirements. The logic is that the initiative should have a greater effect where it is more easily available, meaning that states with lower signature requirements for placing a measure on the ballot should experience greater initiative effects. To explore this idea, Tables 7-9 repeat the analyses of Tables 4-6 with the addition of the signature requirement. This variable is set equal to the signature requirement in years after the initiative has been adopted, and to zero before adoption and for states that never adopted the initiative. In effect, this specification amounts to an interaction between the initiative dummy and the signature requirement. Throughout the analyses, the signature requirement variable is correctly signed in general—meaning oppositely signed relative to the initiative dummy itself—but not significant. For instance, model (1) of Table 8 shows that, for local government funding, the initiative dummy is positively signed and the signature requirement is negatively signed, meaning that the initiative has a smaller effect in states where the signature requirement is higher, as predicted. The model suggests that an initiative state with signature requirement of 5 percent would spend about 50 percent more, while an initiative state with a 10 percent signature requirement would spend about 26 percent more, than a state without the initiative. Again, however, the interaction is not significant.

17

Perhaps more concerning for this line of analysis is that the same interaction between the initiative and the signature requirement is evident in the decade before any state had adopted the initiative, as shown in Table 10. In these models, states are coded according to their ultimate initiative status and signature requirements even though no state had the initiative yet during the time period 1889 to 1898. If anything, the signature requirement interaction is larger and more precisely estimated in these models, suggesting that states that were to adopt more stringent signature requirements were already different from those that would adopt weaker signature requirements at a time when none had yet adopted the initiative. Again, however, the differences fall short of significance at the conventional 5 percent level. 4.4 Did the Initiative Cut Education Spending in the Second Half of the 20th Century? The results from the preceding section suggest that the voter initiative had no effect on education funding—either positive or negative—in the first half of the 20th century, challenging one cornerstone of the existing literature (Matsusaka 2000). Another major finding from prior studies, explored in Part I above, is that the initiative led to reductions in spending in the second half of thee 20th century. The evidence offered in prior studies amounts to a cross-sectional comparison of initiative and non-initiative states nearly a century after the institution was adopted. Had the initiative been adopted at random, it would be credible to compare the long-run outcomes for the “treated” and “control” states, just as one might compare long-run outcomes for participants in a medical trial. But the adoption of the initiative was not random, as demonstrated in the preceding section, and contemporary cross-sectional correlations should be viewed with

18

caution in light of the pre-existing differences between would-be initiative and noninitiative states revealed in the preceding section. Such concerns notwithstanding, the broad trajectory of government spending is at least consistent with the notion that the initiative led to long-run reductions in spending. As illustrated in Figure 2, government spending on education was higher in would-be initiative states prior to the adoption of the institution and gradually declined in relative terms over time, ultimately falling below non-initiative state spending. It is at least plausible that the initiative could be part of the explanation. Table 11 presents analyses of education funding using the entire 120-year time series. Model (1) controls only for year fixed effects, meaning that the coefficient on the initiative dummy simply represents the average difference in spending between the two groups of states across the entire time period, essentially a long-run correlation. The coefficient is roughly 5 percent (though not significant). Model (2) adds state fixed effects and thereby transforms the analysis into a difference-in-differences, which is negative and significant, consistent with the long-run relative decline in initiative-state spending shown in Figure 2. However, this is a valid causal estimate only if the parallel trends assumption is satisfied; that is, only if initiative and non-initiative states would have followed identical spending trajectories but for the adoption of the initiative. Such an assumption is dubious in light of the other differential trends documented in Figures 6 to 9, which provide good reason to believe that the relative spending of the two groups of states would have changed even if the initiative had not been introduced. Indeed, when I add a set of time-varying covariates in Model (3), the initiative effect disappears. Income, in particular, stands out as a confounder, with an elasticity of about 1, which further

19

codifies the relationships depicted in Figure 7. In other words, the change in relative spending can be explained by other differential trends over time between initiative and non-initiative states. The covariates added in model (3) capture only readily observable time-varying differences across the states. There were likely other, possibly unobservable, factors changing over time differentially between the states. For instance, the income convergence between initiative and non-initiative states is likely only one component of broader regional economic convergence in the 20th century (e.g., Barro et al. 1991). To capture differential time trends more generally, models (4) and (5) add regional and statespecific linear time trends, respectively. The addition of these differential trends further reduces the magnitude of the initiative coefficient, although the difference relative to model (3) is negligible. Estimates for state and local education funding separately (reported in the Appendix) similarly show no significant effect of the initiative after accounting for time-varying covariates and/or regional or state-specific trends. The difference-in-differences estimates are based on comparisons of all the years after a state adopted the initiative to all the years before. Given the extremely long time span under consideration, such a comparison may be too crude. Based on Figure 2, it appears that there was a long and gradual period of decline in relative initiative state spending, which may not be fully captured by comparing the average of all the years before to the average of all the years after adoption of the initiative. I explore this question by respecifying the models to allow the initiative coefficient to vary over time

20

after adoption. Specifically, I create dummy variables for each decade after adoption, which allows the initiative estimate to vary non-parametrically over time.16 Table 12 shows the results of the analyses, which are again conducted in a differences-in-differences framework. Regardless of specification, there is no indication of an initiative effect within the three decades after adoption. In Model (1), which excludes controls, there is evidence of lower spending in initiative states starting in the fourth decade after adoption. However, adding controls in Model (2) causes the differential to disappear for everything less than 5 decades after adoption, and adding regional or state time trends completely eliminates the difference between initiative and non-initiative states. Table 12 thus provides further evidence that differential trends between initiative and non-initiative states—both observable and unobservable—account for the apparent different trends in spending. 4.4.1 Recent Adoptions Even if they exist, identifying effects of the initiative in the latter 20th century is especially challenging, given that there were apparently no effects immediately after adoption. To claim that the initiative affected fiscal policy in the later 20th century amounts to claiming that the relationships depicted Figures 2 through 4, at some point changed from being correlational to being causal. An alternative approach, which sidesteps the methodological challenges of identifying the onset of effects decades after a treatment, is to examine those states that adopted the voter initiative in the latter half of the 20th century. As figure 1 shows, there are three such states: Wyoming (1968); Florida

16

For parsimony, I truncate the decades-after-initiative indicator at “5 or more.”

