Pivotal Politics and Initiative Use in the American States

Pivotal Politics and Initiative Use in the American States Frederick J. Boehmke Tracy L. Osborn Emily U. Schilling University of Iowa June 08, 2015 ABSTRACT We incorporate the role of pivotal players in the legislature into a spatial model of the direct initiative process to study the interplay between these two policymaking institutions. We show that the size of the standard gridlock interval relates to the ability of the legislature to preempt initiatives, and that the presence of the initiative process often reduces the size of the traditional gridlock interval. Further, we find that initiatives can also arise in situations without legislative gridlock because pivotal players sometimes prefer to block legislation in order to ensure an initiative proposal passes. Specifically, we find that as the distance between the median voter and the pivot player closest to the initiative proposer increases, initiative use also increases. Empirical analysis of initiative use in the American states supports this prediction. Corresponding author: frederick-boehmke@uiowa.edu. Boehmke is Professor of Political Science and Director of the Iowa Social Science Research Center; Osborn is Associate Professor of Political Science; Schilling is a Ph.D. candidate in Political Science. We thank Edward Lascher, Justin Esarey, and seminar participants at Emory University, the Unversity of North Carolina, Chapel Hill, and Washington University for helpful comments and suggestions. 1

Introduction The direct legislation process, particularly the initiative and referendum, is portrayed as a way for citizens and organized interests to influence public policy when state legislatures refuse to act. As envisioned by Progressive reformers over a century ago, institutions such as the direct initiative process would serve as a gun behind the door to encourage legislatures to pass policies concordant with citizens preferences. Despite valid concerns about the contemporary practice and use of direct legislation, existing scholarship suggests that the initiative process still generally serves the interests of the many rather than the few (Matsusaka s 2004; Gerber 1999; Boehmke 2005a). The relatively infrequent use of the initiative process comports with its intended role as a corrective rather than a primary means of policymaking. Even during the resurgence in initiative use in recent decades, the most frequent users only experience an average of two successful measures per year (Boehmke 2005a). The success of such measures embodies the direct effect of the initiative process, whereby citizens and organized interests influence policy directly through the passage of proposals rather than through the legislative arena. In addition, theoretical accounts of the initiative process highlight a less observable though perhaps more important indirect effect. These models suggest that legislators, cognizant of the threat of a ballot measure, strategically pass policies closer to the median voter in order to preempt a successful proposal (see, e.g., Gerber 1996; Matsusaka and McCarty 2001; Boehmke 2005a). Both forms of influence thereby lead direct legislation to facilitate policy responsiveness to the median voter. An alternate stream of work focuses on the role of legislative institutions that foster gridlock and thus hinder policy responsiveness (Binder 1999; Brady and Volden 1998; Cox and 2

McCubbins 1993). Krehbiel (1998) embodies one strand of this research by arguing that procedures such as the filibuster and the executive veto create obstacles that may thwart elected officials ability to implement policy change and thereby decrease policy responsiveness. As political institutions, therefore, the initiative process and legislatures often sit in opposition to each other: direct legislation provides an incentive for policymakers to moderate policies towards voters; whereas, supermajoritarian legislative procedures can have a deterrent effect on policy movement. While academic work has overlooked the interplay between these institutions, the initiative process ability to counteract legislative inaction has long been recognized in the popular dialogue. For example, the Los Angeles Times stated that groups are trying to fill the Sacramento power vacuum by bypassing the legislative gridlock and going directly to the voters (Skelton 2011). We show that simultaneously considering supermajoritarian legislative procedures and direct legislation in the policymaking process advances our understanding of both institutions. Further, by assessing gridlock and the initiative in tandem, we also speak to the common perception that direct legislation serves as a release valve when gridlock thwarts policy change. We do this by developing a formal model of the policymaking process in initiative states that fuses the pivotal politics model of legislative policymaking (e.g., Krehbiel 1998) with spatial models of the initiative process (e.g., Gerber 1996). Our integrated model generates predictions about the conditions under which legislative preemption fails and initiatives appear on the ballot. Primarily, we find that initiatives are not a response to legislative gridlock; rather, they occur because pivotal actors in the legislature prefer to block legislation in favor of their preferred outcome offered by the initiative process. We also find that the initiative process generally reduces the size of the traditional legislative gridlock 3

interval. We assess the first prediction by modeling initiative use as a function of the relative locations of pivotal actors with three distinct tests drawing on different measures of state legislators ideal points, finding consistent support for our theory. Overall, our work highlights the importance of considering the interplay between multiple institutions in a political system. Modeling the Initiative Process and Pivotal Politics In this section we present our model that combines policymaking via the legislature and the initiative process, provide an overview of equilibrium conditions, and discuss the model s testable predictions. Particularly, we focus on predictions about the conditions under which the initiative process will change policy outcomes vis-à-vis the simple pivotal politics model. The theoretical appendix to the paper presents a detailed presentation of the model. Models of Direct Legislation Models of the direct legislation process typically originate from Romer and Rosenthal s (1978; 1979) one dimensional spatial model of agenda setting in which the setter makes a take it or leave it offer to voters. If the offer fails, policy reverts either to the status quo or to some legislatively prescribed reversion point. Gerber (1996) adapts this model to direct legislation by giving proposal power to an outside actor, (e.g. an organized interest group). After the legislature passes its policy, the proposer can pay some cost, c, to place an alternate proposal on the ballot. Voters then decide between the two options. The model s central conclusion highlights the indirect effect of the initiative process: by usurping the legislature s policymaking authority, the initiative process forces the legislature to moderate its proposals toward the median voter in order to preempt ballot measures. With this 4

