Citation
Direct democracy and policy innovation in the American states : exploring the citizen initiative's impact on policy formulation through the lens of political competition

Material Information

Title:
Direct democracy and policy innovation in the American states : exploring the citizen initiative's impact on policy formulation through the lens of political competition
Creator:
Harp, K. W.
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Doctorate ( Doctor of philosophy)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
School of Public Affairs, CU Denver
Degree Disciplines:
Public affairs
Committee Chair:
Teske, Paul
Committee Members:
Martell, Christine
Gerber, Brian
Robinson, Tony

Notes

Abstract:
This dissertation explores the role played by direct democracy in the states regarding the state-level adoption of various social and economic policy innovations. Using a modified event history analysis, this dissertation explores two independent variables of interest – the simple presence of direct democracy in the states, and the difficulty of ballot placement for each state – against the single adoption of eight policy variables. It then explores the same independent variables in a repeating-events model, exploring the variable impacts on the multiple adoptions of tax-and-expenditure limitations and medical marijuana policies. In the modified event history analysis, both measures of direct democracy play statistically significant roles to increase the adoption of three policy variables – family cap exemptions, medical marijuana, and tax and expenditure limitations. Further, mandatory bicycle helmets for children and primary seatbelt laws are significantly impacted by both measures of direct democracy, but in these cases direct democracy acts as a deterrent to policy adoption. In the repeating-events model, both measures of direct democracy are significantly related to the adoption of repeating tax-and-expenditure limitation policy adoptions, but are not significant for the repeated adoptions of medical marijuana policies.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright K.W. Harp. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
DIRECT DEMOCRACY AND POLICY INNOVATION
IN THE AMERICAN STATES:
EXPLORING THE CITIZEN INITIATIVE’S IMPACT ON POLICY FORMULATION THROUGH THE LENS OF POLITICAL COMPETITION
by
K.W. HARP
B.S., University of Colorado Boulder, 2000 M.A., University of Colorado Denver, 2006
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Public Affairs Program
2017


This thesis for the Doctor of Philosophy degree by K.W. Harp
has been approved for the Public Affairs Program by
Paul Teske, Chair Christine Martell, Advisor Brian Gerber Tony Robinson
Date: December 16, 2017


Harp, K.W. (PhD, Public Affairs Program)
Direct Democracy and Policy Innovation in the American States: Exploring the Citizen Initiative’s Impact on Policy Formulation through the Lens of Political Competition Thesis directed by Associate Professor Christine Martell
ABSTRACT
This dissertation explores the role played by direct democracy in the states regarding the state-level adoption of various social and economic policy innovations. Using a modified event history analysis, this dissertation explores two independent variables of interest - the simple presence of direct democracy in the states, and the difficulty of ballot placement for each state - against the single adoption of eight policy variables. It then explores the same independent variables in a repeating-events model, exploring the variable impacts on the multiple adoptions of tax-and-expenditure limitations and medical marijuana policies.
In the modified event history analysis, both measures of direct democracy play statistically significant roles to increase the adoption of three policy variables - family cap exemptions, medical marijuana, and tax and expenditure limitations. Further, mandatory bicycle helmets for children and primary seatbelt laws are significantly impacted by both measures of direct democracy, but in these cases direct democracy acts as a deterrent to policy adoption.
In the repeating-events model, both measures of direct democracy are significantly related to the adoption of repeating tax-and-expenditure limitation policy adoptions, but are not significant for the repeated adoptions of medical marijuana policies.
The form and content of this abstract are approved. I recommend its publication.
Approved: Christine Martell
m


ACKNOWLEDGEMENTS
Thanks to my wife Victoria for her support and patience (mostly) during my completion of this dissertation.
Thanks to my two sons Henry and Owen for being awesome.
Thanks to my parents for their infinite support.
Thanks to Dr. Christine Martell for her persistence.
Thanks to my committee members - Drs. Paul Teske, Brian Gerber, and Tony Robinson - for years of support during my academic journey.
Thanks to my colleague Kegan Reiswig for STATA help.
IV


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION................................................1
II. LITERATURE REVIEW...........................................6
Direct Democracy............................................7
Political Competition......................................10
Policy Innovation..........................................17
Review of Empirical Literature.............................22
Literature Review Conclusion...............................28
III. RESEARCH QUESTION AND HYPOTHESES...........................31
IV. RESEARCH DESIGN............................................35
Phase 1....................................................36
Phase 2....................................................37
Explanation of Variables...................................39
V. RESULTS....................................................57
Phase 1 Analysis...........................................58
Phase 2 Analysis...........................................90
VI. DISCUSSION & CONCLUSION...................................106
Limitations and Future Research...........................113
Final Thoughts............................................116
REFERENCES..........................................................118
APPENDIX............................................................126
v


CHAPTER I
INTRODUCTION
In his seminal work on direct democracy in the American states, Matsusaka (2004) argues that the days of questioning the value of direct democracy - whether it is good or bad - will soon become irrelevant, “much like questions about whether democracy itself is a good or bad thing” (p. 128). Future research in direct democracy will turn toward understanding how it interacts with other democratic institutions and how these various institutions can effectively work together. To address these questions, new frameworks and theories must be developed to study the function of direct democracy in the states within America’s representative form of government.
One such framework to consider the citizen initiative’s role within the American republic is the competition framework. As outlined by Matsusaka (2004), the competition framework views the citizen initiative as an avenue for the creation of public policy, and it competes against the traditional political parties (Republicans and Democrats) in the creation of new laws. Just as Republicans and Democrats compete in the arena of public policy, with votes and public support as the prize, the citizen initiative must also compete for votes at the ballot box. Ultimately, the citizen initiative eliminates the monopoly that elected officials otherwise have over the lawmaking process.
In the private sector, product innovation is a necessary tool to gain or keep a competitive advantage over other firms in the marketplace. The focus of product innovation is to meet the evolving needs and expectations of consumers, the markets, and other stakeholders in the business world (Rainey, 2005). In a decentralized marketplace, individual firms and people experiment with new innovations to satisfy consumer demand
1


that is not possible in a monopoly. “Innovation is an absolute necessity to avoid becoming obsolete” (Rainey, 2005, p.12).
Applying this basic economic principle to policymaking, it follows that increased competition in the lawmaking process will promote greater levels of policy innovation (Matsusaka, 2004) as policy entrepreneurs are allowed entry into the lawmaking process. Where direct democracy does not exist, a policy entrepreneur must either run for elected office or successfully lobby an elected official to give attention to his policy innovation or idea. In states where the citizen initiative is available, access to the policymaking marketplace is open to any individual with the means to place a ballot question before the voters. While this may not include every citizen in a state, it no doubt opens the doors to allow significantly more policy entrepreneurs to participate in the creation of new laws.
At the time of publication, Matsusaka (2004) wrote that there exists no research viewing the citizen initiative through the competition perspective. This research is intended to contribute to the first empirical exploration of direct democracy’s impact on the policymaking process through the lens of a competition framework.
Throughout this dissertation, “innovation” is used to indicate a policy that is new to a state1. “Direct democracy” and “initiative” are used interchangeably to indicate that a state allows citizens to propose a new statutory law or constitutional amendment through a ballot measure or referendum2. Putting an initiative on the ballot is generally accomplished through a petition, requiring a certain number of valid signatures under different rules, which vary by state, to place a question on a ballot. This dissertation does account for varying state rules
1 See the Literature Review section for a more detailed definition.
2 Popular referendum, which allows citizens to approve or reject legislation passed by the legislature, are allowed in 24 states though compared to the initiative are rarely used (Initiative & Referendum Institute, 2011).
2


and regulations regarding the initiative process by accounting for the “difficulty of ballot placement” in each initiative state. Currently, there are 24 initiative states - 18 allow constitutional amendments and 21 allow statutory changes (Initiative & Referendum Institute, 2011) (see Appendix A).
While Oregon was the first state to see an initiative on the ballot in 1904, use of the initiative increased significantly after California passed the tax-cutting Proposition 13 in 1978 (Initiative Use, 2010). The 1990s saw a record 377 statewide initiatives throughout the country, of which 177 were approved. There were 420 initiatives in the 2000s (2000-2010), of which 178 were approved by voters (Initiative & Referendum Institute, 2017). Since 1904, almost 45 percent of all citizen initiatives in the United States were placed on the ballot in the last three decades alone (Initiative Use, 2010), which indicates a sustained popularity to use the initiative to drive public policy in the American states. Additionally, Coloradans have seen 215 statewide initiatives (through 2010), which lags behind only Oregon (355) and California (340) (Initiative & Referendum Institute, 2017).
Further, more than 70 percent of the American population now lives in a state or city where direct democracy is available. No state has ever repealed a law allowing for direct democracy, and states are adding the initiative at a rate of about one state per decade since WWII (Lupia & Matsusaka, 2004).
It is evident that use of the citizen initiative in policy formation is a permanent reality for almost half of the American states, but the role the initiative plays in policy innovation is largely untested. “Today political scientists understand that direct democracy is not only a control variable to be included in studies of state politics and policy but also a subject worthy of investigation in its own merit” (Smith & Tolbert, 2007, p. 417). For policymakers, policy
3


entrepreneurs, academics, and civic-minded citizens, understanding what role the initiative plays in policy formation and how it interacts with other democratic institutions is fast becoming a prerequisite to fully understanding the American policy process.
The overriding research question for this dissertation is: Does the citizen initiative promote policy innovation at the state level? To address this question, two hypotheses are tested: 1) direct democracy is positively related to policy innovation in the states, and 2) the difficulty of placing a question on a statewide ballot will be negatively related with higher levels of policy innovation.
The test these hypotheses, this research tests multiple variables across eight various policy adoptions in 48 states. There are two models in this research. 1) Policies are treated as single-event policy adoptions (the first policy adoption in each state). Once a state has adopted the policy in question, it drops from the data set. 2) For two policy variables, the second phase explores repeating policy adoptions (analyzing each policy adoption event even if a prior policy adoption has already occurred).
The simple presence of direct democracy as a dichotomous variable is used to test the first hypothesis. The second hypothesis is tested by employing a state-by-state Qualification Difficulty Index (Bowler & Donovan, 2004), which considers various factors required to place a question on the ballot, (e.gsignature requirements).
This dissertation proceeds as follows. The next chapter explores existing literature in direct democracy, political competition, and policy innovation, and provides an overview of the relevant empirical literature. Chapter III details the research questions and hypotheses. Chapter IV covers the research design for Phase 1 and Phase 2 of this research and provides an explanation of all variables used. Chapter V details results of both statistical analyses
4


individually and in comparison. Finally, Chapter VI explores the research findings’ implications for theory and policy practice, while also detailing research limitations and areas for future research.
5


CHAPTER II
LITERATIVE REVIEW
While the words “democracy” and “democratic” are often used to define American government, the federal government was intentionally set up as a republic - a system of government where people elect representatives. Madison (2012/1787) wrote about the dangers of a pure democracy, which would provide no cure for “the mischiefs of faction” (p. 26). A republic “will be more consonant to the public good than if pronounced by the people themselves” (p. 27). The support for a republican form of government was adopted in the United States because those who wrote the Constitution “had a vivid Calvinistic sense of human evil and damnation and believed with Hobbes that men are selfish and contentious” (Hofstadter, 1989, p. 5). Further, “Throughout the secret discussions at the Constitutional Convention it was clear that this distrust of man was first and foremost a distrust of the common man and democratic rule” (Hofstadter, 1989, p. 6).
While not ignoring the democratic elements within America’s government, a pure democracy was intentionally avoided.3 While direct democracy has been avoided at the federal level, America’s inherent distrust of government helped propel the implementation of the citizen initiative at the state level beginning more than a century ago. Direct democracy in the states predates women’s suffrage and the federal income tax, just to name two accepted American institutions. After South Dakota became the first state to adopt the initiative in 1898, 19 more states followed suit by 1918 as part of the Progressive movement. To this day, opinion polls consistently show strong support for the initiative process at the state and local level (Matsusaka, 2004).
3 See Hofstadter (1989) for a more complete historical overview.
6