21

(1968); Mississippi (1992).17 If the voter initiative caused reductions in spending during this period, the effects should be readily apparent in these states. Because so few states changed initiative status during this period, I use synthetic control methods for the analysis.18 That is, I compare each “treated” state to a “synthetic” control state, which is a weighted average of other states selected to be observationally equivalent to the treated state at the time of treatment. I trace each state from 10 years before the adoption of the initiative to 20 years after adoption. The results are shown in Figure 10. None of the states that adopted the initiative most recently exhibits a relative decline in spending after adoption. Wyoming is often excluded from analyses of state finance because of its heavily reliance on revenue from oil and gas, which might explain its massive increase in spending in the 1970s and early 1980s. So the results for Wyoming should be taken with a grain of salt. More notably, however, both Florida and Mississippi also exhibit relative increases in spending after initiative adoption. Mississippi further illustrates the perils of drawing causal inferences from crosssectional comparisons. As evident from Figure 10B, it was not possible to construct a close synthetic match for Mississippi. This is because Mississippi was already the lowest spending state in the nation before it adopted the voter initiative, meaning that no combination of other states could yield average spending as low. Looking at Mississippi only after the adoption of initiative could leave the spurious impression that its low spending is the product of direct democracy. Instead, spending in Mississippi actually increased after initiative adoption, relative to the synthetic control unit.

17 18

Alaska adopted the initiative in 1956 but is excluded from the analysis for other reasons. For an overview of synthetic control methods, see Abadie et al. (2015).

22

The experience of the three states that most recently adopted the initiative further casts doubt on the notion that the institution reduced spending in the latter half of the twentieth century. 5. Generalizability The preceding analysis is the first to study changes in spending within states over time before and after the adoption of the voter initiative, drawing on 120 years of data. My analysis is made possible by the long historical record of school finances compiled by the Department of Education. As such, however, my analysis is also limited in scope and may not generalize to other categories of spending. In Section 3, above, I showed that most of the spending differential between initiative and non-initiative states today is due to education, and so even if my findings only applied to this category, they would be important to the literature more generally. I conclude by investigating whether trends over time in education spending track total spending for the years in which both are available. Figure 11 shows the ratio of spending between initiative and non-initiative states computed from three separate data sources. The first is the education finance data set I constructed for this paper, which was used in the preceding analyses. The second is the Sylla, Legler, and Wallis (1995) data set used in Matsusaka (2000), which covers the period 1902 to 1942. The third is Census of Governments data on direct general expenditures per capita for state and local governments, which is available starting in 1942. The three data sources give a consistent picture of the trends in relative spending across time. While the consistency between education and total spending in the years of common data availability does not prove that my results generalize to non-education

23

spending, given the results shown in Section 3 and Figure 11, there is no obvious reason to suspect that education is unrepresentative of the general pattern of relative spending. Another important caveat is that, like most of the prior literature, I have asked whether the adoption of the voter initiative changed fiscal outcomes, not whether specific initiatives have had effects. Others, however, have investigated the impact of a wide range of winning initiatives and concluded that their effects are often diluted in the process of implementation (Gerber et al. (2001), but see New (2010)). Moreover, according to canonical theories, the initiative operates through the “threat effect” and therefore the passage of any individual initiative is out-of-equilibrium behavior (Matsusaka and McCarty 2001). That said, nothing in my analysis rules out the possibility that specific initiatives had fiscal effects, even if fiscal policy does not differ between initiative and non-initiative states on average. 6. Interpretation It is important to emphasize just how different the conclusions are that emerge from this historical analysis compared to examining only recent years of data. The post1970 data, seen in isolation from the historical record, would appear to indicate that initiative states spend less, and that they do so primarily be reducing spending on education. What does the historical analysis imply for conventional wisdom about the fiscal effects of the voter initiative? First, the historical analysis underscores the difficulty of identifying the causal effects of the initiative. The voter initiative is not as good as randomly assigned and “pretreatment” differences can be seen across many variables. There is even evidence of pre-existing variation in spending related to future signature

24

requirements. As such, cross-sectional comparisons of outcomes in initiative vs. noninitiative states are unlikely to reveal the causal effect of the institution and should be interpreted cautiously. Even difference-in-differences analyses are open to question, as the parallel trends assumption is suspect given the differential trends over time in economic and demographic variables between the two groups of states. Second, the preceding cautions notwithstanding, I find no evidence that the voter initiative caused increases in spending the first half of the 20th century. Rather, would-be initiative states were already spending more than non-initiative states before any state had adopted the institution, and there is little evidence of within-state increases in spending after adoption. Third, lower spending by state governments in the latter half of the 20th century appears especially unlikely to have been the product of the voter initiative. But for a handful of years surrounding WWI, state governments have always spent less in initiative states, even before the adoption of the initiative. Figure 4 and the accompanying regression analyses seem to indicate that there is a persistent difference between the two groups of states that was unaffected by the adoption of the initiative itself. Of course, one could construct a counter-factual history in which state governments would have increased their relative spending were it not for the adoption of the initiative, but that story is not obvious from the literature. An open question in the history of state government spending is what accounts for the rapid rise and fall of initiative state spending relative to non-initiative state spending around WWI. Pursuing this question may shed more light on the early history of the voter initiative.

25

Fourth, to the extent that the historical record supports a fiscal effect of the voter initiative, it is for local government, not the state government. Local government spending in would-be initiative states was higher before the adoption of the initiative and lower afterward. While my results suggest that the change was not caused by the initiative, but rather by a more general pattern of economic convergence between the two groups of states, further analysis of the evolution of local government spending in over the 20th century would be important to unpack the details beyond the relatively simple difference-in-differences analysis conducted here. If indeed the initiative has affected local government education spending, the underlying explanation of the mechanism would likely differ in important respects from the standard story of how the initiative influences fiscal policy. The standard explanation relies heavily on the threat effect. Yet the threat effect is harder to maintain for local government policy. From the perspective of any individual local government, the threat of a statewide voter initiative in response to its policies is relatively remote. That is, if each local government is small relative to total state population, the ability of that jurisdiction’s voters to threaten a statewide initiative is equally small. As such, local governments face a collective action problem. Any individual jurisdiction might chose to reduce spending in order to please its voters and forestall an initiative, as in the standard threat effect story. But doing so is a sort of public good for all the other local jurisdictions, and we expect public goods to be underprovided for all the usual reasons. Any explanation for how the voter initiative affects local government spending would likely have to lean less heavily on the threat effect as the mechanism, or offer a different interpretation for how the threat effect operates for large group of independent

26

jurisdictions. Scholarly, attention may also need to turn to voter initiatives in local government, as distinct from the state-level initiatives that have been the subject of most of the existing literature (see Primo 2010). 5. Conclusion My aim in this paper has been to advance scholarship on direct democracy by elucidating the specific fiscal policy differences between initiative and non-initiative states and investigating their origins using a novel historical data set. While most of the prior literature on the subject attributes contemporary policy differences between initiative and non-initiative states to the institution itself, my historical analysis shows no evidence of a causal effect of the initiative on fiscal outcomes. If a causal interpretation of the voter initiative is to be salvaged, fleshing out the complex history of 20th century state and local government finance is a major challenge for scholars and a tremendous opportunity for future research. The burden now shifts to those who believe in the effectiveness of the voter initiative to provide more compelling evidence of it.