policy moderation, the legislature deters the group from using the initiative process to pass a less desirable policy. This legislative moderation occurs whenever the legislature and proposer compete over the median voter s ideal point: either when the voter lies between the legislature and the proposer or when the proposer lies between the voter and the legislature. Variations on this basic model (e.g., Matsusaka and McCarty 2001; Boehmke 2005a) allow for uncertainty over the median voter s ideal point; yet, the equilibrium remains similar in spirit. 1 Empirical tests of this model focus on whether the ability to propose initiatives engenders policies that better reflect the median voter s preferences. On balance, the evidence supports this prediction across a variety of policies, such as tax rates (Matsusaka 1995, 2004), abortion laws (Burden 2005; Gerber 1999), gaming policy (Boehmke 2005a), or laws restricting same sex marriage (Lupia et al. 2010); though not all studies find such an effect (e.g., Burden 2005; Lascher, Hagan, and Rochlin 1996; Lax and Phillips 2009). The initiative stage of our model begins from the same premises as Gerber s (1996) canonical model. Policymaking and ideal points exist in a one dimensional policy space with three relevant actors: the legislature, L, which acts first by selecting a policy, denoted x L ; the proposer, P, who then has the option of paying a cost, c, to qualify a measure, x P, for the ballot; and the median voter, V, who chooses between the policies of the legislature and the proposer, provided the latter opts to make a proposal. We adapt this model by relaxing the assumption that the legislature (and executive) is a unitary actor and instead include multiple players in the government as in Krehbiel s (1998) pivotal politics model. 2 Modeling Legislative Policymaking 5

In the pivotal politics model, passing legislation requires the support of pivotal actors. Legislative rules and procedures determine the identities and the spatial location of these actors. The first pivot point is the filibuster. The filibuster allows a minority group of legislators to stall a vote by (the threat of) indefinite debate until cloture is invoked, often via a supermajority vote. In a one-dimensional spatial model, the filibuster pivot, F, represents the member whose support must be won in order to ensure sufficient votes to invoke cloture, allowing movement from the status quo, denoted by x 0. Once a bill receives a majority vote in the legislature the executive must approve it; however, the executive possesses veto power. In the states, we refer to the pivotal role of the Governor, G, who has varying ability across the states to veto legislation. Should such a veto occur, the legislature could override the veto with a supermajority vote. The override rule identifies a second pivot known as the veto override player. The roles of the governor and veto override player interact since the veto override legislator, O, only matters if the governor s veto poses a credible threat. The legislative process in the pivotal politics model often results in gridlock. In order to pass, new legislation must benefit at least three pivotal players: the median legislator, since a majority of legislators must prefer a new proposal to the status quo; the filibuster pivot, since a minority can indefinitely stall a proposal until a successful cloture vote; and either the governor or the veto override pivot, since if the former chooses to veto, the latter must vote to override it. As with the initiative process, adding the filibuster and override rules into legislative policymaking models empowers certain actors (i.e. pivots) and leads to final policy outcomes that can diverge from the median legislator s ideal point. Further, the combination of these requirements leads to 6

a subset of policies falling between the veto and filibuster pivots in which policy cannot be moved the gridlock interval. Combining the Initiative and Pivotal Politics Models We combine these two models by retaining the legislative process in the pivotal politics model and adding two additional stages from the initiative model. 4 First, after the pivotal politics legislative process ends with policy, x L, the initiative proposer can pay a cost, c, to place its own proposal, x P, on the ballot. Second, if the measure, x P, appears on the ballot, the median voter, V, in the state selects her preferred policy, x L or x P. The game then ends. The equilibrium outcome of our game depends on the arrangement of the six players ideal points, but we can reduce the number of combinations due to restrictions on the ordering of legislative actors ideal points, i.e., F L O. The median legislator must always lie between the filibuster and veto override players by construction. 5 Further, we assume that the proposer has an extreme ideal point to capture the idea that organized interests tend to represent either more liberal or conservative positions. For presentational purposes we assume a liberal interest group; in the appendix we discuss the results if we assume a conservative proposer. Similarly, we assume that the governor lies outside the gridlock interval, just to the right of the veto override legislator. If the governor lies inside the gridlock interval between the median legislator and the override legislator, then the governor becomes the veto pivot and similar results hold. Finally, we locate all actors ideal points and the status quo between zero and one for presentation. We present the results of our model based on four different configurations of ideal points given by the location of the median voter: to the left of the gridlock interval (V < F); between 7