The insertion of the citizen initiative at the state level plays a vital role in the creation of state policy and laws. Three relevant themes are relevant to support the general hypothesis that greater competition in the policy arena through the inclusion of direct democracy will result in greater policy innovation at the state level. The following sections explore the literature in direct democracy, political competition, and policy innovation. This is followed by a review of the relevant empirical literature.
Direct Democracy
Despite the first ballot initiatives appearing on Oregon’s 1904 ballot, the academic literature on direct democracy has just recently come of age in the last 25 years (Smith & Tolbert, 2007), and much of the early literature was purely descriptive or normative (Lupia & Matsusaka, 2004). This literature has focused on a variety of topics related to direct democracy, including its impact on voter preferences, minority interests, policy implementation, civic engagement, voter competencies, interest groups, and more (Smith & Tolbert, 2007).4 This review focuses on the junction between direct democracy and public policy.
While most research exploring direct democracy’s impact on public policy are almost exclusively focused on individual or a limited number of policy areas, it is clear that the initiative does impact public policy in the states. There is, for example, a large amount of literature available on tax-and-expenditure limitations (TELs) and the initiative, which blossomed after California voters passed Proposition 13 in 1978. Wallin (2004) found that access to the ballot box through citizen initiative was a better predictor of TEL passage in a state than any other variable. This was especially true where legislatures were dominated by
4 See Smith & Tolbert, 2007; and Lupia & Matsusaka, 2004, for a more complete picture of all literature on direct democracy.
7


a single party. Mullins and Wallin (2004) found that TEL expansion was more rapid in states where the citizen initiative was available. Not only did access to direct democracy play a vital role in the expansion of America’s tax revolt, but the initiative also led to more stringent TELs. New (2001) and Krol (2007) both found that TELs passed by initiative were more effective at limiting government growth than TELs passed by a state’s legislature.
In other fiscal policy areas, Blume, Muller and Voigt (2009) found welfare spending to be lower in countries with mandatory referendums. Wagschal (1997) found direct democracy to be an effective constraint on expansive spending and taxation as countries with some form of direct democracy have significantly lower levels of public expenditure and taxation.
Literature has also explored the initiative’s impact on policy areas outside the fiscal arena. For example, Gerber (1996) found that initiative states are more likely to pass parental-consent laws (concerning abortion) closer to the median voter preference. But the ballot box was also a predictor of passing more “extreme” urban growth boundaries in California (Gerber & Phillips, 2005). Initiative states are also more likely to adopt term-limit policies for legislators (Tolbert, 1998; Bowler & Donovan, 1995), campaign finance restrictions (Pippen, Bowler & Donovan, 2002), and laws concerning English as the official language (Hero & Tolbert, 1996).
Matsusaka (2000) concluded that the initiative process is not inherently a government-reducing institution, but rather it is a tool that brings the legislature in line with voter preferences of the time through increased information. Matsusaka (1995, 2000, 2004), the lead scholar in this policy area, found that in the latter half of the 20th century, state spending was about 4 percent lower on average in initiative states (1995). To the contrary,
8


however, when looking at the first half of the 1900s, initiative states had higher state and local expenditures and revenues (Matsusaka, 2000). Much of the literature on direct democracy has focused on fiscal policy, since 1950, with most literature finding that initiative states had lower overall spending than non-initiative states.
The literature on direct democracy clearly indicates that the initiative does impact public policy at the state level. “States with the initiative, especially those that use it frequently, tend to have different policies than non-initiatives states” (Smith & Tolbert, 2007, p. 422). But other questions then emerge - how does public policy differ across states and what is the initiative’s role in this policy divergence?
Literature shows (Lupia & Matsusaka, 2004) that the initiative can impact policy in both a direct and indirect fashion. The direct model simply implies that the initiative impacts policy when voters approve a measure. The indirect effect is a measure of a change in legislative behavior resulting from the potential or the threat of an initiative (Lupia & Matsusaka, 2004; Gerber, 1996). The impact on policy adoption from the direct effect is many times larger than the indirect effect (Matsusaka, 2014).
Bowler and Donovan (2004) also show that direct democracy should empirically be measured as more than a simple dummy variable, which assumes that all initiatives states are the same. In reality, each initiative state has different rules for placing a question on the ballot. While many of the procedural activities are similar, differences do exist among the states. Notably, the number of signatures required to place a question on the ballot varies by state. Also, some states have geographic requirements (e.g., a certain number of signatures must be collected in each congressional district). Does the difficulty of placing a question on the ballot impact the use of the initiative process? Bowler and Donovan (2004) conclude it
9


does matter, as those states with an easier process (less restrictions) had more initiatives on the ballot than in states where the process is more difficult.
Laws concerning the initiative in Colorado make it the second-easiest state for citizens to place a question on the statewide ballot (during the review period of this research), tied with California and North Dakota (Bowler & Donovan, 2004). In 2016, however, Colorado voters passed Amendment 71, which requires, for constitutional measures, a certain number of signatures be collected in each of the state’s 35 senate districts to place a question on the ballot and a supermajority requirement (55%) to approve the measure. Amendment 71 increases the difficulty for a citizen to place a question on the ballot. Until Amendment 71 passed, only Oregon’s initiative process allowed for easier access to the ballot (Bowler & Donovan, 2004).
This literature review shows that the presence of a citizen initiative process in a state does impact public policy. The literature fails, however, to show from a macro perspective exactly how and why the initiative impacts policy. Additionally, the cupboard is empty in viewing the phenomenon through a theoretical perspective beyond the direct and indirect impacts of the initiative. This dissertation begins to fill the cupboard by exploring the initiative’s impact on policy innovation in the states through a political competition theory.
Political Competition
While the Progressive movement is generally credited with the expansion of direct democracy in the states, Smith and Fridkin (2008) argue that political competition was the catalyst for the adoption of the very policy innovation being explored in this research - the adoption of the initiative in the American states. They found that interparty legislative
10


competition, weaker party organizational strength, and the presence of third parties were the most powerful predictors of a legislatively enacted ballot measure to allow direct democracy.
Downs (1957) applied economic theory to the political world to explain party politics, ideologies, and voter behavior. Just as firms seek to maximize profits and consumers their utility, parties seek to maximize political support while voters seek to maximize benefits from the government. By “political support,” Downs explicitly intends vote maximization and thus election to office. “Just as firms that do not engage in the rational pursuit of profit are apt to cease to be firms, so politicians who do not pursue votes in a rational manner are apt to cease to be politicians” (Kelley, 1965, p. x). Downs (1957) assumes that both parties and voters are rational beings. He also notes that policies are created to win elections, and denies that candidates seek office to pursue virtuous policy endeavors.
To gain election (or re-election), a political party must seek to maximize votes by moving party ideology to the middle of voter preference (Downs, 1957), which is the root of the median voter theorem5 (Llavador, 2000). This often results in two party platforms converging toward the middle and ultimately coming to resemble one another (Downs,
1957). However, where uncertainty and/or polarization exist, party platforms may evolve in opposite directions. Regardless of where party ideology ultimately rests, the motivation is derived from the rational goal of gaining or keeping power through the election process.
This economic model is predicated on competition between political parties.
Stigler (1972) further applied the principles of economic competition to the political world. He asserted that the more competition over repeated election cycles that exists between the parties, the more responsive the political system becomes to the desires of the
5 Looking at policy and how it relates to median voter preferences in not explored in this dissertation, but is further discussed in the section on future research.
11


majority voter. But he also found, however, that electoral results (how large or small the electoral victory) impact the responsiveness. If one party achieves, for example, only 49 percent of the seats in a legislative chamber, then the probability that a single party will achieve its entire platform before the next election is reduced. If a party has a strong majority over a long period of time (think of a monopoly of power), however, then that party is better able to control the policy agenda where there is less risk in departing from median voter preferences. In either case - a monopoly of power or electoral competition - Stigler (1972) found evidence that political competition (or lack thereof) impacts how near or far a party’s policy agenda is to the median voter. Crain and Tollison (1976) found additional support for Stigler’s findings. Bueno de Mesquita, Morrow, Siverson, and Smith (2001) found that prolonged control and dominance by a single party stifles competition and leads to cronyism and corruption. Conversely, they also found political competition leads to economic growth and prosperity. Kessler (2005) inserted information asymmetries into political competition. Where information asymmetries exist, policies formed in a representative system were further away from the median voter than those policies created through direct democracy. This indicates that direct democracy helps reduce information asymmetries - the initiative process signals voter preferences to lawmakers. This allows for competing lawmakers to produce more innovative policy ideas based on voter preference.
Berry and Berry (1990) found that states are more likely to adopt innovative policies in an election year. This indicates that election-year competition does produce an incentive to promote innovative policy ideas. This idea was anecdotally supported from the Colorado 2012 legislative session, which preceded a presidential election. Two separate democratic lawmakers, who ultimately ran for Congress and lost in the 2012 election, ran bills that
12


“strike us as being more about their Democratic sponsors’ political ambitions than any pressing need on the part of the citizens of Colorado” (Bad Bills, 2012, p. 10B). One bill would have prevented discrimination of motorcycle riders based on their dress. The other would have withheld pay from lawmakers if the legislature failed to pass a budget on time. Both failed. “Perhaps these bills [were] signs of electoral desperation, because they surely aren’t the good public policy Coloradans have the right to expect” (Bad Bills, 2012, p. 10B).
Other scholars have expanded the political competition literature with more complex additions to the theory. For example, Tavits (2007) finds that welfare maximizing policy leads to vote increases at the polls, but policy shifts in core belief areas (typically social- or value-based issues such as minority rights or social justice) lead to feelings of betrayal among a party’s base and will result in repercussions on Election Day. Persico, Pueblita, and Silverman (2011) explored intraparty competition through factions within each party and found that elected officials from non-competitive districts are more successful in maintaining policies (even if inefficient) and capturing monetary advantages (pork-barrel spending). Rogers and Rogers (2000) applied political competition to the executive branch and found that a close race in a gubernatorial election acts as a check against bigger government (measured as revenue and expenditure relative to personal income and population). Additionally, the median preference of voters must be adjusted to take into account voter abstention, which could lead to a winning policy bearing little resemblance to the preferences of the median voter (Llavador, 2000).
The literature thus far does indicate that parties do compete for votes, and that the level of competition does impact party platform, and thus public policy. While the literature
13


on political competition is expanding its theoretical base, the inclusion of direct democracy in the states as a variable of interest is largely absent in this literature.
The exception is Bowler and Donovan’s (2006) study on the impact of direct democracy on political parties, though it does not explicitly reflect on direct democracy’s impact on policy formation. They found that the presence of direct democracy in a state leads to a more restrictive legal framework and institutional constraints for political parties -such as the implementation of term limits, civil service regulation, limitations of campaign contributions, and the ability to recall elected officials - thus regulating and weakening their autonomy. This weakening effect gives direct democracy more relative power at the very least. Direct democracy was also associated with weaker party organizations as measured by elements of hierarchy, level of control over candidates, and patronage, among other measures. They also found that political parties are beginning to adapt to the presence of direct democracy by indirectly funding initiative campaigns, notably those that may drive a wedge in the opposition party’s base (Republicans promoting gay-marriage bans, for example). This finding indicates that political parties may be competing with, or possibly leveraging, the citizen initiative in some way.
To further bolster this claim, Boehmke (2002) found that the presence of direct democracy in a state increases that state’s interest group population by 17 percent over noninitiative states. Polsby (1984) suggested that interest groups typically operate through ‘“experts’ who focus routinely ... on the cultivation of policy-related ideas” (p. 167). If one accepts that interest groups are policy entrepreneurs who do impact the public policy agenda, then it implies that an increase in interest groups will impact the public policy agenda, and thus create more competition for elected officials. In non-initiative states, a policy
14


entrepreneur must run for office or lobby an elected official. In initiative states, however, where the legislative policy-making monopoly is broken, citizens and interest groups have direct access to the ballot. “Legislatures are gradually becoming eclipsed as the primary creators of public policy, and in some states they already have assumed a secondary role” (Matsusaka, 2005, pp. 157-158) when forced to react to successful initiatives.
Mintrom and Norman (2009) define policy entrepreneurs as “advocates of policy change” (p. 649). While they found that imminent legislative action minimized the role played by policy entrepreneurs, they did find entrepreneurs to be effective in cases where existing policy-change systems are inadequate to address real or perceived challenges. In short, policy entrepreneurs are effective when challenging the status quo. “A key part of policy entrepreneurship involves seizing such moments to promote major change” (p. 650). While these policy entrepreneurs must be willing to invest time, energy, reputation and money, allowing direct access to the ballot lessens the investment and risk to challenge the policy status quo (Minstrom & Norman, 2009).
Mintrom and Vergari (1996) argue the exploration of policy entrepreneurship helps explain periods of dynamic policy change. The entrepreneurs do so by identifying policy needs and innovative solutions, bearing financial and emotional risk, and coordinating individuals and organizations. “Policy entrepreneurs can be thought of as doing for the policymaking process what business entrepreneurs do for the marketplace. That is to say, policy entrepreneurs serve to bring new policy ideas into good currency” (p. 422). Identified by their actions and not by their positions, “entrepreneurs are not content simply to push for changes at the margins of current policy settings. Rather, they seek to change radically current ways of doing things” (p. 423). Mintrom and Vergari (1996) also note that during the
15