27

REFERENCES Abadie, Alberto, Alexis Diamond, and Jens Hainmueller. 2015. “Comparative Politics and the Synthetic Control Method.” American Journal of Political Science 59(2): 495-510. Arceneaux, Kevin, “Direct Democracy and the Link between Public Opinion and State Abortion Policy,” State Politics and Policy Quarterly, December 2002, Vol. 2(4), 372-387. Baicker, Katherine, and Nora Gordon. 2004. “The Effect of Mandated State Education Spending on Total Local Resources.” NBER Working Paper 10701. Bailey, Stephen J., and Stephen Connolly, “The Flypaper Effect: Identifying Areas for Further Research,” Public Choice, 95 (June 1998), 335-61. Barro, Robert J., Xavier Sala-I-Martin, Olivier Jean Blanchard and Robert E. Hall. “Convergence Across States and Regions.” Brookings Papers on Economic Activity Vol. 1991, No. 1 (1991), pp. 107-182 Berne, Robert, and Leanna Stiefel. The Measurement of Equity in School Finance: Conceptual, Methodological, and Empirical Dimensions (Johns Hopkins University Press, 1983). Berry, Christopher. 2007. “The Impact of School Finance Judgments on State Fiscal Policy.” In School Money Trials: The Legal Pursuit of Educational Adequacy, Martin West and Paul Peterson, eds. Washington, D.C.: Brookings Institution Press, 213-240. Berry,  William  D.,  Evan  J.  Ringquist,  Richard  C.  Fording,  and  Russell  L.  Hanson,  “Measuring   Citizen  and  Government  Ideology  in  the  American  States,  1960-­‐93,”  American   Journal  of  Political  Science,  42  (January  1998):  327-­‐48.   Besley, Timothy, and Stephen Coate. 2008. “Issue Unbundling via Citizens’ Initiatives.” Quarterly Journal of Political Science 3: 379-397. Boehmke, Frederick J. 2005. The Indirect Effect of Direct Democracy: How Institutions Shape Interest Group Systems. The Ohio State University Press. Boehmke, Frederick J., and R. Michael Alvarez. 2005. “The Influence of Initiative Signature Gathering Campaigns on Political Participation.” Working paper. Boehmke, Frederick J., and John W. Patty. 2007. “The Selection of Policies for Ballot Initiatives: What Voters Can Learn from Legislative Inaction.” Economics & Politics 19(1): 97-121. Bowler, Shaun and Todd Donovan. 1995. “Popular Responsiveness to Taxation.” Political Research Quarterly 48: 79-100. Bowler, Shaun and Todd Donovan. 2004. “Measuring the Effects of Direct Democracy on State Policy: Not All Initiatives Are Created Equal,” State Politics and Policy Quarterly Vol. 4(3), 345-363. Bowler, Shaun and Todd Donovan. 2004b. “Evolution in State Governance Structures: Unintended Consequences of State Tax and Expenditure Limitations.” Political Research Quarterly 57: 189-96. Broder, David S. 2000. Democracy Derailed: Initiative Campaigns and the Power of Money. New York: Harcourt Brace. Camobreco, John F. 1998. “Preferences, Fiscal Policies, and the Initiative Process.” Journal of Politics 60: 819-29. Card, David, and A. Abigail Payne. 1998. “School Finance Reform, the Distribution of School Spending, and the Distribution of Student Test Scores,” Journal of Public Economics 83, no. 1 (2002): 49–82. Cronin, Thomas E. 1989. Direct Democracy: The Politics of Initiative, Referendum, and Recall. Cambridge, MA: Harvard University of Press. Donovan, Todd, and Shaun Bowler. 1998. "An Overview of Direct Democracy in the American States," in Shaun Bowler, Todd Donovan, and Caroline Tolbert, eds., Citizens as Legislators. Columbus, Ohio: Ohio State University Press.. Donovan, Todd, Caroline J. Tolbert, and Daniel A. Smith. 2008. “Priming Presidential Votes by Direct Democracy,” Journal of Politics 70: 1217-31.

28

Ellis, Richard. 2002. Democratic Delusions: The Initiative Process in America. Lawrence: University of Kansas Press Erikson, Robert S., Gerald C. Wright; John P. McIver, 2007, "Replication data for: Public Opinion in the States: A Quarter Century of Change and Stability", (http://php.indiana.edu/~wright1/cbs7603_pct.zip). Gerald C. Wright, Distributor. Fischer, Justina A.V. 2008. “Direct Democracy and Public Education in Swiss Cantons.” In N. C. Soguel and P. Jaccard (eds.), Governance and Performance of Education Systems. London: Springer 137–153. Fletcher, D., and Kenny, L.W. (forthcoming). “The Influence of the Elderly in a Median Voter Framework.” Education Finance and Policy. Galbraith, James K., and Travis Hale. 2006. “State Income Inequality and Presidential Election Turnout and Outcomes.” The University of Texas Inequality Project Working Paper 33. Garrett, Elizabeth. 2008. “Direct Democracy and Public Choice.” USC Legal Studies Research Paper No. 08-20. Garrett, Elizabeth. 1999. “Money, Agenda Setting, and Direct Democracy.” Texas Law Review 77: 1845-90. Gerber, Elisabeth. 1996. “Legislative Responses to the Threat of Popular Initiatives.” American Journal of Political Science 40. ______. 1999. The Populist Paradox: Interest Group Influence and the Promise of Direct Legislation. Princeton, NJ: Princeton University Press. Gerber, Elisabeth, and Arthur Lupia. 1995. “Campaign Competition and Policy Responsiveness in Direct Legislation Elections.” Political Behavior 17: 287-306. Gerber, Elisabeth, Arthur Lupia, Mathew D. McCubbins, and D. Roderick Kiewiet, Stealing the Initiative: How State Government Responds to Direct Democracy, Upper Saddle River, NJ: Prentice Hall, 2001. Guthrie, James. “Twenty-First-Century Education Finance: Equity, Adequacy, and the Emerging Challenge of Linking Resources to Performance.” In Money, Politics, and Law: Intersections and Conflicts in the Provision of Educational Opportunity; 2004 Yearbook of the American Education Finance Association, edited by Karen DeMoss and Kenneth K. Wong (Larchmont, N.Y.: Eye of Education, 2004). Hahn, Harlan, and Sheldon Kamieniecki. 1987. Referendum Voting: Social Status and Policy Preferences. Westport, CT: Greenwood Press. Hero, Rodney, and Caroline Tolbert. 1996. “A Racial/Ethnic Diversity Interpretation of Politics and Policy in the States of the U.S.” American Journal of Political Science 40: 851-71. Holcombe, Randall G., and Lawrence W. Kenny. 2008. “Does Restricting Choice in Referenda Enable Governments to Spend More?” Public Choice 136: 887-101. Hoxby, Caroline M. 1998. “All School Finance Equalizations Are Not Created Equal.” NBER Working Paper 6792. Cambridge, MA: National Bureau of Economic Research. Kouser, Thad, and Mathew D. McCubbins. 2005. “Social Choice, Crypto-Initiatives, and Policymaking by Direct Democracy.” Southern California Law Review 78(4): 949-84. Kousser, Thad, Mathew D. McCubbins and Kaj Rozga (2008), ‘When Does the Ballot Box Limit the Budget? Politics and Spending Limits in California, Colorado, Utah, and Washington’, in Garrett, Elizabeth, Elizabeth A. Graddy and Howell E. Jackson (eds), Fiscal Challenges: An Interdisciplinary Approach to Budget Policy, Cambridge: Cambridge University Press, pp. 290-321. Lascher, Edward L., Michael G. Hagen, and Steven A. Rochlin. 1996. “Gun Behind the Door? Ballot Initiatives, State Policies, and Public Opinion.” Journal of Politics 58: 760-75. Lupia, Arthur. 1994. “Shortcuts versus Encyclopedias: Information and Voting Behavior in California Insurance Reform Elections.” American Political Science Review 88: 63-76. Magleby, David B. 1984. Direct Legislation: Voting on Ballor Propositions in the United States. Baltimore: Johns Hopkins University Press.