the filibuster pivot and the median legislator (F < V < L); between the median legislator and the veto pivot (L < V < O); and, finally, to the right of the veto pivot (O < V). The equilibrium outcome then depends on the location of the status quo, creating multiple possible policy outcomes within each of these four configurations. To summarize these, we present the equilibrium policy outcome (focusing on subgame perfect equilibria) in Figure 1. The horizontal and vertical axes represent the policy space, and notations along the axes indicate the location of each actor s ideal point. The red line plots the equilibrium outcome of our gridlockinitiative game on the vertical axis against the location of the status quo on the horizontal axis. For comparison, the blue line does the same for the solely pivotal politics game, while the orange line plots the optimal initiative for every policy (i.e., initiatives without a legislature). We set the cost of proposing an initiative to 0. In the appendix we present a similar graph with a small cost of proposal; generally, as long as the cost of the initiative is not too large, the main results hold with only minor modifications to reflect the cost of proposal. [Figure 1 about here.] Figure 1 shows that the possibility of initiatives often changes the equilibrium outcome in the pivotal politics model. Generally, these deviations result from the threat of an initiative. The initiative threat either leads the legislature to alter its policy choice to preempt an initiative or leads the proposer to place a measure on the ballot. Initiatives can occur in equilibrium when the legislature can make no better policy through the legislative process (i.e., the preemptive policy change fails to happen). This occurs when one of the non-median pivotal actors in the legislature blocks legislation to secure something better from the initiative than what the legislature will do. Such a failure of preemptive legislation can only happen when the median voter and the proposer sit on opposite sides of that pivotal actor. As we show in the appendix, initiatives only occur 8

when the optimal ballot measure falls into the gridlock interval, though this constitutes a necessary rather than a sufficient condition. Intuitively, when a ballot measure falls into the gridlock interval at least one of the veto or filibuster pivots would be made worse off by passing any other policy, thereby creating an incentive to block the legislative process in favor of the initiative policy. Overall, the equilibrium outcome represents a combination of the forces operating in Gerber s initiative model and Krehbiel s pivotal politics model. In configuration 1, for example, when the median voter sits between the proposer and the filibuster pivot, the result mirrors Gerber s (1996) model: the legislature preempts an initiative by moving policy to the voter s ideal point. Moving policy anywhere to the right of the voter s ideal point results in an initiative that pivots around the median voter towards the proposer. In the other three configurations, however, the median voter s ideal point lies to right of the filibuster pivot, leading to conflict among the pivotal players over where to move policy given the threat of an initiative. If the legislature passes a policy to the right of the voter s ideal point, the proposer can select an initiative that it prefers to the left of the median voter that is also closer to the median voter s preferred policy. This decision by the proposer changes the dynamic in the legislature, since gridlock now leads to the proposer s policy via the ballot instead of reverting to the status quo ante in the legislative process. When the proposer s policy lies in the gridlock interval, the legislature again finds itself trapped; in order to preempt a ballot measure in the gridlock interval, the median legislator would have to move policy away from at least one of the two other pivots. This move is impossible, since both of the pivots must approve of the proposed policy change and can thwart the change if they disapprove. Thus, the legislature can, at best, propose the same policy as the proposer, rendering itself indifferent between legislative 9

action and a ballot proposal. Once the optimal initiative moves back out of the gridlock interval, there again exists a policy that the legislature prefers to the expected initiative policy. This legislative policy makes at least two pivots strictly better off (and the other weakly better off) thus precluding use of the initiative process. Findings and Predictions Table 1 summarizes the predictions of our gridlock-initiative game for each of the four main configurations. We highlight three main testable propositions. First, our game implies that the presence of the initiative process in a state alters the traditional gridlock interval. When the initiative is considered in tandem with the legislative process, gridlock only occurs when the legislative gridlock interval overlaps with the initiative gridlock interval. This interval may be zero, as in configuration 1, smaller than the traditional gridlock interval as in configurations 2 or 3, or identical, as in configuration 4. Scholars testing pivotal politics style theories in the states should be cognizant of these changes stemming from direct democracy. [Table 1 about here] Second, our model demonstrates that the initiative process can make the median voter better or worse off relative to legislative-only policy outcomes. The possibility of being worse off has previously occurred in models with uncertainty (Matsusaka and McCarty 2001), but here it occurs due to the actions of the pivotal players. For example, consider case three in Figure 1 with a status quo just to the right of the veto override player. In the game without the initiative, policy would pivot around O back towards the median voter. With the initiative process, however, the filibuster pivot blocks this move since it prefers the initiative that pivots around V (taking the outcome further from the median voter). Further, note that this utility loss occurs despite a 10