policymaking process, policy entrepreneurs often work outside the policy subsystem, which is to say that entrepreneurs look for ways to push their policy agenda outside the traditional legislative scope. In states where more policymaking avenues exist outside the legislative system, clearly the more opportunity exists for entrepreneurs to push an agenda and create dynamic and innovative policy changes.
While the literature generally supports the conclusion that increased competition leads to an increase in policy ideas and policy adoption, within the context of direct democracy there are at least two conceivable ways in which the indirect effects of direct democracy (still increasing competition) could act as deterrents to policy adoption. First, it is conceivable that the threat of a reactionary initiative to a legislatively adopted policy would dissuade a legislature from adopting the policy in question. The literature on the indirect effects of direct democracy, however, tend to focus on policies that are legislatively adopted based on the threat of an initiative (Matsusaka, 2014), rather than policies that are not adopted due to the threat of a reactionary ballot measure. Second, Matsusaka (2014) defines a secondary indirect effect of direct democracy as one that provides new information (a communication or signaling effect) to a legislature. Under this measure of the indirect effect, he found that in initiative states where a conservative ballot measure was defeated at the polls, that state was 21 percent less likely than a non-initiative state to adopt the policy in question in future years.
This provides a bit of evidence that political competition through direct democracy can decrease policy innovation. There is otherwise little to no evidence, scientific or anecdotal, that policymaking competition resulting from direct democracy actually decreases the potential for policy adoption. As such, this research continues under the notion that
16


political competition would, at least generally if not overwhelmingly, support an increase in policy innovation.
The literature in this section indicates three main themes: 1) political parties do compete for votes through policy platforms, 2) direct democracy does have a direct impact on traditional political parties, and 3) the initiative significantly expands both the number of policy entrepreneurs who are able to impact the formulation of public policy and the avenues available to them. It logically follows from these three findings that direct democracy in the states does play a competitive role in the American policy process.
Policy Innovation
In the literature, the generally accepted definition of a policy innovation is a program or policy that is new to the state adopting it without regard for whether the policy has already been adopted in other states (Walker, 1969; Berry & Berry, 2007; Boehmke & Witmer,
2004). This demands a distinction be made between policy innovation and policy invention, which is defined as the adoption of a policy never seen before in any state. Further, policy innovation is not a value statement, meaning that innovation, as used throughout this dissertation, in no way implies a given policy is good or bad - only that it is a new policy for the state.
Matsusaka (2005) holds that direct democracy is increasingly responsible for shaping the direction of state policy. He identified a bevy of high-profile state policy issues that emerged through the initiative process; including affirmative action, illegal immigration, medical marijuana, TELs, and school vouchers. “Now try to compose a comparable list of important policy developments that emanated from legislatures; it is difficult to identify more than a handful. ... It does not seem an exaggeration to say that policy innovation is now
17


being driven as much by voter initiatives as by legislatures and governors” (pp. 161-162). While this particular article was descriptive in nature, Matsusaka’s assertion exemplifies his conceptual idea (2004) that the citizen initiative creates competition in the lawmaking process leading to policy innovation in the states. There is limited empirical work elsewhere to support Matsusaka’s nascent competition framework.
Kotsogiannis and Schwager (2006) found that in a federal system with political competition for office, there is a strong incentive to “experiment” in innovative policy positions. Candidates use innovative policy ideas to signal their ability to the electorate. For example, a legislator with aspirations for higher office will want to promote innovative policy ideas to be well received in future elections. This provides evidence that a federal system will generate more innovation than a unitary system. Adding to the policy innovation incentive, a federal system like the U.S. also allows states to act as natural “policy laboratories” where competition exists at the state level (Kotsogiannis & Schwager, 2006). This paper supports Matsusaka’s (2004) competition framework in that it signals an elected policy maker’s acknowledgment of voter preferences.
Further support for the competition framework can be drawn from Williams (2003), who found that policy innovation in criminal justice policy was predicted more by the presence of and advocacy by policy entrepreneurs than by state characteristics. In considering Boehmke’s (2002) finding that initiative states have a 17 percent higher interest group population, it follows that initiative states with more interest groups (policy entrepreneurs) and avenues for policy change could have more policy innovation. Obinger (1998), while concluding that the effects of federalism and direct democracy in Switzerland reduce political power related to the expansion of the Swiss welfare state, acknowledged
18


“this slow-down of decisionmaking does not necessarily affect the innovation capacity and the quality of legislation. ... Federalism and the popular initiative are actually important institutional channels favouring innovation.” (p. 260).
Most of the academic discussion on policy innovation has taken place in the literature on policy diffusion (Berry & Berry, 2007; Boehmke & Witmer, 2004; Karch, 2006). Outside of the diffusion literature, policy innovation is little explored. But the literature that does exist provides a foundation from which to build this research.
In Walker’s (1969) article, which first stimulated interest in the process of state policy diffusion (Welch & Thompson, 1980), policy innovation was the dependent variable, with various sociodemographic factors, political characteristics, and other factors acting as the independent variables. To operationalize his dependent variable, Walker looked at the rate of adoption of 88 policies across the states to give each state an “innovation score.” He found that wealthy, higher populated, and industrial states had the highest innovation scores, but he did not account for states with direct democracy (the presence of the citizen initiative was one of the 88 policies considered for the innovation score, but was not considered as an independent variable that may impact the rate of policy innovation). He did measure political competition by the competitiveness of gubernatorial elections and found that it did not impact a state’s policy innovation score. A better measurement of party competition within the legislature can be attained for this study following the ideas of Stigler (1972),
Crain and Tollison (1976), and Bueno de Mesquita et al. (2001). Walker (1969) does concede that “it would seem that parties which often faced closely contested elections would try to out-do each other by embracing the newest, most progressive programs and this would naturally encourage the rapid adoption of innovations” (p. 885). While he did not find
19


empirical support for this assertion, he seems to suggest that competition is not fully captured by party competition when he wrote “party competitiveness does not seem to be consistently related to the innovation score, at least as it is measured here” (p. 886).
Walker (1969) also found that legislative competition based on a measure between urban and rural representation impacted innovation scores from 1930-1966. While he did not measure legislative competition as a function of party, the finding still indicates that legislative competition leads to policy innovation. This finding by Walker is supported by historical evidence. In the Civil War years, more than 80 percent of Americans lived in rural areas (Matsusaka, 2004). By 1920, a majority lived in urban areas, thus creating a strong demand for new and different services such as water sanitation, sewage treatment, and garbage collection. This increased demand from a growing urban population for government services is a unique measure of legislative competition appropriate for that time. While it pits two competing and diverse interests (urban versus rural demands), there are parallels to the concept of two competing interests in the form of party competition leading to policy innovation. Further, today’s urban areas and rural areas tend to support legislative candidates from different parties. Consider that Denver and Boulder’s state legislators tend to be Democrat, while representation of rural Colorado is dominated by Republican legislators. It is not a stretch to speculate rural and urban competition in the early part of the 20th century is similar to contemporary times when legislative competition is measured as a function of partisanship.
Boehmke and Skinner (2012a) revisited Walker’s (1969) study and increased the number of policy innovations to 189, all treated as non-repeating events. They found that states with higher populations, greater per capita income, and higher rates of urbanization
20


were associated with higher innovation scores. This study did include the presence of the citizen initiative and found no consistent effect on a state’s level of innovativeness. The authors, however, did not focus on the specific time frame that could have isolated the period from 1978 to present when initiative use reached its peak (their study explored policy adoptions beginning in 1912). Further, they treated the initiative variable as dichotomous, which did not take into account the varying institutional rules across the states.
In their study of innovation in e-govemment, Tolbert, Mossberger, and McNeal (2008) also found that wealthier states were more likely to engage in policy innovation, as were states with higher levels of education and legislative professionalization. Political, economic and social characteristics are also important factors leading to policy innovation in internal determinants models (models looking at a state’s internal characteristics) (Berry & Berry, 2007). Welch and Thompson (1980) also found that policies with federal monetary incentives lead to higher rates of policy innovation in the states.
In the policy innovation literature, the concept of party competition is only briefly explored, most notably by Walker (1969) and Boehmke and Skinner (2012a). Further, while internal characteristics play a role in policy innovation, external influences from other states must also be considered. Past research on policy innovation has traditionally explored either internal characteristics within a state or diffusion effects (external influences) from other states to explain a policy change, but rarely were both captured in a single study (Berry & Berry, 1990). As explored further in the next section, diffusion literature is the leading arena in which to discuss policy innovation in the states, though none of the literature expressly considers the potential role of direct democracy.
21


This research incorporates ideas presented in the diffusion literature and adds the element of the citizen initiative as a possible determinant of policy innovation. While there exists a hint of direct democracy in the innovation literature, this dissertation puts the initiative on center stage in exploring policy innovation in the states.
Review of Relevant Empirical Literature To remedy a methodology shortfall of failing to capture both internal characteristics of a state and external diffusion influences, Berry and Berry (1990) developed event history analysis (EHA) for the social sciences, which accounts for both internal and external influences on state policy adoption. Additionally, this approach captures influences that may impact policy adoption in each state in each year to account for changing circumstances in each individual state.
Berry and Berry (2007) proposed the following unified model of state government innovation that captures both internal determinants and the external influence of policy diffusion. Their proposed model is as follows:
ADOPTi:t=f (MOTIVATIONi,t, RESOIJRCES/OBSTACLESi?t, OTHER-POLICIES,,t, EXTERNAL,,,)
In their model (Berry & Berry, 2007), the unit of analysis is the individual state that will potentially adopt a policy. ADOPT, the dependent variable, is the probability that an individual state (z) will adopt a given policy in a given year (I). ADOPT is a function of internal and external variables (see Table 1). The first three subsets of variables relate to internal determinants. MOTIVATION contains variables that indicate the motivation of public officials, such as the proximity to an election, public opinion, and electoral competition. RESOURCES/OBSTACLES is concerned with obstacles to innovation and resources that are available to overcome those obstacles. Examples are a state’s level of economic development, legislative professionalism, and the presence of policy entrepreneurs.
22


Table 1
Explanation of Bery and Berry (1990) Model Variables
Policy Adoption is a function of... Internal Characteristics of a State Diffusion effects
Model variable subset Motivation Resources/Obstacles Other Policies External
What it is Captures variables concerning the motivation Obstacles to innovation and resources available to Other policies exist that might impact likelihood of Diffusion effects across states
of public officials overcome policy adoption
Examples Proximity to an election, electoral competition Direct democracy and difficulty of ballot placement. Economic development, legislative professionalism, presence of policy entrepreneurs. E.g., if the policy is gay marriage, the presence of civil unions in a state may impact policy adoption. Number of contiguous states with a given policy; number of states within a fixed region with a given policy.


The presence of direct democracy and difficulty of ballot placement variables fall in this category. OTHER-POLICIES would indicate whether or not other policies exist that might increase or decrease the likelihood of the policy in question being adopted in any given year. EXTERNAL is concerned with diffusion effects across state boundaries and is not concerned with any internal determinants within a state.
Berry and Berry (1990) analyzed pooled cross-sectional time-series data via event history analysis to test their model of state lottery adoptions. They used a discrete time model, meaning the time element was divided into years (any fixed period of distinct time units would be a discrete time model). As such, the data are one observation per state for each year the state might possibly adopt the given policy. Event history analysis is explicitly structured to look at trends over time, while a simple regression or probit analysis would not account for error structures over time. The dichotomous dependent variable is coded as “0” if the state has not adopted the policy, and then coded as “1” in the year the policy was adopted. Or, in Berry and Berry’s (1990) study, for example, once a lottery was adopted, a state is no longer “at risk” of adopting the policy - it is a non-repeating event. Further, because the variable is dichotomous, Berry and Berry (1990) used probit maximum likelihood estimation. The estimates of the coefficients for the independent variables measure the predicted impacts of these independent variables on the likelihood a state will adopt the specific policy innovation (Berry & Berry, 2007). “.. .The coefficient estimates can, in turn, be used to generate predictions of the probability that a state with any specified combination of values on the independent variables will adopt the policy in a given year” (pp. 243-244). Further, the model estimates the probability of adoption when any of the independent variables change while holding other variables constant.
24


Since their 1990 study on state lottery adoptions, some refinements to event history analysis have been made. In the early literature, the most common measure of diffusion (external influences) was to quantify the number or percentage of contiguous states adopting a policy (Berry & Berry, 2007). More recently, fixed-region diffusion has been utilized. It first defines regions of the country and measures the number or percentage of states from a region that have adopted a policy to account for diffusion across non-contiguous states (Mooney & Lee, 1995; Andrews, 2000; Allen, Pettus, and Haider-Markel, 2004), as states can learn from other states, even when there is no common border. Allen, Pettus, and Haider-Markel (2004) also measured vertical influence to capture any effects of federal policies.
The OTHER-POLICIES variables, which capture influence from previously adopted policies within a state, were not included in early event history analysis studies (Berry & Berry, 2007). More recently, Balia (2001) measures the impact of previously adopted HMO measures that were complementary to the policy in question (HMO Model Act). And Soule and Earl (2001) studied whether a state was more likely to adopt hate crime laws if other hate crime legislation had previously been approved. Further, Seljan and Weller (2011) found that states were less likely to propose a TEL - through the legislature or citizen initiative - if a state in close geographic proximity had already rejected a TEL.
Buckley and Westerland (2004) offered three fundamental criticisms of Berry and Berry’s (1990) event history analysis model, but also offered recommendations to overcome those problems. The three major concerns over the EHA model are duration dependence, functional form, and corrected standard errors.
25