29

Matsusaka, John. 1992. “Economics of Direct Legislation.” Quarterly Journal of Economics 107 (May). _______. 1995. “Fiscal Effects of the Voter Initiative: Evidence from the Last 30 Years.” Journal of Political Economy 103. _______. 2004. For the Many or the Few: The Initiative, Public Policy, and American Democracy. Chicago: The University of Chicago Press. _______. 2005. “Direct Democracy Works.” Journal of Economic Perspectives 19(2), 185-206. _______. 2007. “Direct Democracy and Social Issues.” Working paper: Marshall School of Business, University of Southern California. Matsusaka, John, and Nolan McCarty. 2001. “Political Resource Allocation: Benefits and Costs of Voter Initaitives.” Journal of Law, Economics, and Organization 17: 413-48. McCubbins, Colin. 2008. “It’s All About the Benjamins: Growth in Government and Revenue Initiatives in the American States.” Working Paper, UC San Diego. Mullins, Daniel R., and Bruce A. Wallin. 2004. “Tax and Expenditure Limitations: Introduction and Overview.” Public Budgeting and Finance (Winter): 2-15. Murray, Shelia, William N. Evans, and Robert Schwab, “Education Finance Reform and the Distribution of Education Resources,” American Economic Review 88 (September 1998): 789–812. National Center on Education Statistics, Digest of Education Statistics (U.S. Department of Education, Institute of Education Sciences, various years) (www.nces.ed.gov/programs/digest [September 2006]); Common Core of Data (U.S. Department of Education, Institute of Education Sciences, various years) (www.nces.ed.gov/ccd [September 2006]). New, Michael. 2010. “U.S. State Tax and Expenditure Limitations: A Comparative Political Analysis.” State Politics and Policy Quarterly 10(1): 25-50. Nicholson, Steven. 2005. Voting the Agenda: Candidates, Elections, and Ballot Propositions. Princeton, NJ: Princeton University Press. Peltzman, Sam. 1992. “Voters as Fiscal Conservatives.” Quarterly Journal of Economics 107 (May): 327-61. Poole, Keith T. and Howard Rosenthal. 1991. “Patterns of Congressional Voting.” American Journal of Political Science 35 (Feb.): 228-78. Primo, David. 2010. “The Effect of Initiatives on Local Government Spending.” Journal of Theoretical Politics 22(1): 6-25. Romer, Thomas, and Howard Rosenthal. 1979. “Bureaucrats versus Voters: On the Political Economy of Resource Allocation by Direct Democracy.” Quarterly Journal of Economics 93 (Nov.): 563-87. Sabato, Larry J., Howard R. Ernst, and Bruce A. Larson, eds. 2001. Dangerous Democracy: The Battle over Ballot Initiative in America. Lanham, MD: Rowman and Littlefield. Schrag, Peter. 1998. Paradise Lost: California’s Experience, America’s Future. New York: New Press. Smith, Daniel A. 1998. Tax Crusaders. London: Routledge. Smith, Daniel A. 2004. “Peeling Away the Populist Rhetoric: Toward a Taxonomy of Anti-Tax Ballot Initiatives.” Public Budgeting and Finance (Winter): 88-110. Smith, Daniel A., and Dustin Fridkin. 2008. “Delegating Direct Democracy: Interparty Legislative Competition and the Adoption of the Initiative in the American States.” American Political Science Review 102(3): 333-350. Smith, Daniel A., and Caroline J. Tolbert. 2004. Educated by Initiative: The Effects of Direct Democracy on Citizens and Political Organizations in the American States. Ann Arbor: University of Michigan Press. Sutch, Richard. 2006. “ Gross domestic product: 1790–2002 [Continuous annual series].” Table Ca9-19 in Historical Statistics of the United States, Earliest Times to the Present:

30

Millennial Edition, edited by Susan B. Carter, Scott Sigmund Gartner, Michael R. Haines, Alan L. Olmstead, Richard Sutch, and Gavin Wright. New York: Cambridge University Press. Sylla, Richard E., John B. Legler, and John Wallis. 1995. STATE AND LOCAL GOVERNMENT [UNITED STATES]: SOURCES AND USES OF FUNDS, CENSUS STATISTICS, TWENTIETH CENTURY [THROUGH 1982]. New York, NY: Richard E. Sylla, New York University/Athens, GA: John B. Legler, University of Georgia/College Park, MD: John Wallis, University of Maryland [producers]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor]. Turner, Chad, Robert Tamura, Sean Mulholland, and Scott Baier. 2006. “Education and Income of the States of the United States, 1840-2000.” Journal of Economic Growth 12: 101-158. Weingast, Barry R., and William J. Marshall. 1988. “The Industrial Organization of Congress; or, Why Legislatures, Like Firms, Are Not Organized as Markets.” Journal of Political Economy 96 (Feb.):132-63. Weingast, Barry R., Kenneth A. Shepsle, and Christopher Johnsen. 1981. “The Political Economy of Benefits and Costs: A Neoclassical Approach to Distributive Politics.” Journal of Political Economy 89 (Aug.): 642-64.