successful initiative. This prediction speaks to debates over whether the initiative is a positive or negative institutional feature. Third, our model shows that initiatives occur over a wide range of status quo points. Thus, initiatives can occur with regularity where permitted by state law, because a multitude of legislative configurations result in equilibria involving a successful ballot measure. 6 Since this third prediction forms the basis for our subsequent empirical tests, we explore it in more detail. Consider the outcomes summarized in Table 1. 7 In configuration 1, no initiative occurs since the legislature can always preempt it. In configurations 2 and 3, initiatives can occur whenever the proposer will benefit and the resulting optimal measure, given the status quo ante, lies in the gridlock interval. However, the proposer cannot pass anything until the status quo lies opposite the median voter, so the region in which initiatives can occur begins at L. When policy moves to the right of L, the median legislator and veto pivot would like to preempt an initiative at V, but the filibuster pivot stops them since the resulting initiative will be closer to his ideal point. The resulting initiative measure continues to lie in the gridlock interval until it crosses F. This happens when the status quo passes V + V F. Therefore, the size of the policy space that can lead to initiatives therefore corresponds to V F. Configuration 4 looks similar, except that the legislature can initially preempt a ballot measure when the status quo crosses V. Thus, the optimal initiative would lie to the right of the veto override. An initiative to the right of the veto override allows the median legislator to propose a legislative policy equidistant to the left of O, which it and the filibuster pivot prefer. Once the optimal initiative moves into the gridlock interval, pivoting ends since no policy to the left of the ballot measure would make the veto override pivot better off. Thus, the permissive initiative interval starts at V + V O and continues as before until V + V F. The size of the 11

policy space that can lead to initiatives in configuration 4, therefore, corresponds to O F, the familiar pivotal politics gridlock interval. 8 Pivotal Politics and Initiative Use In order to evaluate the relationship between the size of our gridlock intervals and initiative use, we require data on the number of citizen-sponsored initiatives appearing on statewide ballots and the location of various actors ideal points. 10 The Initiative and Referendum Institute s website provides information on the former from the year the state adopted the initiative process through 2010. Of the twenty-four states that currently permit initiatives, some allow them only on general election ballots, while others allow them on primary and off-year ballots. We follow convention in the literature by analyzing the number of ballot measures per biennium. To measure the location of pivotal actors ideal points, we draw on three available data sources. First, we rely on Clark, Osborn, Winburn, and Wright s (2009) data on state legislative ideology for the 1999-2000 and 2003-2004 sessions. For these data, one concern is the ideology scores are calculated one state at a time, so that they are not directly comparable across states. Because the intervals measure the difference between ideal points, however, problems of location do not affect our measures in these data. Differences in scale across chambers and states do remain a concern, though. Second, we use Shor and McCarty s (2011) data on state legislative ideal points for 1993-2009. 11 These data cover a longer time period and are comparable over time and across states, as the authors use the Project Vote Smart National Political Awareness Test (NPAT) to put legislators ideological scores in a common ideological space. 12 Finally, we utilize Masket s (2004) data on California Assembly ideal points from 1901-2003. These data provide us an extended time series for one state. Like the Shor and McCarty data, Masket s data 12

allow ideological comparison across legislators in different sessions on a common ideological scale. These data sources represent a near universe of the available roll call data collection in the U.S. states, and thus the available ideological estimators of legislators positions, at the time of our study. Measuring Initiative Intervals Across our four configurations of the median voter s location, we find that three different status quo regions permit initiatives. When we have a liberal proposer, the size of the permissible status quo region in configurations 2 and 3 corresponds to the distance between the median voter and the filibuster pivot, i.e., F-V. With a conservative proposer the distance between the median voter and the veto override pivot, O-V, measures the analogous interval and gives our second permissible status quo region. From configuration 4, the third status quo region is the traditional gridlock interval given by F-O. To identify the filibuster and veto pivots, we examined state legislative procedures to determine the veto override threshold and whether a state allows the filibuster and the threshold necessary to invoke cloture. Only thirteen of the fifty states allow filibusters, including five of the twenty-four initiative states. In states that do not, the median legislator effectively serves as the filibuster pivot. For the veto pivot, the legislative bound of the permissive status quo range depends on whether the governor or the veto threshold legislator lies closer to the median voter. Unfortunately, we lack data on gubernatorial ideal points and therefore must calculate the veto interval as the distance between the median voter and the veto override pivot. 13 For the California data, we initially calculate the veto interval as the distance between the median voter and the 66 th percentile legislator for the veto pivot. 14 Two rule changes, however, 13