Duration Dependence
At a very minimum, scholars must account for duration dependence - that is, scholars must account for the impact that the passage of time has on the likelihood of policy adoption - in EHA models (Buckley & Westerland, 2004). The Berry and Berry (1990) EHA model, as with other EHA models, assumes a flat hazard rate; EHA models assume that the probability of policy adoption does not change over time. To correct this problem, one must relax the flat-hazard rate assumption. The most basic solution to this problem is to simply add a linear time counter to the model, adding one year to the measurement for each year in duration. The result is the following revised model:
ADOPTr,t=f (MOTIVATION,t, RESOURCES/OBSTACLESu, OTHER-POLICIES^, EXTERNAL^ TIME COUNTER,t.)
This approach still assumes that the risk of policy adoption as a function of time changes at a constant rate, but this revised model is superior to the previous model that did not account for time’s impact on the adoption of policy.
Functional Form
While logit and probit analyses are still appropriate, they should not be the preferred functional form for EHA modeling (Buckley & Westerland, 2004). Buckley and Westerland (2004) assert that a complementary log-log function (cloglog) can remedy this problem. In EHA, states are removed from the dataset once a given policy has been adopted, therefore creating a dataset with many more 0s than Is. Using the cloglog function instead of probit or logit, Buckley and Westerland (2004) find that the hazard rate approaches 1 (policy adoption) quicker and “suggests that the cloglog function may be more theoretically appropriate for rare event discrete EHA” (p. 102). In other words, it is preferred to use the cloglog function
26


for EHA because it accounts for the disproportionately high number of Os (no policy adoption) in EHA’s dichotomous dependent variables.
Corrected Standard Errors
The final criticism of traditional EHA modeling is that it remains possible that reported standard errors are incorrect (Buckley & Westerland, 2004). In general linear regression models (cloglog included), a requirement is an assumption that data are independent across time and in space; in EHA models, “this assumption is clearly violated” (p. 104) as EHA depends on correlation over both time and space. The solution is to use clustered standard errors, which relaxes the assumption of independent observations in the data (Buckley & Westerland, 2004). One fix for this problem is to cluster standard error based on region, as defined by the U.S. Census Bureau, to account for regional similarities among the states. This corrects for bias in EHA standard errors caused by spatial autocorrelation.
Repeating Events
Berry and Berry’s (1990) look at lottery adoptions was based on a policy event that was non-repeating. In other words, once a state adopted the policy in question, there was no further exploration in that particular state as it was no longer “at risk” of adopting the policy. In the Berry and Berry model, the only notable event is the first event, which makes an assumption that “the first event will be representative of all events” (Jones & Branton, 2005, p. 430). This is problematic as “analyzing only the first event discards information unnecessarily” (Jones & Branton, 2005, p. 430). To account for policy events that can be repeated after the initial adoption, Jones and Branton (2005) propose a Cox Duration Model with a conditional gap time model modification - they call it the stratified Cox model of
27


repeated events. In this model, once a state passes a policy, instead of disappearing from the data set, it becomes at risk for a second event. The duration time counter resets to 1 after each event. This model is explored further in the research design section below.
Literature Review Conclusion
From this literature review, three major themes must be recognized. The first is the finding that direct democracy does impact public policy in various capacities at the state level but limited empirical evidence shows how. The second important theme is party competition - the literature clearly indicates that political parties compete against each other for votes and electoral victory. Within this theme it should also be recognized that the use of the initiative, gaining in popularity, has increasingly broken the policy-making monopoly impacting both parties and policy directly and indirectly. Finally, the literature on policy innovation suggests that political competition is one of many possible explanatory factors leading to policy innovation.
Combined, these themes indicate that direct democracy in the states could be a competitive factor in the formulation of public policy. Coupled with basic economic ideas on product innovation, it logically follows that direct democracy, through increased competition against the traditional parties, will enhance policy innovation in the American states.
As an anecdotal example, consider the passage of California’s tax-cutting Proposition 13 in 1978. This proposal was considered an outlandish idea at the time of its passage (Matsusaka, 2004) and was dismissed by the established parties and political experts, yet it was approved at the polls by a 65-35 margin. Proposition 13, followed by Massachusetts’ Proposition 2'A in 1980, is largely credited as setting off the “tax revolt” of the 1970s and
28


1980s (Resnick, 2004). Currently, 46 states have some form of tax and expenditure limitation (TEL) in place at the statewide level (Amiel, Deller, & Stallman, 2009). While it cannot be known if the TEL movement would have taken off without the citizen initiative, it is clear that the initiative brought TELs to the forefront of public policy in the states for multiple decades.
Finally, the empirical models to evaluate policy innovation and diffusion have advanced significantly since Berry and Berry (1990) laid the foundation for using event history analysis in the study of policy innovation. Newer models and modifications account for the duration of time’s impact on potential policy adoption and favor other analysis tools over probit and logit. Other refinements account for spatial autocorrelation that was present in early EHA models and allow for the modeling of multiple-event policies. These more advanced models and methods continue to be refined, but the relevant empirical models in existing literature are capable of producing results for this research that are more empirically rigorous than the original Berry and Berry model. These more advanced models also allow for testing theory in a new way, allowing for the insertion of various measures of direct democracy to test a competition theory alongside well established variables, such as external influences (diffusion), internal factors within a state, and the impacts of time passage.
Having established that direct democracy affects public policy at the state level, that political competition exists and shapes policy agendas, and that political competition is one possible explanation for policy innovation, this research now moves to the exploration of direct democracy’s impact on policy innovation. As direct democracy increases the number of players available to affect change policy, there is more competition in the policy arena. This research attempts to quantify the initiative’s impact on policy innovation in the states
29


and shed some light on the role of direct democracy’s impact on policy formulation at the state level. As such, this research hypothesizes that the existence of direct democracy in the states, via increased political competition, will impact policy innovation at the state level.
30


CHAPTER III
RESEARCH QUESTION AND HYPOTHESES
The overarching research question is: Does the citizen initiative promote policy innovation at the state level'? Two hypotheses (see Table 2) will be tested.
• Hypothesis 1: Direct democracy, through increased political competition, is positively related to policy innovation in the states.
• Hypothesis 2: The difficulty of placing a question on a statewide ballot, with the magnitude of competition decreasing as access to the ballot is more difficult, will be negatively related with higher levels of policy innovation.
Table 2
Explanation of Hypotheses
Operationalization of
Independent Variables of Independent Variables Dependent Variables
Li terest of Interest (Policy Innovations) Expected Relationship
Hypothesis 1 Presence of Direct Dictotomous no = 0 yes = 1 Adoption of 8 Policy Positive
Democracy in a State Innovations
Hypothesis 2 Difficulty of Initiative Process Interval: Qualitication Difficulty Index* Adoption of 8 Policy Innovations Negative
Note. *From Bowler and Donovan (2004)
Each hypothesis is tested using a variety of policy innovations from two policy subsets at the state level: 1) matters of social policy, and 2) matters of fiscal policy. In exploring two policy subsets, this research allows one to potentially draw distinctions between different types of policies and the independent variables that may or may not influence policies within the subsets. Within the social policy subset, this dissertation uses medical marijuana policies, mandatory bicycle helmets for minors, hate crime laws, and primary seat belt laws. For fiscal policies, this research explores strategic planning for economic development, state enterprise zones, family cap exemptions for welfare policy, and tax and expenditure limitations (see Table 3).
31


Table 3
Dependent Variables
Social Policies Fiscal Policies
Policy First Adopted Last Adopted States with Policy Policy First Adopted Last Adopted States with Policy
Symbolic Medical Marijuana Policy 1978 2008 31 Strategic Planning Economic Development 1981 1992 34
Mandatory Bicycle Helmets Minors 1992 2007 21 State Enterprise Zones 1981 1992 38
Hate Crime Laws 1978 1994 33 Family Cap Exemptions (welfare) 1992 1998 21
Primary Seat Belt 1984 2004 21 Tax and Expenditure Limitations 1976 1994 26
Note. Data fromBoehmke and Skinner (2012b)
Each policy was selected from Boehmke and Skinner’s (2012b) study on policy innovation, which looked at more than 180 policy innovations. In selecting these eight specific policies, several criteria were used. First, use of the initiative ballooned after 1978’s Proposition 13 in California, so policy innovations from 1978 to present day were isolated to capture the period in American history when initiative use has been at its peak (Initiative Use, 2010). TEL policy (first adopted in 1976) fails this criterion, but TEL literature indicates most TELs were adopted after California’s Proposition 13 in 1978. The first hate crime and medical marijuana laws were adopted in 1978, so the first adoptions would not have been influenced during the peak days of initiative use (not to say that the policy was not influenced by the initiative at all). But again, most of these laws were adopted after 1978. For the remaining five policy variables, the first adoption took place in 1981 or later.
Second, it was preferred that policy adoption occurred over at least a 10-year span so as to exclude trendy punctuations in policy adoption. This allows the research to look at more sustained policy adoptions over time. Adoptions of family cap exemptions over 21
32


states occurred only during a seven year period, and it is the only policy selection outside of this criterion. To satisfy other concerns as noted above, it was necessary to include this exception. Third, it was also preferred that at least 15 states had adopted any given policy so that the sample size was sufficient. Fourth, fringe or extreme policies were not selected as to avoid extremely partisan and contentious issues. These might include gay marriage bans or parental notification laws for abortions by minors.
This variety of policies over two policy subsets allows for the hypotheses to be evaluated across different policy areas in the specified time frame of interest. This study also controls for political bias and other factors.
This dissertation fills several research gaps. As detailed in the literature review, a majority of EHA and policy innovation research has focused on single policies. Further, the two studies (Walker, 1969; Boehmke & Skinner, 2012a) that took a holistic approach to studying innovation in the states failed to account for several areas of interest: There was no detailed look at how policy innovation may have varied in different time frames or in different policy subsets; neither looked at the citizen initiative specifically during the time frame when initiative use reached its peak; Walker (1969) did not account for the citizen initiative, while Boehmke and Skinner (2012a) treated the initiative as a simple dichotomous variable; and both studies treated all policy innovations in a similar dichotomous fashion without accounting for repeating events.
This research contributes to the literature by:
• Looking at the initiative’s impact on policy innovation through a competition framework;
33


• Providing a deeper analysis of the initiative’s impact during a time frame when competition was maximized through higher use of the initiative;
• Using a modified event history analysis that controls for the passage of time;
• Exploring multiple policies over two separate policy subsets to differentiate the effects the initiative may have in different policy areas;
• Excluding punctuated diffusion by looking only at policies adopted over an extended period of time;
• Treating the initiative as a dichotomous variable indicating the presence of direct democracy, but also taking a more in-depth account into the varying rules and laws of each state’s initiative function;
• Providing an in-depth look at two policies where repeated adoptions were the norm across the states and over multiple decades.
• And exploring explanatory variables to policy innovation through the lens of political competition created by the presence and form of direct democracy in the states.
The research in this dissertation relies heavily on the work of previous scholars in the policy innovation field, but also fills in many research gaps until now left unexplored. As scholars, lawmakers and practitioners work to fully understand policy innovation in the states and the citizen initiative’s impact on that policy innovation, this dissertation provides fresh perspective from new angles.
34


CHAPTER IV
RESEARCH DESIGN
This dissertation answers the research question by exploring four policies each within two policy subfields - fiscal policy and social policy - for a total of eight policies. STATA is the primary tool used for analysis. The individual state is the unit of analysis with a fixed time period of one year, which is the accepted standard for research on policy innovation in the states. There are two main phases of analysis: first, an event history analysis is completed for each of the eight policy variables; second, a repeating-event analysis is completed for the tax-and-expenditure limitations variable and the medical marijuana variable. For each phase, data were collected for 48 states from 1976 through 2015 and does not consider the District of Columbia or any U.S. territory. Illinois is excluded because the initiative process there can only be employed to modify the organization of the state legislature, and fiscal initiatives are not allowed (Matsusaka, 2004). Wyoming is excluded for several reasons, most notably because the state’s signature requirements make it an outlier, and because there are subject restrictions on revenue and appropriation measures (Matsusaka, 2004). Hawaii and Alaska are excluded from this study when exploring diffusion effects using the contiguous-state measure of diffusion, but are included in the western region when using the fixed-region diffusion measure. The contiguous state measure was used in early event history analyses to account for external influences on policy innovation. Later measures used the fixed-region diffusion measure, again to account for external influences while accounting for the idea that policy learning is not limited to only neighboring states.
35