31

Figure 1: Initiative States with Year of Adoption

Note: Initiative states shown in blue with year of adoption. Alaska (not shown) adopted the initiative in 1956.

32

100

Total Funding per Pupil (Log Scale) 1000 5000 10000

Figure 2A: Average Combined (State + Local) Education Funding 1889-2010

SDUT SD UTOR

1900

MT OK MO ME MI CO AR C CA AZ WA OH NVND NV NE A ID ND

MA

WY FL

1925

1950

Initiative States Avg.

MS

1975

2000

Non-Initiative States Avg.

Note: Funding is expressed in constant 1996 dollars. States are grouped by ultimate initiative status. State postal codes denote year of initiative adoption.

-.2

0

.2

.4

.6

.8

Figure 2B: Difference in (State + Local) Education Funding 1889-2010

1900

1925

1950 start Point Estimate

1975

2000

90% CI

Note: The red line represents the coefficient from a rolling regression of log spending per pupil against a dummy for would-be initiative status, estimated annually from 1889 to 2010.

33

100

Local Funding per Pupil (Log Scale) 1000

5000

Figure 3A: Average Local Education Funding, 1889-2010

SDUT SD UTOR

1900

MT OK MO ME MI CO AR C CA AZ WA OH NVND NV NE A ID ND

MA

WY FL

1925

1950

Initiative States Avg.

MS

1975

2000

Non-Initiative States Avg.

Note: Funding is expressed in constant 1996 dollars. States are grouped by ultimate initiative status. State postal codes denote year of initiative adoption.

-.5

0

.5

1

1.5

Figure 3B: Difference in Local Education Funding, 1889-2010

1900

1925

1950 start Point Estimate

1975

2000

90% CI

Note: The red line represents the coefficient from a rolling regression of log spending per pupil against a dummy for would-be initiative status, estimated annually from 1889 to 2010.

34

State Funding per Pupil (Log Scale) 50 100 1000

5000

Figure 4A: Average State Education Funding, 1889-2010

SDUT SD UTOR

1900

MT OK MO ME MI CO AR C CA AZ WA OH NVND NV NE A ID ND

MA

WY FL

1925

1950

Initiative States Avg.

MS

1975

2000

Non-Initiative States Avg.

Note: Funding is expressed in constant 1996 dollars. States are grouped by ultimate initiative status. State postal codes denote year of initiative adoption.

-2

-1

0

1

Figure 4B: Difference in State Education Funding, 1889-2010

1900

1925

1950 start Point Estimate

1975

2000

90% CI

Note: The red line represents the coefficient from a rolling regression of log spending per pupil against a dummy for would-be initiative status, estimated annually from 1889 to 2010.

35

State Funding per Pupil (Log Scale) 50 100 1000

5000

Figure 4C: Median State Education Funding, 1889-2010

SDUT SD UTOR

MT OK MO ME MI CO AR C CA AZ WA OH NVND NV NE A ID ND

MA

1900

WY FL

1925

1950

Initiative States Median

MS

1975

2000

Non-Initiative States Median

Note: Funding is expressed in constant 1996 dollars. States are grouped by ultimate initiative status. State postal codes denote year of initiative adoption.

0

Total Funding per Pupil 20 40 60

80

Figure 5: Average Education Funding (Nominal Dollars), 1889-1940

1890

1900

1910

Initiative States Avg.

1920

1930

1940

Non-Initiative States Avg.

Note: Funding (combined state and local) is expressed in nominal dollars. States are grouped by ultimate initiative status.

36

0

Real Income per Capita 20000 40000 60000

80000

Figure 6: Real Income per Worker (2010 Dollars), 1890-2010

1900

1925

1950

Initiative States Avg.

1975

2000

Non-Initiative States Avg.

Note: Income per private sector worker from BEA for 1929 onward and from Turner et al. (2006) for earlier years. Values are in constant 2010 dollars. States are grouped by ultimate initiative status.

1.6 1.4 1.2 1 .8

Ratio: Avg. Initiative State/Avg. Non-Initiative State

1.8

Figure 7: Income Ratio vs. Education Funding Ratio, 1889-2010

1900

1925

1950 year

Education Funding per Pupil

1975

2000

Income per Capita

Note: The dashed line represents the ratio of education funding per pupil for the average initiative state relative to the average non-initiative state. The solid line represents the ratio of income per worker in the average initiative state relative to the average non-initiative state. States are grouped by ultimate initiative status.

37

48

Population Percent Male 50 52 54

56

Figure 8: Population Percent Male, 1890-2010

1900

1925

1950

Initiative States Avg.

1975

2000

Non-Initiative States Avg.

Note: Male percent of adult population States are grouped by ultimate initiative status.

30

40

Percent Urban 50 60

70

Figure 9: Urbanization, 1890-2010

1900

1925 Initiative States Avg.

1950

1975

2000

Non-Initiative States Avg.

Note: Percent urban population. States are grouped by ultimate initiative status.

38

Figure 10A: Synthetic Control Analysis for Florida

1000

2000

Spending per Pupil 3000 4000 5000

6000

Florida

1960

1970

1980

1990

Year treated unit

synthetic control unit

Figure 10B: Synthetic Control Analysis for Mississippi

3000

Spending per Pupil 4000 5000 6000

7000

Mississippi

1980

1990

2000

2010

Year treated unit

synthetic control unit

39

Figure 10C: Synthetic Control Analysis for Wyoming

2000

Spending per Pupil 4000 6000 8000

10000

Wyoming

1960

1970

1980

1990

Year treated unit

synthetic control unit

40

Ratio of Initiative State to Non_Initiative State .8 1 1.2 1.4 1.6 1.8

Figure 11: Trends in Education Spending and Total Spending, by Data Source

1900

1950 Census Education

2000 Wallis

Notes: “Wallis” is total spending per capita from the Sylla, Legler, and Wallis (1995) data set. “Census” is total state and local direct expenditures per capita, from the Census of Governments. “Education” is total expenditures per pupil from the data set constructed by me for this paper.