potentially increase the filibuster interval for revenue bills. In June of 1933 voters passed Proposition 1, requiring a two-thirds vote to pass a budget with more than a 5% increase in spending. 15 In 1978, voters passed Proposition 13, which applied this same supermajority requirement to tax bills. These two supermajority requirements potentially mimic the possibility of a filibuster by requiring a supermajority to pass certain bills. We calculate the filibuster pivot using the 33 rd percentile legislator during the period 1979-2003. We use this filibuster measure in all analyses for the post Proposition 13 period since the 1933 change targeted only the budget and only under certain circumstances; whereas, Proposition 13 s tax requirement likely affects more bills and has spillover effects into other policy areas that entail raising revenue. 16 These calculations provide us with direct measures of two of the three points the filibuster pivot and the override pivot. For the third key point, no data as of yet provide sufficiently comprehensive coverage of the location of the median voter in a common space with legislators for our use here. Bafumi and Herron (2010) estimate the median voter at the national level with a battery of questions about legislative proposals; Masket and Noel (2012) create common space legislator and median voter ideal point estimates using legislator and district votes on referendums as a bridge. Unfortunately, neither approach currently provides the broad crosssectional and temporal coverage we require to test our model. The extant work on median voter estimates suggests a reasonable interim solution, however. To place the median voter, we measure the distance between the median voter and the veto and filibuster pivots with the distance between the median legislator and each pivot. While imperfect, we believe this provides a good proxy measure for our theoretical concept. While the median voter and median legislator will not necessarily have the same ideal point, our approach only requires a positive correlation between the relevant intervals: as long as V F 14

correlates with L F (and similarly for the veto interval), the latter will provide a useable proxy for our preferred measure. We subject this assumption to an extensive robustness check through a series of simulations. A second complication lies with identifying the correct preference configuration. Without knowing the median voter and proposer s ideal points, we do not know which of our measures of the permissive status quo interval to use. 17 In configurations two and three our model predicts that V F or V O measures the correct interval (depending on the location of the proposer); whereas, in the fourth configuration, O F captures it. But since V F + V O = F O whenever the median voter lies between O and F we can not include all three measures, since they would be perfectly collinear. Since V F matters with a proposer on the left and V O matters with a proposer on the right, our first approach includes both. Thus, if an observation corresponds to a configuration where one of them matters, our theory predicts a positive relationship; if it does not, our theory predicts no relationship. These expectations simply make it less likely that we would find a significant effect since in some cases the included interval will be incorrect and have no effect. Our second approach includes just the gridlock interval, which forms the permissive zone in case four (and is captured by the sum of the two separate intervals in the first approach). This effectively captures the two separate intervals if the data are roughly evenly split between left and right proposers in configurations two and three and has the added benefit of not requiring us to measure the median voter s ideal point. 18 Additional Control Variables In addition to our key independent variable the size of the initiative-permissive status quo intervals we also control for a number of other factors known to influence the frequency of 15

initiative usage across the states (see, e.g., Banducci 1998, Boehmke 2005b, and McGrath 2011). To protect empirical parsimony, we focus only on a small number of additional variables. First, in order to rule out the possibility that divided government rather than the gridlock interval per se leads to more initiatives, we utilize Klarner s (2003) data to construct a measure of whether one party controls all three branches of state government from 1959-2007. Second, demographic characteristics of a state also influence initiative use. Matsusaka and McCarty s (2001) model predicts that legislators in larger states will experience greater uncertainty about constituent preferences which then increases the frequency of initiatives. We add a control for state population from the Bureau of Economic Analysis (BEA) from 1929-2009. We also control for real per capita state income using BEA data. Third, liberal states tend to see more initiatives. This tendency may stem from the initiative s Progressive Era origins in the United States or from the fact that the initiative process tends to be utilized at a greater rate by citizen-oriented groups (Boehmke 2005b), which form in direct democracy states at a greater rate (Boehmke 2002). For our purposes, a greater number of potential proposers may exist in more liberal states; the presence of more proposers should increase initiative use. We control for state liberal citizen ideology using Berry, Ringquist, Fording, and Hanson s (1998) measure for 1960-2010. Finally, the rules governing ballot measure qualification explain much of the variation in initiative use in the American states. Previous studies find that signature requirements, distribution requirements, petition circulation periods, and whether states permit statutory or constitutional measures influence the number of measures (see, e.g., Boehmke 2005b; McGrath 2011; Hicks 2013). 19 We measure the signature requirement in percentage terms and average it across distinct requirements for statutory and constitutional measures if both are allowed. We 16

measure the circulation period with two variables: one for the maximum number of days in circulation and an indicator variable for those without a maximum. We account for the distribution requirement with a dichotomous indicator for whether the state requires thresholds for signatures across geographic subunits in addition to overall (e.g., the statewide signature threshold must be met in 40% of counties). Initiative Use across the American States We link our cross-sectional time-series data on initiative use to Clark et al. (2009) and Shor and McCarty s (2011) data on state legislator ideal points, from which we construct a measure of the filibuster and veto intervals for each legislative session. The first sample we use is the two legislative sessions (1999-2000 and 2003-2004) in the Clark et al. data (2009). 20 We also consider an extended sample for these data, in which we use initiative use data from 1997-2006, substituting the gridlock interval measure from 1999-2000 for the missing one in 1997-1998, the 2003-2004 value for 2005-2006, and the mean of the 1999-2000 and 2003-2004 values for 2001-2002. This quadruples our sample size, but likely introduces some error into our key independent variable. 21 We then use the Shor and McCarty data, which cover a longer time period. [Table 2 about here.] We present the results of our negative binomial models of initiative use across states in Table 2. We find generally strong support for our hypotheses, with the coefficient on the filibuster interval positive and significant in all models, the coefficient on the gridlock interval significant in two of three models, and the coefficient on the veto interval significant in one model. In all three cases, the standard errors decrease as the sample size increases and, notably, all variables achieve significance in the model with the greatest number of observations. First difference 17