This section first explores the two phases of this study and explains the model used for each phase. This is followed by an in-depth discussion of all the variables used for each phase.
Phase 1
For phase one of this dissertation, event history analysis is employed on each policy individually to determine if the presence of a citizen initiative and the ease of placing a question on the ballot are explanatory factors for each policy innovation. This phase employs the Berry and Berry (1990) model of event history analysis with several modifications derived from Buckley and Westerland (2004), as discussed in detail in the empirical section of the literature review. Specifically, these modifications include adding a linear time counter, replacing logit and probit with a cloglog function to account for the disproportionately high number of Os in the dichotomous dependent variables, and the clustering of standard errors by region. By using EHA, this study accounts for both internal and external influences on policy innovation. This approach captures influences that may impact policy adoption in each state in each year to account for changing circumstances in each individual state.
Adding a linear time counter to the Berry and Berry model (Buckley & Westerland, 2004), the model for phase 1 is:
ADOPT,,t=f (MOTIVATION^, RESOIJRCES/OBSTACLESi?t, EXTERNAL,,t, TIME COUNTER^)
The “Other Policies” component, not included in early EHA models, was removed from this phase as identifying potential policies that may impact the policy variable in question would be somewhat arbitrary and subjective. As utilized in previous studies, for example, Balia (2001) measured the impact of previously adopted HMO measures on the adoption of the HMO Model Act. Also, Soule and Earl (2001) studied hate crime policy
36


adoption as a function of whether or not some form of hate crime law already existed in the state. Identifying and researching every potential policy in every state over 40 years that may impact the adoption of the eight policy variables researched here would have been time prohibitive, while being somewhat arbitrary and providing limited value in the context of this research given the number of independent variables already utilized. This minor limitation is overcome in the Phase 2 repeating events model, as it by definition accounts for similar policies already in place for two policy variables.
Following Buckley and Westerland (2004), a complementary log-log function (cloglog) is employed for analysis rather than standard probit or logit, which were used in early EHA models given the dichotomous dependent policy variables. Further, Phase 1 corrects for bias in EHA standard errors caused by spatial autocorrelation (Buckley & Westerland, 2004) by clustering standard error based on region, as defined by the U.S.
Census Bureau, to account for regional similarities among the states.
Phase 2
Jones and Branton (2005) focus on a failure of event history analysis treating all policies as non-repeating events. In their model on lottery adoption, Berry and Berry (1990) did focus on a policy that generally occurred only one time. However, some policy events are considered to be repeating. Notably, as analyzed in phase 1, Boehmke and Skinner (2012a) treat tax and expenditure limitations and medical marijuana laws as non-repeating events. In other words, a state drops out of the data set once the first policy adoption occurs. However, some states have adopted multiple TELs and medical marijuana laws over time.
For example, Colorado alone has four TELs currently in place - the 1977 Kadlecek Amendment, which imposes a 7 percent annual spending limit; the 1982 Gallagher
37


Amendment, which caps residential property taxes; the 1991 Arveschoug-Bird Amendment, a limit on general fund expenditures; and the Taxpayer Bill of Rights, approved by voters in 1992 (Martell & Teske, 2007). Also, some states have passed multiple marijuana laws over the years, with many states first approving a medical marijuana law and then refining the law to add restrictions and/or expansions. When a policy innovation study only looks at the first policy adoption, it ignores information and provides potentially faulty conclusions (Jones and Branton, 2005).
In looking at obscenity legislation across the states in both single-event and repeating-event models, Jones and Branton (2005) found statistically inconsistent findings. In this light, Phase 2 of this dissertation looks at TEL policy and medical marijuana policy using Jones and Branton’s repeated events model. This provides further insight into the initiative’s impact not only on policy adoption, but on repeated adoptions in the same policy area. This allows examination of the role of direct democracy in ongoing policy change.
Jones and Branton (2005) use a stratified Cox model to study policy innovation, where the baseline hazard rate is left unspecified. Further, unlike other models of EHA, the Cox model is easily extended to handle repeating events. The Cox model is as follows:
h,(t) = ho(t) exp (P’x)
where ho(t) is the baseline hazard function and // ’x are the covariates and regression parameters. “Although the baseline hazard rate is not directly estimated from the data, the covariate parameters are, using a partial likelihood approach” (Jones & Branton, 2005, p. 424). To modify the model to account for repeating events, no state drops out of the dataset once the first policy is adopted. Using Colorado’s multiple TEL laws as an example (see Table 4), the conditional time counter measures the number of years since the last policy
38


event. It starts at 1 in the first year and increases by one each year until another policy adoption. Then, the conditional time counter resets with a 1 in the year following another Table 4
Repeating Events Example
Year State Event Overall time (years) Risk sequence Conditional time (years)
1976 CO 0 1 1 1
1977 CO 1 (Kadlec ek) 2 1 2
1978 CO 0 3 2 1
1979 CO 0 4 2 2
1980 CO 0 5 2 3
1981 CO 0 6 2 4
1982 CO 1 (Gallagher) 7 2 5
1983 CO 0 8 3 1
1984 CO 0 9 3 2
1985 CO 0 10 3 3
1986 CO 0 11 3 4
1987 CO 0 12 3 5
1988 CO 0 13 3 6
1989 CO 0 14 3 7
1990 CO 0 15 3 8
1991 CO 1 (Arveschoug-Bird) 16 3 9
1992 CO 1 (TABOR) 17 4 1
Note . Modified from Jones and Bran ton (2005); Data fromMartelland Teske (2007)
policy adoption. The risk sequence indicates the ordering of policy events. It starts at 1. To model repeating events, the modified Cox model is stratified by the event number (risk sequence column) to preserve the ordering of events, and it defines time in the duration data set as the conditional gap time (conditional time column). Combined, these modifications create a conditional gap time Cox model (Jones & Branton, 2005). This is the model used for Phase 2.
Explanation of Variables
For both Phase 1 and Phase 2, all dependent and independent variables are measured in each state in each year from 1976 through 2015. They are summarized in Table 5.
39


Table 5 All Variables
VARIABLE SYMBOL OPERATIONALIZATION SOURCE Expected Relationship
DEPENDENT (Policy Innovation)
Tax and Expenditure Limitations TELs 0=not adopted l=adopted Waisanen, 2010; New, 2010, Martell & Teske, 2007 n/a
State Enterprise Zones EnterZone 0=not adopted l=adopted Wilder & Rubin, 1996 n/a
Family Cap Exemptions (welfare) FamCap 0=not adopted l=adopted Karch, 2007 n/a
Strategic Planning Economic Development EconDev 0=not adopted l=adopted Berry, 1994 n/a
Medical Marijuana Policy MedMar 0=not adopted l=adopted “Medical Marijuana Overview,” 2015 n/a
Hate Crime Laws HateCrime 0=not adopted l=adopted Grattet, Jenness & Curry, 1998 n/a
Primary Seat Belt PrimSeat 0=not adopted l=adopted NHSTA, 2006 n/a
Mandatory Bicycle Helmets Minors KidHelmet 0=not adopted l=adopted Bicycle Helmet Safety Institute, 2016 n/a
INDEPENDENT
1 for first year, 2 for second year, etc. For phase 2 of
T ime Duration TD this study, the counter will reset after each policy n/a Positive
occurrence.


Table 5 (Cont.) All Variables
VARIABLE SYMBOL OPERATIONALIZATION SOURCE Expected Relationship
EXTERNAL (Diffusion)
Contiguous state diffusion cs Percentage of surroundings states with the policy in question Self-calculated Positive
Fixed-region diffusion FR Percentage of states within a region with the policy in question Self-calculated usingU.S. Census region definition Positive
MOTIVATION
Proximity to gubernatorial election GE Number of years away from the next election The Green Papers web site Negative
Partisanship PS Republicans control governorship, either chamber of legislature. (0-3) NCSL, Wikipedia None
Consecutive gubernatorial years of single party GPC Consecutive number of years a single party has controlled the governor’s office. NCSL, Wikipedia Negative
Consecutive years a single party LMC Consecutive number of years that a single party has NCSL Negative
controls both chambers controlled both chambers of the legislature.
Consecutive years party has controlled gov and legislature Consecutive number of years that a single party has
CPC controlled both chambers of the legislature and the governor’s office. NCSL, Wikipedia Negative
RESOURCES/OBSTACLES
Presence of Direct Democracy DD 0 = Initiative not available 1 = Initiative is available Initiative and Referendum Institute Positive
Captures how difficult it is to place a question on the
Difficulty of ballot placement for initiative questions DDi ballot in each state. Considers signature requirements and geographic regulations. Higher number indicates a higher level of difficulty. Interval scale 0-6 (non- Index from Bowler & Donovan (2004) Negative
initiative states receive a 10 score).
Rural population RP As a percentage of total state population U.S. Census via Iowa State U niversity Unknown
Wealth W M edian household income U.S. Census Bureau Positive
Total population P Population U.S. Census Bureau Positive
U.S. Census Bureau (State Tax
Economic conditions EC State revenue by population Collections divided by Positive
Population)
Legislative professionalism LP Index captures measurements of salary and benefits, length of sessions, staff, and resources Index from Squire, 2007 Positive
Education EDU % of state population with college degree U.S. Census Bureau Positive


Dependent Variables
Policy innovation is captured in the eight policy innovations. These dependent variables are tested as a function of the presence of the citizen initiative, the difficulty of ballot placement for each state’s citizen initiative, and a host of other variables derived from the literature (see Table 5).
The eight dependent variables are captured in ADOPTi}t. They are coded in a dichotomous fashion: “0” if a policy has not been adopted in any given year, and “1” if the policy has been adopted in any given year (see Table 6).
Tax and Expenditure Limitations
Tax-and-expenditure limitations (TELs) come in many forms - spending limits, revenue limits, requiring voter approval or legislative supermajorities for tax increases - but all attempt to limit government spending and/or government growth. Using updated figures regarding states with a tax-and-expenditure limitation, 33 states have these policies in place (Waisanen, 2010; New, 2001). About half of TELs are constitutional, with the other half statutory provisions. They have been passed through initiative, referenda, or legislative approval, and can be either statutory or constitutional (Waisanen, 2010).
Four states passed TELs prior to 1976. These TELs are not considered here.
42


Table 6
First Policy Adoptions by State
STATE YEAR OF FIRST ADOPTION
TELs EnterZone FamCap EconDev MedMar HateCrime PrimSeat KidHelmet
ALABAMA 1988 1999 1995
ALASKA 1982 1991 1998 1982 2006
ARIZONA 1978 1988 1994 1991 2010
ARKANSAS 1984 1993
CALIFORNIA 1979 1986 1994 1992 1996 1978 1993 1987
COLORADO 1977 1986 1990 2000 1988
CONNECTICUT 1991 1982 1995 2012 1990 1986 1993
DELAWARE 1978 1985 1995 1992 2011 2003 1996
FLORIDA 1994 1981 1995 1986 1989 1997
GEORGIA 1982 1993 1996 1993
HAWAII 1978 1987 1986 2000 1985 2001
IDAHO 1980 1981 1983
INDIANA 2002 1983 1994 1990 1998
IOWA 1992 1986 1990 1986
KANSAS 1982
KENTUCKY 2000 1982 1989 2006
LOUISIANA 1979 1981 1989 1995 2002
MAINE 2005 1987 1999 1999
MARYLAND 1982 1994 2014 1990 1997 1995
MASSACHUSETTS 1986 1995 2012 1983 1994
MICHIGAN 1978 1986 2008 1988 2000
MINNESOTA 1983 1989 2014 1989
MISSISSIPPI 1982 1984 1995 1994 1996
MISSOURI 1980 1983 1988
MONTANA 1981 1991 2004 1989
NEBRASKA 1992 1994
NEVADA 1979 1984 1985 1998 1989
NEW HAMPSHIRE 2013 1990 2006
NEW JERSEY 1976 1984 1992 1990 2010 1990 2000 1992
NEW MEXICO 2007 1986 2007
NEW YORK 1987 1987 2014 1982 1984 1994
NORTH CAROLINA 1991 1995 1987 1991 1985 2001
NORTH DAKOTA 1998
OHIO 2006 1982 1983 1987
OKLAHOMA 1985 1983 1998 1987 1987 1997
OREGON 1979 1986 1987 1998 1981 1990 1994
PENNSYLVANIA 1983 1982 1995
RHODE ISLAND 1992 1991 2006 1982 1996
SOUTH CAROLINA 1980 1987 1995 2005
SOUTH DAKOTA 1996 1993
TENNESSEE 1978 1984 1996 1987 1989 2004 1994
TEXAS 1978 1988 1993 1985
UTAH 1989 1988
VERMONT 1986 2004 1989
VIRGINIA 1984 1994 1990 1994
WASHINGTON 1979 1998 1981 2002
WEST VIRGINIA 1987 1989 1987 1996
WISCONSIN 2001 1988 1987
43