41

Table  1:  Initiative  vs.  Non-­‐Initiative  State  Spending,  1970-­‐2002       (1)       Dummy  =  1  if  initiative  state     State  per  Capita  Income       Intergovernmental  Revenue  from  Federal,  per  Capita     ln(Population)     State  Pct  Metro  Pop       Pct  Population  Change  Over  Previous  5  Years     Proporiton  of  the  population  aged  >65     Proportion  of  the  population  aged  5-­‐17     Unemployment  Rate  of  the  Civilian  Labor  Force     DW  NOMINATE     Region=Midwest     Region=South     Region=West     Constant       Observations   R-­‐squared  

 

(2)  

 

Total       -­‐161.80   (71.60)**   0.12   (0.02)***   2.07   (0.34)***   25.69   (68.10)   1.52   (1.89)   1,570.45   (492.14)***   11.19   (17.08)   17.29   (21.13)   29.25   (10.25)***   -­‐41.13   (115.17)   205.56   (110.29)*   -­‐117.90   (90.68)   147.66   (96.78)   -­‐1,418.82   (1,293.10)  

Education       -­‐84.16   (36.61)**   0.03   (0.01)***   0.40   (0.09)***   -­‐10.69   (34.55)   0.05   (1.42)   226.46   (474.49)   -­‐9.12   (11.51)   17.67   (12.65)   -­‐0.76   (8.76)   0.71   (43.12)   136.92   (54.86)**   -­‐76.21   (52.56)   114.16   (66.88)*   351.06   (486.74)  

(3)   Non-­‐ Education       -­‐77.63   (50.75)   0.09   (0.01)***   1.67   (0.30)***   36.38   (53.71)   1.47   (1.50)   1,343.99   (599.89)**   20.31   (13.51)   -­‐0.38   (19.03)   30.01   (10.02)***   -­‐41.85   (94.60)   68.64   (80.89)   -­‐41.69   (70.67)   33.50   (73.40)   -­‐1,769.87   (1,112.22)  

  1,505   0.91  

  1,505   0.76  

  1,505   0.89  

Notes:  Standard  errors,  clustered  by  state,  are  reported  in  parentheses.    Analysis  covers  1970-­‐2002,   excepting  2001.  All  models  also  include  year  dummies.  Alaska,  Hawaii,  and  Wyoming  are  excluded.  *   significant  at  10%,  **  significant  at  5%,  ***  significant  at  1%.  

42

Table  2:  Average  Education  Funding  per  Pupil,  1889-­‐1898       Never-­‐Initiative  States   Would-­‐be  Initiative  States   Total    $9.84      $13.13     State    $2.19      $1.97     Local    $7.69      $10.98     Note:  Nominal  dollars.  States  grouped  by  ultimate  initiative  status.   p-­‐values  adjusted  for  clustering  by  state.  

 Difference:   p-­‐value     0.03   0.36   0.03  

Table  3:  Covariate  Balance  Tests,  1889-­‐1898    Difference:   Never-­‐Initiative   Would-­‐be       States   Initiative  States   p-­‐value     Real  Income  per  Worker    9,626.22      11,741.89     0.05   Percent  Male   50.75   56.18   0.00   Percent  Urban   30.21   27.97   0.35   NOMINATE   -­‐0.11   0.08   0.04   Note:  States  grouped  by  ultimate  initiative  status.   Income  per  worker  in  constant  2010  dollars.   p-­‐values  adjusted  for  clustering  by  state.  

43

Table  4:  Total  (Combined  State  and  Local)  Education  Funding,  1889-­‐1940             Initiative  Dummy     Income   per  Worker  (Log)     Population   (Log)     Enrollment   (Log)     Population   Pct  Urban     Population   Pct  Male     DW   NOMINATE     Constant      Observations  

(1)  

(2)  

      0.314**   (0.121)  

      0.149***   (0.0525)   0.892***     (0.0695)     0.165     (0.127)     -­‐0.194     (0.120)     0.00642***     (0.00152)     0.0289***     (0.00632)     0.354***     (0.0586)     4.986***   -­‐4.924***   (0.0980)   (0.718)     1,433   0.401   Yes  

(3)  

(4)  

(5)  

      -­‐0.0246   (0.0457)   0.436***   (0.142)   0.366   (0.296)   -­‐0.281   (0.227)   0.00858   (0.00654)   0.0137   (0.0189)   0.0752   (0.0461)   -­‐1.687   (2.886)  

      -­‐0.0104   (0.0479)   0.329**   (0.132)   0.452*   (0.254)   -­‐0.322   (0.201)   0.00683   (0.00574)   0.0127   (0.0257)   0.0213   (0.0462)   -­‐49.38***   (6.782)  

      0.0513   (0.0423)   0.274**   (0.115)   1.519***   (0.253)   -­‐0.859***   (0.177)   0.00373   (0.00619)   -­‐0.0120   (0.0296)   0.0200   (0.0360)   -­‐39.61***   (6.391)  

  1,372   0.912   Yes  

      1,372   1,372   1,372   R-­‐squared   0.887   0.892   0.919   Year  FE   Yes   Yes   Yes   State  FE   Yes   Yes   Yes       Linear  Trends               Region   State   The  dependent  variable  is  the  log  of  total  education  funding  per  pupil  in  constant  1996  dollars.   The  initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.  Models  that   include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by  state  in   parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1            

44

Table  5:  Local  Government  Education  Funding,  1889-­‐1940             Initiative  Dummy  

(1)  

(2)  

(3)  

(4)  

(5)  

      0.405**   (0.155)  

(0.174)  

      0.132   (0.107)   1.481***   (0.326)   -­‐0.298   (0.274)   0.244   (0.265)   0.00561   (0.00339)   0.0237   (0.0244)   0.567***   (0.156)   -­‐9.387***   (2.214)  

      -­‐0.0756   (0.0905)   0.782*   (0.404)   -­‐0.820   (0.532)   0.631   (0.397)   0.0104   (0.00884)   0.0173   (0.0282)   0.109   (0.131)   -­‐0.319   (4.994)  

      -­‐0.0650   (0.0869)   0.546   (0.382)   -­‐0.745   (0.560)   0.548   (0.451)   0.00867   (0.00804)   0.0337   (0.0296)   -­‐0.00296   (0.138)   -­‐52.91***   (8.272)  

      0.0100   (0.0785)   0.751***   (0.271)   0.151   (0.783)   -­‐0.328   (0.404)   0.00307   (0.00635)   0.0604*   (0.0317)   -­‐0.000382   (0.147)   -­‐60.56***   (8.823)  

  1,400   0.267   Yes  

  1,345   0.808   Yes  

 Income  per  Worker  (Log)    Population  (Log)    Enrollment  (Log)    Population  Pct  Urban    Population  Pct  Male    DW  NOMINATE    Constant      Observations  