calculations show that an increase from the 16 th to the 84 th percentile for the veto interval produces about 1.9 more initiatives per biennium in this model. The same change in the filibuster interval produces 0.2 more initiatives per biennium with the Shor and McCarty data and 0.4 to 0.7 more initiatives per biennium using the Clark et al. data. 22 The first difference for the gridlock interval ranges from 1 to 1.7 to 2.1 over the three models. Over the ten-year time period in our extended sample, states average 3 measures per biennium, making these sizable effects. Our other variables perform largely as expected. As expected, distributional requirements and greater signature requirements decrease the number of initiatives on the ballot while the number of circulation days, the population of the state, and the real income per capita have a positive effect. Initiative Use in California, 1911-2003 Our second test uses a purely time series data set on the number of ballot measures in California from 1911-2003 using Masket s first dimension DW-nominate estimates of legislator ideal points. We again use the legislative session as our unit of analysis and estimate a negative binomial model of the number of initiatives. [Figure 2 about here.] Figure 2 presents a smoothed, lowess plot of initiative use per biennium and the filibuster and veto intervals over time. Initiative use follows the familiar pattern of high frequency early in the twentieth century shortly after adoption, followed by a long fallow period and resurgence in use in the early 1970s. The filibuster interval follows a broadly similar trend, with a moderate but steady decline from 1911 to the late 1950s, followed by a small surge in the 1960s and a steep 18

increase in 1979 with the introduction of the supermajority requirement for budgets and tax measures in 1979. The veto interval trends with the filibuster interval until 1979, after which it experiences a steep decline. We now turn to a regression analysis to determine if this correlation remains with the addition of various control variables. Because most of our independent variables only have valid measures back to about 1960, we estimate two models. The first model includes the two permissive initiative intervals or the gridlock interval as the substantive independent variables and a cubic spline in order to model changes in other forces that we do not measure directly. We use a natural cubic spline with knots placed at each quintile. 23 Using a cubic spline with more knots should capture possibly complex changes in omitted variables over time, though robustness checks indicate that our results do not depend on this particular functional form. In the second model we add demographic and ideological control variables. Since this restricts our sample to the period 1961-2003, we account for complex changes stemming from time with a quadratic function of time rather than with splines. [Table 3 about here.] Table 3 presents the results for all four models. In the models with the separate permissive zones, the coefficient for the filibuster interval is always positive and significant. The veto interval coefficient is only significant in the full sample. In the two models that use the gridlock interval, we again find significance in both models. Calculations show the gridlock interval led to between one and five more initiatives per biennium prior to Proposition 13 in 1979 while the combination of the filibuster and increased gridlock interval led to five to fourteen more initiatives per year thereafter (see appendix). This change in legislative rules demonstrates handily the real effect of legislative gridlock on initiative use. 19

Robustness Check While we argued earlier that the distance between the median legislator and the pivots serves as a good proxy for the distance between the median voter and the pivots, we attempt to bolster this claim by summarizing the results of an extensive robustness check reported in detail in our supplemental appendix. To execute this test, we assume the existence of a representation function that maps district median voter ideal points into legislator ideal points. While we do not know the exact shape of such a function, we set the general structure based on reported results in the literature at the state (Masket and Noel 2012; Tausanovitch and Warshaw 2012) or federal level (Bafumi and Herron 2010). We then simulate district median voter ideal points from our observed legislator ideal points to impute a predicted state median voter ideal point. We assume that the relationship between the median voter and a district s legislator is generally linear with a slope near one, an intercept near zero, and an extremism bias such that a representative is more extreme. We add a random error term to capture variation around this relationship. We repeat this exercise over 1800 different combinations of the parameters of this representation function, calculate the two theoretical variables of interest, and run our regression model with these imputed values. In brief, this exercise indicates that our findings for both key variables prove quite robust and do not change sign or lose significance across broad combinations of changes, though the findings for the veto interval appear slightly more persistent. Not surprisingly, though, as the size of the random error increases, the results eventually weaken. Conclusion 20