State Enterprise Zones
State-initiated enterprise zones (EnterZones) are specific geographic areas that provide tax subsidies to businesses and are meant to spur development through tax incentives and regulatory relief (Wilder & Rubin, 1996). This research collected policy adoption data from Wilder and Rubin (1996). While 39 states have adopted enterprise zones6, not all enterprise zones are currently active.
Family Cap Exemptions
Family cap exemption policies (FamCap) exempt a certain number of children from welfare family caps, which deny welfare aid to children bom into families already receiving money through a TANF-supported welfare program (Gutierrez, 2013). Many states exempt a certain number of children up to eight from this cap based on certain criteria. For example, many states allow exemptions in cases of rape or incest, while California also allows an exemption in cases of failed contraception (Karch, 2007). This research analyzes 19 states that have adopted this policy, using Karch (2007) as the source for states with policy adoptions.
Strategic Planning (Economic Development)
Strategic planning for economic development (EconDev) is defined as “a management process that combines four basic features” (Berry, 1994). First, a clear statement of an organization’s mission. Second, identification and recognition of external constituencies and their assessment of the agency. Third, a delineation of the agency’s goals and objectives in a 3-5 year plan. Fourth, strategic development to achieve those goals and objectives. If these features were met, a strategic planning policy was considered to be
6 Including the District of Columbia
44


adopted. Berry found 24 states had policy adoptions fitting these criteria. Due to the survey methodology, this research uses the policy adoption data set as provided by Berry (1994).
While the survey identified nine areas of strategic development, this research focuses on strategic planning for economic development.
Medical Marijuana
Medical marijuana (MedMar) is defined as a policy that allows marijuana for medical use based on data from the Marijuana Policy Project (“Medical Marijuana Overview,” 2015). Twenty-two states have adopted a qualifying policy. This is more restrictive than the medical marijuana definition used by Boehmke and Skinner (2012b), which included state policies that were only a simple recognition of the benefits of medical marijuana. Policy adoption data were derived from the Marijuana Policy Project (“Medical Marijuana Overview,” 2015).
This policy variable, along with tax-and-expenditure limitations, is used in the repeating events policy analysis. Please see below for more details regarding medical marijuana and repeating policy adoptions.
Hate Crime Laws
Simply put, hate-crime laws (HateCrime) are the “criminalization of hate-motivated intimidation and violence” (Grattet, Jenness & Curry, 1998, p. 286). The growth of hate-crime laws derived from the perception of increased racial, ethnic, and religious forms of conflict, among other categories, in the 1970s. Through the late 1980s and 1990s, states began to push legislating hate-crime laws (Grattet, Jenness & Curry, 1998). Data were collected from Grattet, Jenness and Curry (1998). California was the first state to pass a hate-crime law, and this research analyzes 32 states with hate crime policies. Across the
45


states, these laws take the form of criminal penalties to penalty enhancements on existing law.
Primary Seat Belt Laws
A primary seat belt law (PrimSeat) allows law enforcement to give a driver a citation for not wearing a seat belt. Updating the Boehmke and Skinner (2012b) statistics used in their paper, there are 25 states plus Puerto Rico and the District of Columbia with primary seat belt laws (NHSTA, 2006). Data are collected from the National Highway Traffic Safety Administration (NHSTA, 2006). Twenty-four states carry a secondary enforcement provision, which allows a citation to a driver for not wearing a seat belt only if another violation occurred. These secondary laws are not included in this research.
Mandatory Bicycle Helmets for Minors
Mandatory bicycle helmet laws for minors (KidHelmet) are laws that require helmet use for minors while riding a bicycle. From the same source as Boehmke and Skinner (2012b) but updated, there are 22 states with mandatory bicycle helmet laws for minors (Bicycle Helmet Safety Institute, 2016). The age limit for mandatory helmet use varies from those under 12 years of age to under 18, though some local laws cover all ages. All of these laws are included regardless of the age considerations. Some states also have passenger-only helmet laws, which are not included. The year of policy passage here includes laws for minor riders, not for minor passengers. Three states with passenger laws - California, Massachusetts, and New York - also have laws covering riders.
Independent variables
The TIME COUNTER variable (TD) in the modified Berry and Berry model for Phase 1 is a nominal measurement for each year in duration, coding “1” for the first year and
46


adding 1 unit for each year of observation. In Phase 2, this variable is captured again by starting at 1 and increasing one unit for each year passed. However, the conditional time counter used in Phase 2 resets after every policy adoption or policy change.
External Variables
These variables capture the diffusion effects on policy innovation. This research captures both contiguous-state diffusion (CS) and fixed-region diffusion (FR). For the former, the variable is captured by counting the percentage of contiguous states that have adopted the policy in question, necessarily excluding Hawaii and Alaska. Using a series of formulas in Excel, a percentage of states with the relevant policy were calculated for each state. For fixed-region diffusion, the states are divided into regions as defined by U.S.
Census Bureau (Census Regions and Divisions of the United States, 2016). Using Excel formulas, the percentage of states in each region that have adopted the policy was calculated. Motivation Variables
These variables capture internal characteristics of a state; specifically they indicate the motivation of public officials to pass a policy innovation. In this study, MOTIVATION is captured by several partisan and election-related factors.
Walker (1969) found an increase in legislative rural-urban competition was positively related to innovation. He also concedes that parties that face tight elections would be more apt to promote policy innovation. Single-party dominance of both chambers in the legislature make it more likely that a party is able to control the policy agenda (Stigler, 1972; Crain & Tollison, 1976), creating an easier path to policy innovation. However, prolonged political dominance has also been found to reduce political competition (Bueno de Mesquita, Morrow, and Smith, 2001) and therefore policy innovation would be expected to decrease.
47


To capture these phenomena, this research measures the proximity to gubernatorial elections (GE). These data were collected from The Green Papers (All-Up Chart of Governors by Election Cycle, 2012). It also captures partisanship using a 0-3 scale for Republican control (PS), with 1 point each for controlling the governor’s office, the lower legislative chamber, and the upper legislative chamber. These data were collected from the National Conference of State Legislatures (State Partisan Composition, 2016) and from Wikipedia data regarding the state history of governors by state (List of Current United States Governors, 2016). This study also measures prolonged party dominance by capturing the consecutive number of years a single party has controlled the governorship (GPC), both chambers of a legislature (LMC), and both at the same time (CPC) using the same data sources. Prolonged dominance is expected to reduce competition.
Resources/Obstacles Variables
The independent variables of interest - the presence of direct democracy (DD) and the difficulty of ballot placement (DDi) - are captured in RESOURCES/OBSTACLES. The simple presence of direct democracy in a state is a dichotomous variable. These data are collected from the Initiative and Referendum Institute (2011).
The measure for difficulty of ballot placement is derived from an index in Bowler and Donovan (2004), which gives each initiative state a “Qualification Difficulty Index” score between 0 and 6, with 0 representing the least difficult requirements. Non-initiative states are scored a 10, following Bowler and Donovan (2004). Each state has different signature requirements for placing a question on the ballot, and some states require a certain number of signatures to be gathered in specific geographic locations within the state. Further, some states limit the time available to collect signatures (Bowler & Donovan, 2004). Banducci
48


(1998) and Matsusaka and McCarty (2001) found that stricter signature requirements reduce the number of initiatives in a state. It makes sense, then, that stricter signature requirements will reduce competition and therefore policy innovation.
These variables also look at other internal characteristics of a state. As legislative competition not only exists between the parties, but also between urban and rural representation (Walker, 1969), a state’s overall rural population (RP) is captured as a percentage of total state population. Rural population data were collected from the U.S. Census Bureau via Iowa State University (Urban Percentage of the Population for States, Historial, 2010). Data were available by decade (1970, 1980, 1990, 2000, and 2010), so, for example, 2000 figures were used in the data from 2000 through 2009.
Wealth by median household income (W) is a measure of median household income for each state. Data were collected from the U.S. Census Bureau (Historical Income Tables: Households, 2016). Data for all years were not available. As such, data figures from 1969 were used for 1976 through 1978. Data figures from 1979 were used from 1979 through 1983. Annual figures were available from 1983 forward.
Total state population (P) was again captured from data from the U.S. Census Bureau (Population and Housing Unit Estimate, (2016). Population figures were available for each year in this research.
Economic conditions (EC) in the states were captured by tax revenue per capita.
First, U.S. Census Bureau data regarding state historical tax collections were captured (State Government Tax Collections, 2016). These data were available for all years. The tax revenue figures were then divided by the population figures to attain a measure of tax revenue per capita.
49


It has been found that states with higher levels of legislative professionalism contributed to policy innovation in e-govemment (Tolbert, Mossberger, & McNeal, 2008). Therefore, this dissertation captures each legislature’s level of professionalism (LP) using an index by Squire (2007), which is based on measures of salary and benefits, length of legislative sessions, and availability of staff and resources. Squire (2007) provides ranks for 1979, 1986, 1996, and 2003 for each state. The 1979 score is used for the data years 1976 through 1985, the 1986 score from 1986 through 1995, the 1996 score from 1996 through 2002, and the 2003 score from 2003 through 2015.
Finally, this research uses a measure of education in each state based as a percentage of the state population with at least a bachelor’s degree (EDU). Data were available by decade from the U.S. Census Bureau (A Half-Century of Learning: Historical Census Statistics on Educational Attainment in the United States, 1940 to 2000: Detailed Tables, 2016; American Community Survey, 2010 1-Year Estimates, Table S. 1501, 2016). Data from 1970 were used from 1976 through 1979, data from 1980 were used from 1980 through 1989, data from 1990 were used from 1990 through 1999, data from 2000 were used from 2000 through 2009, and data from 2010 were used from 2010 through 2015.
Dependent Variables for Repeating Events Model
Again, Phase 2 of this dissertation explores policy adoptions of tax-and-expenditure limitations and medical marijuana laws using a repeating events model. Unlike Phase 1, where all policy adoptions are treated as single events and a state drops from analysis once the first policy adoption occurs, the repeating events model considers all policy adoptions of an issue from 1976 to 2015. Table 7 details the number of policy adoptions by policy and by state. Table 8 provides a vertical timeline of policy adoptions by year.
50


Table 7
Repeating Event Adoptions by State - All Adoptions by Year
Tax and Expenditure Limitations Medical Marijuana
State Count of Adoptions Year(s) of Adoptions State Count of Adoptions Year(s) of Adoptions
ALASKA 1 1982 ALASKA 2 1998, 1999
ARIZONA 2 1978, 1992 ARIZONA 1 2010
CALILORNIA 1 1979 CALIFORNIA 3 1996, 2003, 2015
COLORADO 4 1977, 1982, 1991, 1992 COLORADO 2 2000, 2010
CONNECTICUT 2 1991, 1992 CONNECTICUT 1 2012
DELAWARE 2 1978, 1980 DELAWARE 1 2011
LLORIDA 1 1994 HAWAII 2 2000, 2013
HAWAII 1 1978 MAINE 5 1999, 2001, 2009, 2010, 2011
IDAHO 1 1980 MARYLAND 2 2014, 2015
INDIANA 1 2002 MASSACHUSETTS 1 2012
IOWA 1 1992 MICHIGAN 1 2008
KENTUCKY 1 2000 MINNESOTA 1 2014
LOUISIANA 2 1979, 1993 MONTANA 2 2004, 2011
MAINE 1 2005 NEVADA 4 1998, 2000, 2001, 2013
MASSACHUSETTS 2 1986, 2002 NEW HAMPSHIRE 1 2013
MICHIGAN 2 1978, 1994 NEW JERSEY 3 2010, 2012,2013
MISSISSIPPI 2 1982, 1992 NEW MEXICO 1 2007
MISSOURI 2 1980, 1996 NEW YORK 1 2014
MONTANA 1 1981 OREGON 3 1998, 2005, 2013
NEVADA 2 1979, 1996 RHODE ISLAND 4 2006, 2007, 2009, 2012
NEW JERSEY 2 1976, 1990 VERMONT 4 2004, 2007, 2011, 2014
NORTH CAROLINA 1 1991 WASHINGTON 4 1998, 2010, 2011, 2015
OHIO 1 2006
OKLAHOMA 2 1985, 1992
OREGON 4 1979, 1996, 2000, 2001
RHODE ISLAND 1 1992
SOUTH CAROLINA 2 1980, 1984
SOUTH DAKOTA 1 1996
TENNESSEE 1 1978
TEXAS 1 1978
UTAH 1 1989
WASHINGTON 2 1979, 1993
WISCONSIN 1 2001
51