                        4.607***  

      1,345   1,345   1,345   R-­‐squared   0.747   0.760   0.846   Year  FE   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Linear  Trends                   Region   State   The  dependent  variable  is  the  log  of  local  government  education  funding  per  pupil  in  constant   1996  dollars.  The  initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.   Models  that  include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by   state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1              

                             

45

  Table  6:  State  Government  Education  Funding,  1889-­‐1940             Initiative  Dummy    Income  per  Worker  (Log)    Population  (Log)    Enrollment  (Log)    Population  Pct  Urban    Population  Pct  Male    DW  NOMINATE    Constant      Observations  

(1)  

(2)  

(3)  

(4)  

(5)  

      -­‐0.0930   (0.301)  

(0.246)  

      0.226   (0.279)   0.160   (0.503)   1.980*   (1.067)   -­‐1.943*   (1.000)   0.00543   (0.0145)   -­‐0.0599   (0.0663)   -­‐1.124***   (0.344)   0.453   (5.891)  

      -­‐0.0132   (0.297)   -­‐0.369   (0.476)   1.748*   (0.949)   -­‐0.874   (0.658)   0.0157   (0.0261)   -­‐0.104   (0.119)   -­‐0.631**   (0.264)   -­‐2.354   (13.80)  

      0.0885   (0.302)   -­‐0.474   (0.505)   2.282**   (0.904)   -­‐1.106*   (0.656)   0.0144   (0.0268)   -­‐0.214   (0.134)   -­‐0.670***   (0.239)   -­‐28.43   (28.96)  

      0.0866   (0.316)   -­‐0.976*   (0.572)   5.890***   (1.208)   -­‐0.893   (0.823)   0.0225   (0.0306)   -­‐0.308   (0.206)   -­‐0.312   (0.192)   47.56   (34.89)  

  1,411   0.167   Yes  

  1,352   0.281   Yes  

                        2.621***  

      1,352   1,352   1,352   R-­‐squared   0.337   0.345   0.515   Year  FE   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Linear  Trends                   Region   State   The  dependent  variable  is  the  log  of  state  education  funding  per  pupil  in  constant  1996  dollars.  The   initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.  Models  that  include   state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by  state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1              

46

 

 

Table  7:  Total  Funding,  Including  Signature  Requirement,  1889-­‐1940           Initiative  Dummy     Signature   Requirement  

(1)  

(2)  

(3)  

(4)  

(5)  

      0.445**   (0.186)   -­‐0.0193   (0.0289)  

      0.340**   (0.156)   -­‐0.0284   (0.0209)   0.899***   (0.0697)   0.184   (0.131)   -­‐0.213*   (0.124)   0.00627***   (0.00149)   0.0292***   (0.00644)   0.347***   (0.0585)   -­‐5.023***   (0.733)  

      0.0310   (0.174)   -­‐0.00783   (0.0223)   0.448***   (0.154)   0.363   (0.298)   -­‐0.278   (0.228)   0.00881   (0.00691)   0.0135   (0.0194)   0.0739   (0.0457)   -­‐1.788   (2.812)  

      -­‐0.0403   (0.158)   0.00427   (0.0198)   0.322**   (0.147)   0.456*   (0.257)   -­‐0.324   (0.201)   0.00664   (0.00608)   0.0126   (0.0249)   0.0217   (0.0461)   -­‐49.51***   (7.003)  

      0.0712   (0.161)   -­‐0.00288   (0.0213)   0.276**   (0.115)   1.516***   (0.253)   -­‐0.858***   (0.177)   0.00376   (0.00619)   -­‐0.0119   (0.0295)   0.0198   (0.0359)   -­‐39.62***   (6.398)  

  Income   per  Worker  (Log)     Population   (Log)     Enrollment   (Log)     Population   Pct  Urban     Population   Pct  Male     DW   NOMINATE     Constant  

                        4.986***   (0.0980)  

              Observations   1,433   1,372   1,372   1,372   1,372   R-­‐squared   0.402   0.913   0.887   0.892   0.919   Year  FE   Yes   Yes   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Linear  Trends                   Region   State   The  dependent  variable  is  the  log  of  total  education  funding  per  pupil  in  constant  1996  dollars.  The   initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.  Models  that  include   state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by  state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1  

47

Table  8:  Local  Government  Funding,  Including  Signature  Requirement           Initiative  Dummy     Signature   Requirement  

(1)  

(2)  

(3)  

(4)  

(5)  

      0.716***   (0.228)   -­‐0.0456   (0.0336)  

      0.504**   (0.226)   -­‐0.0548*   (0.0306)   1.496***   (0.326)   -­‐0.259   (0.282)   0.204   (0.271)   0.00529   (0.00336)   0.0242   (0.0244)   0.555***   (0.156)   -­‐9.580***   (2.233)  

      0.205   (0.313)   -­‐0.0395   (0.0426)   0.842*   (0.432)   -­‐0.835   (0.537)   0.648   (0.402)   0.0115   (0.00929)   0.0159   (0.0288)   0.102   (0.134)   -­‐0.809   (5.136)  

      0.114   (0.253)   -­‐0.0255   (0.0346)   0.586   (0.411)   -­‐0.771   (0.575)   0.557   (0.455)   0.00976   (0.00809)   0.0348   (0.0313)   -­‐0.00544   (0.139)   -­‐52.16***   (8.634)  

      0.242   (0.286)   -­‐0.0337   (0.0408)   0.771***   (0.274)   0.118   (0.797)   -­‐0.317   (0.409)   0.00341   (0.00619)   0.0620*   (0.0319)   -­‐0.00220   (0.147)   -­‐60.66***   (8.925)  

  Income   per  Worker  (Log)     Population   (Log)     Enrollment   (Log)     Population   Pct  Urban     Population   Pct  Male     DW   NOMINATE     Constant  

                        4.607***   (0.174)  

              Observations   1,400   1,345   1,345   1,345   1,345   R-­‐squared   0.269   0.810   0.748   0.760   0.847   Year  FE   Yes   Yes   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Linear  Trends                   Region   State   The  dependent  variable  is  the  log  of  local  government  education  funding  per  pupil  in  constant   1996  dollars.  The  initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.   Models  that  include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by   state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1                        

48

Table  9:  State  Government  Funding,  Including  Signature  Requirement           Initiative  Dummy     Signature   Requirement  

(1)  

(2)  

(3)  

(4)  

(5)  

      -­‐1.211   (0.847)   0.164   (0.104)  

      -­‐0.892   (0.865)   0.165   (0.102)   0.116   (0.493)   1.863*   (1.081)   -­‐1.821*   (0.999)   0.00638   (0.0141)   -­‐0.0614   (0.0637)   -­‐1.087***   (0.337)   1.033   (5.877)  