Uniting standard spatial models of the initiative process with the pivotal politics model of legislative policymaking leads to a more nuanced understanding of each. We identify the conditions under which the initiative process breaks legislative gridlock, and alternatively, when the legislature cannot successfully preempt a ballot measure. Thus, we provide predictions about when the initiative might serve as the Progressives proffered gun behind the door via legislative policy moderation or direct voter decision making. Importantly, our theoretical results indicate that initiatives do not merely result from legislative gridlock; they can in fact create legislative gridlock that would not have existed otherwise. Initiatives do not arise simply because the status quo lies in the gridlock interval; rather, they arise because the legislature s optimal policy in the absence of the initiative lies in the gridlock interval. Pivotal actors exploit the threat of an initiative to force policy changes that make them better off, whether through the legislature or the ballot box. Ultimately, this exploitation can leave the median voter worse off than without the initiative process, meaning a state institution designed to express the public s wishes may actually thwart the public will. We find evidence consistent with our hypotheses across several state data sets, suggesting that our findings do not emerge purely due to the limitations of our data. Extensive robustness analysis also bolsters our results. As state data sets continue to improve, we believe it will be possible to test more precisely our predictions and yield further evidence of the success or failure behind one of the Progressives key reforms. Our model produces a number of novel theoretical and substantive implications for legislative and state politics. First, our results highlight the interdependence of political institutions and, consequently, the potential unintended consequences of reforming one institution on the other institutions in the state. Just as state legislative term limits 21

unintentionally limit the legislature s bargaining power with the governor (Kousser 2005), the response of pivotal legislative actors to the initiative process determines whether the initiative generates helps produce policy closer to the median voter s ideal point. Yet, multi-institutional studies that trace the dynamics between institutions are rare in American politics relative to single institution studies, particularly about legislative dynamics. Considering multiple institutions in tandem offers greater insight into how a rule change in one institution affects the other. For example, supermajority requirements may slow the legislative process and make it more cautious, but they also increase the permissive space for the initiative process. In turn, the initiative process can both decrease overall gridlock and change the balance of power between pivotal actors in the legislature. As Figure 2 shows, California s two-thirds majority tax rule introduced by Proposition 13 (i.e., the initiative process itself changing the gridlock interval!) resulted immediately in a huge expansion in the filibuster-median voter interval, which was promptly followed by a doubling in the number of initiatives. Second, our results identify nuance in translating the pivotal politics model to state legislatures. In states that allow initiatives, the initiative process can break the traditional legislative gridlock interval, leading to a smaller initiative-adjusted gridlock interval. Policies that formerly would have been gridlocked can now be moved, either through legislative preemption or through the ballot box. Further, the variety in state legislative rules affects the placement of the filibuster and veto pivots, and thus the applicability of congressional theory to legislatures more generally. As we note earlier, not all states allow for filibusters; similarly, a few states, such as Arkansas, place their veto override threshold at a simple, rather than a super, majority. From our initial application of the pivotal politics model to the states, additional variations in rules, such as supermajority vote requirements on particular types of bills (which 22

exist in at least nine states) and line-item and/or amendatory veto powers, could be added to further specify the identification of the filibuster and override pivots. 25 These observations speak to the generalizability of theories across institutions, as the application of pivotal politics to state legislatures (or more generally, comparative legislatures) is not always straightforward. Furthermore, a number of testable outcomes regarding legislative behavior should be pursued in future work. For example, differences in patterns of bill sponsorship, legislative vote margins, rolling of the legislative majority, and bill passage rates might all result from changes to the effective gridlock interval under the initiative process. Our theory could also inform models of the sequential process of legislative bargaining under initiative and non-initiative conditions. Other legislative rules, such as the thresholds required for discharge petitions in the U.S. states, could also place additional pivotal actors on the spatial spectrum. As state legislative data become more commonly available, these institutional variations to our basic initiative-gridlock model will become testable propositions. In the meantime, the model we present in this work demonstrates the importance of both institutional interplay in the creation of public policy and the importance of assessing the consequences of government reform. 23

References Bafumi, Joseph, and Michael C. Herron. 2010. Leapfrog Representation and Extremism: A Study of American Voters and their Members in Congress. American Political Science Review 104 (3): 519-542. Banducci, Susan A. 1998. Direct Legislation: When is it Used and When Does it Pass? In Citizens As Legislators: Direct Democracy in the United States, edited by Shaun Bowler, Todd Donovan, and Caroline J. Tolbert. Columbus: Ohio State University Press. Berry, William; Evan Ringquist; Richard Fording; and Russell Hanson. 1998. Measuring Citizen and Government Ideology in the American States, 1960-1993. American Journal of Political Science 42 (1): 327-48. Binder, Sarah A. 1999. The Dynamics of Legislative Gridlock, 1947-1996. American Political Science Review 93 (3): 519 33. Boehmke, Frederick J. 2002. The Effect of Direct Democracy on the Size and Diversity of State Interest Group Populations. The Journal of Politics 64(3): 827 844. Boehmke, Frederick J. 2005a. The Indirect Effect of Direct Legislation: How Institutions Shape Interest Groups Systems. The Ohio State University Press. Boehmke, Frederick J. 2005b. Sources of Variation in the Frequency of Statewide Initiatives: The Role of Interest Group Populations. Political Research Quarterly 58(4): 575 585. Brady, David W., and Craig Volden. 2006. Revolving Gridlock: Politics and Policy from Jimmy Carter to George W. Bush. Boulder, CO: Westview Press. Burden, Barry C. 2005. Institutions and Policy Representation in the States. State Politics and Policy Quarterly (4): 373-393. 24