Table 8
Timeline of Repeating Event Adoptions
T ax and Exp enditure Limitations Year
NJ 1976
CO 1977
TX TN MI HI DE AZ 1978
WA OR NV LA CA 1979
SC MO ID DE 1980
MT 1981
MS CO AK 1982
1983
SC 1984
OK 1985
MA 1986
1987
1988
UT 1989
NJ 1990
NC CT CO 1991
OK MS IA CT CO AZ 1992
WA LA 1993
MI FL 1994
1995
SD OR NV MS 1996
1997
1998
1999
OR KY 2000
WI OR 2001
MA IN 2002
2003
2004
ME 2005
OH 2006
2007
2008
2009
2010 2011 2012
2013
2014
2015
Medical Marijuana
CA
AK NV OR WA
AK ME
CO HI NV
ME NV
CA
MT VT
OR
RI
NM RI VT
MI
ME RI
AZ CO ME NJ WA
DE ME MT VT WA
CT MA NJ RI
HI NV NH NJ OR
MD MN NY VT
CA MD WA
52


Fifteen states have adopted more than one TEL, which constitutes 19 repeating events. Colorado and Oregon each have four TELs. Colorado’s four policy adoptions are detailed above. Oregon first passed a TEL in 1979, passed by the state legislature, which required immediate refunds of surplus revenue to taxpayers. The refund provision was established in the state Constitution in 2000, and the next year was modified by the legislature (New, 2010). Further, Oregon, by referendum in 1996, approved the requirement of a three-fifths legislative supermajority to raise taxes (Waisanen, 2010).
Thirteen other states have adopted two TEL policies. Arizona passed a constitutional spending limit in 1978 and a legislative supermajority requirement to raise taxes in 1992. Connecticut passed a statutory spending limit in 1991 and a constitutional limit in 1992. Delaware in 1978 passed a constitutional spending limit, adding a legislative supermajority to raise taxes in 1980. Louisiana enacted its first TEL legislatively in 1979 and added a constitutional spending limit in 1993. Massachusetts passed a spending limit in 1986, which was amended in 2002. Michigan’s first TEL was approved in 1978, and a supermajority to raise taxes was passed in 1994 regarding state property taxes. Mississippi passed two spending limits in 1982 and 1992. Missouri passed a constitutional revenue limit in 1980, and added a constitutional measure requiring voter approval for certain tax increases.
Nevada passed a statutory spending limit in 1979, and through the initiative process added a legislative supermajority provision to raise any taxes. New Jersey’s legislature approved its first TEL in 1976, which expired in 1983. The legislature approved a new spending limit in 1990. Oklahoma in 1985 legislatively approved constitutional spending and appropriations limits (approved through the referenda procedures), and later approved an initiative in 1992 requiring a legislative supermajority to raise taxes. In 1980 and 1984, South Carolina
53


approved constitutional spending limits. And finally, Washington approved a TEL through the initiative process in both 1979 and 1993 (New, 2010; Waisanen, 2010).
Eighteen states have adopted one TEL, while 15 states do not have any TELs. All states, regardless of the number of policy adoptions, are analyzed in this research.
Regarding medical marijuana, 13 states have adopted multiple policies constituting 27 repeating events. The initial wave of medical marijuana laws were approved by initiative, but currently 12 of the 22 laws were legislatively enacted laws with Hawaii being the first in 2000 (“State-by-State Medical Marijuana Laws,” 2015). Of the 13 states with repeating events, Maine has the most repeating events with a total of five policy adoptions (“State-by-State Medical Marijuana Laws,” 2015).
In Maine, the policy was first adopted at the ballot in 1999 and has the most repeating events with five total policy adoptions. In 2001, the legislature clarified protections for patients and caregivers while also increasing the limit for marijuana possession. Another successful ballot question in 2009 established a patient and caregiver registry while also establishing dispensaries. There were two additional legislative changes in 2010 and 2011 (“State-by-State Medical Marijuana Laws,” 2015).
Four states have four policy adoptions. Nevada, though initiative, passed constitutional ballot measures in 1998 and 2000 to require the legislature to implement medical marijuana laws. Two legislative enactments in 2001 and 2013 separately removed criminal penalties and allowed for dispensaries. Rhode Island’s four policies were all enacted by the legislature in 2006, 2007, 2009, and 2012. The initial provision removed state-level criminal penalties for medical marijuana. Subsequent amendments created dispensaries and other regulatory reforms. Upon Vermont’s initial passage in 2004 by the
54


legislature, medical marijuana registration was only available to people with certain diseases, including AIDS and cancer. Subsequent amendments softened eligibility requirements and allowed dispensaries, among other regulatory provisions. Washington’s initial law in 1998 was passed by initiative. Subsequent legislative changes in 2010, 2011, and 2015 allowed patients and caregivers to collectively grow marijuana, allowed professionals to legally recommend marijuana for treatment, and created a voluntary registry (“State-by-State Medical Marijuana Laws,” 2015).
Three states have three policy adoptions. California, by ballot initiative, was the first state to allow for medical marijuana use in 1996, effectively removing criminal penalties for patients who receive a doctor recommendation. In 2003, the California legislature expanded the law “to allow patients and caregivers to collectively or cooperatively cultivate marijuana” (“State-by-State Medical Marijuana Laws,” 2015, p. 7). In 2015, a new regulatory system was legislatively adopted (to be phased in over several years) replacing the collective/cooperative model with a dispensary licensing system. New Jersey has one of the most restrictive laws in the country, requiring mandatory ID cards for patients and caregivers. The initial law, passed legislatively in 2010, also developed a state-regulated dispensary system. This initial law was amended in 2012 and 2013. Oregon’s medical marijuana law was first approved at the ballot in 1998 and removed state-level criminal penalties. Legislative amendments in 2005 and 2013 increased the plant and possession limits, and also established a registry system (“State-by-State Medical Marijuana Laws,” 2015).
Five states have two policy adoptions. Colorado approved Amendment 20 in 2000, with the medical marijuana law taking effect in 2001. In 2010 the legislature adopted
55


dispensary licensing procedures. Hawaii, the first state to adopt a medical marijuana law in 2000 through the legislature, modified the law in 2015 allowing for civil protections and regulations for patient access to marijuana. Alaska approved a ballot measure in 1998 removing state-level criminal penalties, and a legislative bill in 1999 implemented a mandatory registry program. Maryland legislatively removed state-level criminal penalties in 2014, and was amended in 2015. Finally, Montana approved medical marijuana by initiative in 2004. In 2011, the legislature made it more difficult to receive recommendations and placed other regulatory policies in place (“State-by-State Medical Marijuana Laws,” 2015).
Nine states have adopted one medical marijuana policy (“Medical Marijuana Overview,” 2015), while 28 states do not have any medical marijuana policy. All states, regardless of the number of policy adoptions, are analyzed in this research.
56


CHAPTER V
RESULTS
The results overview in this section covers both phases of analysis. Phase 1 analyzes single policy adoptions for 8 policy variables, while Phase 2 analyzes repeating events for two policy variables (TELs and medical marijuana). Both phases of this dissertation focus on two independent variables of interest. To reiterate, the simple presence of direct democracy in a state is a dichotomous variable, with “0” representing no policy adoption, and “1” representing policy adoption. The second independent variable of interest uses a difficulty index established by Bowler & Donovan (2004), giving states a higher score as it becomes more difficult to place a question on the ballot.
• Hypothesis 1: Direct democracy is positively related to policy innovation in the states.
• Hypothesis 2: The difficulty of placing a question on a statewide ballot will be negatively related with higher levels of policy innovation.
Twelve analysis scenarios (see Table 9) with different combinations of the independent variables were run for each phase - six combinations using the dichotomous variable for the existence of direct democracy, and the same six combinations using the direct democracy index variable. For the diffusion component, the contiguous states measure (percentage of bordering states having already adopted the policy in question) and fixed region measure (percentage of states within the same region having already adopted the policy in question) were necessarily run separately. Finally, there are three similar variables for partisanship: 1) Consecutive years of single party control of the governor’s office (GPC), 2) consecutive years a single party controls both chambers of a legislature (LMC), and 3)
57


consecutive years a single party controls both chambers of the legislature and the governor’s office (CPC). For this dissertation, three combinations of the partisanship variables were considered: including all three variables, including only CPC, and a third excluding CPC. These scenarios were developed since CPC is similar to the combination of LMC and GPC. The first four scenarios capturing all three variables are able to capture single party control of both the governor’s office and the legislature (GPC and LMC) individually, while the CPC variable informs whether the same party or different parties controlled both.
Table 9
Analysis Scenarios
SCENARIO VARIABLE OF INTEREST DIFFUSION PARTISANSHIP VARIABLES
1 PRESENCE OF DD CONTIGUOUS STATE ALL VARIABLES
2 PRESENCE OF DD FIXED REGION ALL VARIABLES
3 INDEX CONTIGUOUS STATE ALL VARIABLES
4 INDEX FIXED REGION ALL VARIABLES
5 PRESENCE OF DD CONTIGUOUS STATE EXCLUDES GPC & LMC VARIABLES
6 PRESENCE OF DD FIXED REGION EXCLUDES GPC & LMC VARIABLES
7 INDEX CONTIGUOUS STATE EXCLUDES GPC & LMC VARIABLES
8 INDEX FIXED REGION EXCLUDES GPC & LMC VARIABLES
9 PRESENCE OF DD CONTIGUOUS STATE EXCLUDES CPC VARIABLE
10 PRESENCE OF DD FIXED REGION EXCLUDES CPC VARIABLE
11 INDEX CONTIGUOUS STATE EXCLUDES CPC VARIABLE
12 INDEX FIXED REGION EXCLUDES CPC VARIABLE
Phase 1
Phase 1 looked at each of the 8 policy variables as single-event adoptions. Once a state adopted the policy in question, the state dropped out of analysis. In other words, once the given policy was adopted, the state was no longer “at risk” of another policy adoption.
As such, for each of the 8 dependent variables, the range of years in the analysis varied for different states, so the descriptive statistics for all variables (dependent and independent) also
58


varied depending on the dependent variable. For a full table of the descriptive statistics for all variables for each of the 8 analyses, please see Appendix B.
Tables 10 through 21 present the results for all 8 dependent variables for each of the 12 scenarios (for a total of 96 sets of results), showing the estimates, standard errors, and log pseudolikelihood values. Independent variables found to be related to policy adoption at a statistically significant level are highlighted with asterisks. There are four measures of statistical significance: p values of <0.001, <0.01, < 0.05, and <0.1. The log likelihood value is not used as an index of fit as it is a function of sample size, but can be used to compare the fit of various models relative to each other, where a value closer to zero is desired (Interpreting Log Likelihood, 2017). When clustering standard errors, STATA provides a log pseudolikelihood value (Log Pseudo Likelihood, 2004).
Table 22 summarizes by scenario the independent variables that show any level of statistical significance for each of the eight dependent variables. Table 23 summarizes the number of times each independent variable shows a significant relationship to each policy adoption. These tables give a visual representation of the relative statistical significant of the independent variables on policy adoption.
This section highlights the main findings from the Phase 1 analysis by reviewing findings from the 8 dependent variable policies within scenario pairs, pairing the scenarios with the same measures of direct democracy and partisanship, but differing diffusion measures.
It makes sense to report analysis findings by scenario pair in order to begin parsing out the relevant findings by variable type. For example, by analyzing scenarios 1 and 2 together, the dependent variables, the direct democracy variable, and all other independent
59


variables except the measure of diffusion remain common to the two scenarios. In this way, the impact of the diffusion measures on direct democracy’s influence of policy can be viewed in isolation.
Overall, findings are mixed regarding the role played by either measure of direct democracy on policy adoption. Throughout the various scenarios, the impact of direct democracy was little influenced by the partisanship measure used. Switching up the partisanship measure altered the outcome only for the index measure of direct democracy and only for four total scenarios relating to medical marijuana and TELs. The impact of both measures of direct democracy was influenced by the diffusion measures only when using the fixed-region measure of diffusion and only when analyzing TEL policy, as the direct democracy measures were only statistically significant for TELs when paired with the fixed-region measure. In total, 26 policy adoptions are statistically predicted by either measure of direct democracy in the direction hypothesized, out of a total of 96 possibilities (12 scenarios each for 8 policies). More details on these follow the summary results by scenario.
Further, overall findings show that diffusion plays the largest role in predicting the passage of policy innovation. Out of the 96 analyses, 69 show that diffusion has a statistically significant impact on policy adoption. More specifically, the fixed-region measure of diffusion is able to predict policy innovation in 43 of the 48 analyses on a statistically significant level.
60