      1.011   (0.740)   -­‐0.144   (0.0935)   -­‐0.153   (0.494)   1.691*   (0.942)   -­‐0.813   (0.654)   0.0197   (0.0266)   -­‐0.109   (0.118)   -­‐0.656**   (0.267)   -­‐4.152   (13.89)  

      0.936   (0.757)   -­‐0.121   (0.0953)   -­‐0.285   (0.550)   2.160**   (0.899)   -­‐1.062   (0.660)   0.0196   (0.0276)   -­‐0.208   (0.140)   -­‐0.682***   (0.242)   -­‐24.87   (31.63)  

      0.654   (0.852)   -­‐0.0823   (0.128)   -­‐0.926   (0.570)   5.809***   (1.240)   -­‐0.867   (0.826)   0.0233   (0.0313)   -­‐0.304   (0.210)   -­‐0.317   (0.194)   47.32   (35.18)  

  Income   per  Worker  (Log)     Population   (Log)     Enrollment   (Log)     Population   Pct  Urban     Population   Pct  Male     DW   NOMINATE     Constant  

                        2.621***   (0.246)  

              Observations   1,411   1,352   1,352   1,352   1,352   R-­‐squared   0.177   0.291   0.343   0.349   0.515   Year  FE   Yes   Yes   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Linear  Trends                   Region   State   The  dependent  variable  is  the  log  of  state  government  education  funding  per  pupil  in  constant   1996  dollars.  The  initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.   Models  that  include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by   state  are  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1                        

49

Table  10:  “Pre-­‐Treatment”  Analysis,  1889-­‐1898           Would-­‐be  Initiative  Dummy     Would-­‐be   Signature  Requirement     Constant  

(1)   Total       0.823**   (0.336)   -­‐0.0656   (0.0428)   4.832***   (0.135)  

(2)   Local       1.269***   (0.441)   -­‐0.0862*   (0.0497)   4.310***   (0.254)  

(3)   State       -­‐1.558   (1.505)   0.117   (0.200)   2.960***   (0.273)  

          Observations   336   327   334   R-­‐squared   0.134   0.133   0.078   The  dependent  variables  in  models  (1)  to  (3)  are  the  natural  log  of  per  pupil   education  funding  from  total,  local,  and  state  sources,  respectively.  The  analysis   include  years  1889-­‐1898,  before  any  state  had  adopted  the  voter  initiative.  States   are  coded  according  to  their  future  initiative  status  and  signature  requirements.   Standard  errors  clustered  by  state  are  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1        

50

 Table  11:  Combined  State     and  Local    Education     Funding,  1   889-­‐2010             Initiative  Dummy  

(1)  

(2)  

(3)  

(4)  

(5)  

      0.0485   (0.0627)  

      -­‐0.226**   (0.103)  

                            4.986***   (0.0978)  

                            4.935***   (0.0567)  

      -­‐0.0146   (0.0461)   0.961***   (0.0742)   0.324*   (0.168)   -­‐0.416**   (0.156)   0.00133   (0.00212)   0.0176   (0.0109)   0.0349   (0.0588)   0.0927***   (0.0314)   -­‐4.325***   (0.913)  

      -­‐0.00690   (0.0520)   0.703***   (0.0775)   0.374**   (0.171)   -­‐0.423***   (0.148)   0.00219   (0.00245)   -­‐0.0105   (0.0126)   0.0458   (0.0476)   0.0694**   (0.0303)   -­‐44.21***   (2.551)  

      0.00948   (0.0496)   0.521***   (0.108)   0.650***   (0.162)   -­‐0.570***   (0.159)   0.00134   (0.00343)   -­‐0.0115   (0.0171)   0.0703   (0.0490)   0.0384   (0.0257)   -­‐46.34***   (3.679)  

  3,977   0.928   Yes  

  3,977   0.970   Yes   Yes      

  3,916   0.989   Yes   Yes      

  3,916   0.990   Yes   Yes   Region  

  3,916   0.992   Yes   Yes   State  

 Income  per  Worker  (Log)     Population   (Log)    Enrollment  (Log)     Population   Pct  Urban    Population  Pct  Male     Population   Growth  Rate    DW  NOMINATE     Constant       Observations   R-­‐squared   Year  FE   State  FE   Linear  Trends  

     

The  dependent  variable  is  the  log  of  local  government  education  funding  per  pupil  in  constant   1996  dollars.  The  initiative  dummy  is  equal  to  one  in  years  after  a  state  has  adopted  the  initiative.   Models  that  include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors  clustered  by   state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1            

51

 Table  12:  Differential  Time     Trends,  Total     Education     Spending,  1   889-­‐ 2010       (1)   (2)   (3)   (4)   Decades  After  Adoption                               Less  than  1   0.0509   0.0766   0.0549   0.0249   (0.0626)   (0.0474)   (0.0450)   (0.0407)     2   1  to   -­‐0.000691   0.0566   0.0384   0.0142   (0.0822)   (0.0554)   (0.0670)   (0.0716)     3   2  to   -­‐0.121   0.0105   -­‐0.00404   -­‐0.0357   (0.0783)   (0.0461)   (0.0583)   (0.0640)     4   3  to   -­‐0.209**   -­‐0.0414   -­‐0.0494   -­‐0.0730   (0.0913)   (0.0552)   (0.0726)   (0.0797)     5   4  to   -­‐0.284**   -­‐0.0734   -­‐0.0602   -­‐0.0672   (0.114)   (0.0560)   (0.0669)   (0.0869)   More  t  han  5   -­‐0.444***   -­‐0.137**   -­‐0.0764   -­‐0.0507   (0.130)   (0.0535)   (0.0564)   (0.0906)     Constant   4.929***   -­‐3.425***   -­‐45.02***   -­‐46.98***   (0.0548)   (1.014)   (2.549)   (3.694)               Observations   3,977   3,916   3,916   3,916   R-­‐squared   0.974   0.989   0.990   0.992   Controls   No   Yes   Yes   Yes   Year  FE   Yes   Yes   Yes   Yes   State  FE   Yes   Yes   Yes   Yes   Linear  Trends           Region   State   The  dependent  variable  is  the  log  of  local  government  education  funding  per   pupil  in  constant  1996  dollars.  The  table  reports  the  coefficient  of  the  initiative   dummy  interacted  with  indicators  for  decades-­‐since-­‐adoption.  Models  2-­‐4   include  the  standard  set  of  control  variables  (not  reported).  Models  that   include  state  fixed  effects  report  the  within  R-­‐squared.  Standard  errors   clustered  by  state  in  parentheses.   ***  p<0.01,  **  p<0.05,  *  p<0.1        

52