Clark, Jennifer Hayes; Tracy Osborn; Jonathan Winburn; and Gerald C. Wright. 2009. Representation in U.S. Legislatures: The Acquisition and Analysis of State Roll Call Data. State Politics and Policy Quarterly 9 (3): 356-370. Cox, Gary and Mathew McCubbins. 1993. Legislative Leviathan: Party Government in the House. Berkeley: University of California Press. Cronin, Thomas E. 1989. Direct Democracy: the Politics of Initiative, Referendum and Recall. Cambridge: Harvard University Press. Gerber, Elisabeth R. 1996. Legislative Response to the Threat of Popular Initiatives. American Journal of Political Science 40(1): 99-128. Gerber, Elisabeth R. 1999. The Populist Paradox: Interest Group Influence and the Promise of Direct Legislation. Princeton: Princeton University Press. Hicks, William D. 2013. Initiatives within Representative Government Political Competition and Initiative Use in the American States. State Politics & Policy Quarterly 13(4): 471-494. Keele, Luke J. 2008. Semiparametric Regression for the Social Sciences. New York, NY: John Wiley & Sons. Klarner, Carl. 2003. The Measurement of the Partisan Balance of State Government. State Politics and Policy Quarterly, 3 (3): 309-319. Krehbiel, Keith. 1998. Pivotal Politics: A Theory of U.S. Lawmaking. Chicago: University Of Chicago Press. Lascher, Edward L. Jr; Michael G.Hagen and Steven A. Rochlin. 1996. Gun Behind the Door? Ballot Initiatives, State Policies and Public Opinion. The Journal of Politics 58: 760-775. 25

Lax, Jeffrey R. and Justin H. Phillips. 2009. Gay Rights in the States: Public Opinion and Policy Responsiveness. American Political Science Review 103 (3): 367-386 Lupia, Arthur; Yanna Krupnikov; Adam Seth Levine; Spencer Piston; and Alexander Von Hagen-Jamar. 2010. Why State Constitutions Differ in their Treatment of Same-Sex Marriage. The Journal of Politics 72: 1222-1235. McGrath, Robert J. 2011. Electoral Competition and the Frequency of Initiative Use in the U.S. States. American Politics Research 39 (3): 611-638. Magleby, David B. 1984. Direct Legislation: Voting On Ballot Propositions in the United States. Baltimore: Johns Hopkins University Press. Masket, Seth E., 2004. California Assembly Ideal Points, 1901-2003. University of Denver. http://mysite.du.edu/~smasket/research.html. Masket, Seth E. and Hans Noel. 2012. Serving Two Masters Using Referenda to Assess Partisan versus Dyadic Legislative Representation. Political Research Quarterly 65 (1): 104-123. Matsusaka, John G. 1995. Fiscal Effects of the Voter Initiative. Journal of Political Economy 103: 587-623. Matsusaka, John G. 2004. For the Many or the Few: The Initiative Process, Public Policy, and American Democracy. Chicago: Chicago University Press. Matsusaka, John G. and Nolan M. McCarty. 2001. Political Resource Allocation: Benefits and Costs of Voter Initiatives. Journal of Law, Economics, and Organization 17: 413-448. Romer, Thomas and Howard Rosenthal. 1978. Political Resource Allocation, Controlled Agendas, and the Status Quo. Public Choice 33: 27-44. 26

Romer, Thomas and Howard Rosenthal. 1979. Bureaucrats versus Voters: On the Political Economy of Resource Allocation by Direct Democracy. Quarterly Journal of Economics 93: 563-588. Shor, Boris and Nolan McCarty. 2011. The Ideological Mapping of American Legislatures. American Political Science Review 105 (3): 530-551. Skelton, George. Touting a Tax on Services. Los Angeles Times, November 24, 2011 (online edition). Tausanovitch, Chris, and Christopher Warshaw. 2013. Measuring Constituent Policy Preferences in Congress, State Legislatures, and Cities. The Journal of Politics 75 (2): 330-342.Waters, M. Dane. 2003. Initiative and Referendum Almanac. Durham, NC: Carolina Academic Press. 27

Table 1. Summary of Theoretical Predictions about When Initiatives can Occur in Equilibrium Configuration 1 Configuration 2 Configuration 3 Configuration 4 Location of the Median Voter V<F F<V<L L<V<O O<V Initiative range with left proposer No V<SQ<V+ V-F V<SQ<V+ V-F V+ V-O <SQ<V+ V-F Size of initiative range 0 V-F V-F F-O Initiative range with right proposer V- V-O <SQ<V- V-F V- V-O <SQ<V V- V-O <SQ<V No Size of initiative range F-O V-O V-O 0 Note. P represents the Proposer s ideal point, F the filibuster pivot s ideal point, L the median legislator s ideal point, O the veto override s ideal point, V the median voter s ideal point, and SQ the status quo ante. 28