Table 10
Phase 1, Analysis Scenario 1
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
Presence DD 1.04 (.65) .24 (.75)
Time Duration -.03 (.17) -.02 (.05)
Contiguous State Diff 1.13 (1.55) 2.79 (.50)****
Proxto Gov Election -.03 (.20) -.04 (.25)
Repub Control -.11 (.26) .14 (.19)
Single Party Gov (GPC) .14 (.04)*** -.02 (.03)
Rural Pop -.04 (.03) -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population .00 (.00)* -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)*
Legjs Prof 4.85 (3.38) 1.98 (1.74)
Education -.09 (.01)**** -.02 (.09)
Single Party Leg (LMC) -.01 (.01) .02 (.04)
Single Party Leg and Gov (CPC) -.06 (.05) -.05 (.05)
Log Pseudolikelihood -115.85 -122.15
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.) 1.95 (.97)** -.26 (.17) 2.49 (2.88) .01 (.22) .97 (.48)** -.03 (.04) .01 (.03) -.00 (.00) .00 (.00) .00 (.00)** -.83 (2.14) .18 (.05)**** .10 (.05)** .04 (.05) -78.92
EconDev Estimate (s.e.) -.65 (.59) -.05 (.05)
1.07 (.56)* .33 (.19)* -.67 (.38)* .11 (.06)* .11 (.00)*** .00 (.00)*** -.00 (.00) -.00 (.00)** 2.40 (2.67) -.03 (.06) .04 (.03) -.18 (.05)*** -99.21
MedMar Estimate (s.e.) 2.31 (.58)**** .12 (.03)**** 2.09 (.76)*** -.11 (.17) -1.44 (.48)*** .11 (.06)** -.01 (.03) .00 (.00)
.00 (.00)
.00 (.00)* -11.6 (4.57)** .11 (.03)**** -.00 (.02) -.29 (.09)*** -59.38
HateCrime Estimate (s.e.) .19(1.01) -.10 (.08) 1.96(1.24) -.09 (.14) -.42 (.18)** .11 (.09)
.02 (.02)
.00 (.00)** .00 (.00) -.00 (.00) 6.88(2.26)*** .15 (.05)*** .10 (.05)* -.21 (.09)** -106.78
PrimSeat
Estimate (s.e.)
-1.62 (.26)**** .03 (.07)
1.38 (1.49) -.08 (.12) -.43 (.30)
.08 (.08) -.00 (.01)
.00 (.00)
.00 (.00)*** -.00 (.00)
.66 (3.35) -.04 (.09)
.01 (.01) -.05 (.04) -100.99
KidHelmet Estimate (s.e.) -1.90 (.65)*** -.05 (.06) 3.71 (.68)**** .04 (.09) -.43 (.32) .02 (.05) -.02 (.01) .00 (.00)
.00 (.00) -.00 (.00) 4.93 (.68)**** -.04 (.05) .08 (.01)**** -.03 (.01)**** -81.26


Table 11
Phase 1, Analysis Scenario 2
VARIABLE TELs Estimate (s.e.) EnterZ one Estimate (s.e.)
Presence DD .92 (.44 )** -.10 (.62)
Time Duration -.03 (.08) -.09 (.04)**
Fixed Region Diff 1.68 (1.03) 3.57(1.0)****
Proxto Gov Election -.09 (.13) -.01 (.22)
Repub Control -.12 (.24) .04 (.13)
Single Party Gov (GPC) .14 (.05)*** -.02 (.05)
Rural Pop -.04 (.01)*** -.03 (.02)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 5.11 (2.72)* 1.30 (1.83)
Education -.12 (.05)** -.02 (.10)
Single Party Leg(LMC) -.02 (.02) .02 (.07)
Single Party Leg and Gov (CPC) -.06 (.04) -.03 (.04)
Log Pseudolikelihood -120.68 -126.95
Note. ****<0.001, ***<.01, **<0.05, *<0.1
F amCap Estimate (s.e.) EconDev Estimate (s.e.) MedMar Estimate (s.e.) HateCrime Estimate (s.e.) PrimSeat Estimate (s.e.) KidHelmet Estimate (s.e.)
1.38 (.53)*** -.57 (.47) 1 77 (49)**** .46 (.82) -1.71 (.36)**** -1.62 (.29)****
-17 (.14) -.21 (.05) **** .03 (.05) -.20 (.02)**** -.02 (.05) -.04 (.06)
3.47 (2.38) 4.46 (1.07)**** 3.14(1.28)** 3.87 (.88)**** 1.77 (.55)*** 4.72 (2.18)**
-.01 (.24) .245 (.19) -.13 (.14) -.05 (.12) -.16 (.19) -.00 (.07)
.47 (.25)* -.51 (.36) -1.42 (.47)*** -.22 (.16) -.27 (.13)** -.42 (.23)*
-.02 (.05) .06 (.07) .12 (.04)*** .14 (.09) .05 (.06) .01 (.02)
.01 (.03) .02 (.01)** .02 (.03) .02 (.02) -.01 (.00)* -.04 (.03)
.00 (.00) .00 (.00)*** .00 (.00) .00 (.00)** .00 (.00) .00 (.00)
.00 (.00) -.00 (.00) .00 (.00) .00 (.00) .00 (.00)**** .00 (.00)
.00 (.00)* -.00 (.00) .00 (.00)** .00 (.00) .00 (.00) -.00 (.00)*
.67(1.94) 2.80 (2.61) -2.86 (2.07) 8.61 (1.96)**** 1.23 (2.12) 5.37 (.69)****
.20 (.03)**** -.05 (.04) .01 (.04) .14 (.05)** -.07 (.06) -.02 (.08)
.07 (.04) .02 (.03) -.02 (.02) .08 (.05) .01 (.01) .06 (.02)**
.00 (.03) -.13 (.05)*** -.16 (.05)*** -.19 (.08)** -.02 (.03) -.03 (.02)
-80.26 -104.81 -66.29 -108.63 -110.03 -86.87


Table 12
Phase 1, Analysis Scenario 3
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
DD Index -.13 (.08) -.02 (.11)
Time Duration -.03 (.18) -.02 (.05)
Contiguous State Diff 1.08(1.51 2.75 (.50)****
Prox to Gov Election -.03 (.20) -.04 (.25)
Repub Control -.11 (.27) .13 (.21)
Single Party Gov (GPC) .14 (.04)**** -.02 (.03)
Rural Pop -.04 (.03) -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)* -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)**
Legjs Prof 4.89 (3.32) 2.13(1.82)
Education -.09 (.02)**** -.02 (.09)
Single Party Leg (EMC) -.01 (.01) .02 (.04)
Single Party Leg and Gov (CPC) -.06 (.04) -.05 (.05)
Log Pseudolikelihood -116.25 -122.2
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.) -.30 (.15)** -.27 (.18) 2.51 (2.99) .01 (.22)
.99 (.52)* -.04 (.04) .01 (.03) -.00 (.00) .00 (.00)* .00 (.00)** -1.05 (2.36) 19(06)*** .10 (.05)** .05 (.05) -78.58
EconDev Estimate (s.e.) .08 (.08) -.05 (.05)
1.07 (.56)* .33 (.18)* -.65 (.37)* .11 (.06)** .12 (.00)** .00 (.00)*** -.00 (.00) -.00 (.00)**
2.35 (2.65) -.03 (.06) .04 (.03) -.18 (.05)*** -99.3
MedMar Estimate (s.e.) -.32 (.10)*** .14 (.04)*** 2.13 (.78)*** -.10 (.16) -1.52 (.53)*** .11 (.05)** -.02 (.04) .00 (.00) .00 (.00) .00 (.00)* -11.7(5.57)** .10 (.03)**** -.01 (.02) -.30 (.09)*** -58.95
HateCrime Estimate (s.e.) -.01 (.12) -.10 (.08) 1.99 (1.21) -.09 (.14) -.42 (.18)** .11 (.09)
.02 (.02)
.00 (.00)* .00 (.00) -.00 (.00) 7.00 (2.36)*** .15 (.06)*** .10 (.05)* -.21 (.09)* -106.83
PrimSeat
Estimate (s.e.)
.22 (.04)**** .02 (.07) 1.39(1.38) -.08 (.12) -.45 (.30) .09 (.08) .00 (.01) .00 (.00)
.00 (.00)*** -.00 (.00) .77 (3.22) -.04 (.10) .02 (.01) -.06 (.04) -101.27
KidHelmet Estimate (s.e.)
.25 (.08)*** -.06 (.06)
3.66 (.63)**** .04 (.09) -.43 (.31) .02 (.04) -.02 (.01) .00 (.00)
.00 (.00) -.00 (.00) 4.70 (.82)**** -.03 (.05) .08 (.01)**** -.03 (.01)* -81.86


Table 13
Phase 1, Analysis Scenario 4
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
DD Index -.12 (.07)* .02 (.09)
Time Duration -.03 (.08) -.09 (.05)*
Fixed Region DifF 1.81 (1.04)* 3.57 (.97)****
Prox to Gov Election -.10 (.13) -.01 (.22)
Repub Control -.12 (.25) .04 (.14)
Single Party Gov (GPC) 14(05)*** -.02 (.05)
Rural Pop -.04 (.02)*** -.03 (.02)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legjs Prof 5.24 (2.77)* 1.36 (1.90)
Education -.12 (.05)** -.02 (.11)
Single Party Leg (EMC) -.02 (.02) .02 (.07)
Single Party Leg and Gov (CPC) -.06 (.03)** -.03 (.04)
Log Pseudolikelihood -120.64 -126.94
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.) -.22 (.07)*** -,18(.14)
3.47 (2.36) -.01 (.24) .48 (.26)* -.03 (.06) .01 (.03) .00 (.00) .00 (.00) .00 (.00)*** .40 (2.08) .20 (.03)**** .07 (.04)
.01 (.03) -79.83
EconDev Estimate (s.e.)
.08 (.07)
-.21 (.06)**** 4.44 (1.05)**** .24 (.19)
-.51 (.35)
.06 (.07)
.02 (.01)** .00 (.00)**** -.00 (.00) -.00 (.00)* 2.82 (2.59) -.05 (.05)
.02 (.03) -.13 (.05)** -104.85
MedMar Estimate (s.e.) -.24 (.07)*** .03 (.06) 3.40(1.36)** -.14 (.14) -1.43 (.47)*** .11 (.04)*** .02 (.03) .00 (.00) .00 (.00) .00 (.00)*** -1.87 (2.29) .00 (.05) -.03 (.02) -.16 (.05)*** -66.28
HateCrime Estimate (s.e.)
-.07 (.11) -.20 (.01)**** 3.96 (.69)**** -.05 (.12) -.22 (.17)
.14 (.09)
.02 (.02)
.00 (.00)**
.00 (.00)
.00 (.00) 8.52(1.99)**** .13 (.06)** .08 (.05) -.19 (.08)** -108.61
PrimSeat Estimate (s.e.)
.25 (.05)**** -.02 (.04) 1.83 (.71)*** -.16 (.18) -.30 (.13)** .05 (.06) -.00 (.00) .00 (.00) .00 (.00)*** -.00 (.00)
1.47 (2.28) -.06 (.07) .02 (.01)** -.02 (.03) -109.87
KidHelmet Estimate (s.e.)
.21 (.05)**** -.04 (.06) 4.77 (2.15)** .00 (.07) -.46 (.23)** .01 (.02) -.04 (.03) .00 (.00)
.00 (.00) -.00 (.00)* 5.12 (.95)**** -.01 (.08) .06 (.02)*** -.03 (.02) -87.28


Table 14
Phase 1, Analysis Scenario 5
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
Presence DD .92 (.63) .32 (.72)
Time Duration -.01 (.18) -.03 (.05)
Contiguous State Diff 1.21 (1.48) 2.96 (.70)****
Prox to Gov Election -.07 (.21) -.03 (.23)
Repub Control -.11 (.22) .15 (.21)
Rural Pop -.03 (.02) -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)* -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)*
Legis Prof 6.34 (3.61)* 1.83 (1.52)
Education -.08 (.01)**** -.01 (.08)
Single Party Leg and Gov (CPC) .01 (.05) -.05 (.05)
Log Pseudolikelihood -119.67 -122.38
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.)
1.81 (.82)** -.17 (.11) 3.05 (2.31) .01 (.23)
.47 (.22)** .01 (.03)
.00 (.00)
.00 (.00)
.00 (.00)*** .04 (1.91) .15 (.04)**** .04 (.05) -82.74
EconDev Estimate (s.e.) -.41 (.54) -.03 (.04)
1.38 (.46)*** .29 (.17)* -.58 (.28)** .01 (.01) .00 (.00)*** -.00 (.00) -.00 (.00)* 1.99 (2.31) -.01 (.07) -.09 (.07) -101.17
MedMar Estimate (s.e.) 2.32 (.95)** .06 (.03)** 2.17 (.64)***) -.24 (.17) -1.11 (.29)**** -.01 (.03)
.00 (.00)
.00 (.00)
.00 (.00)** -11.94 (6.26)* .11 (.08) -.21 (.13)** -61.51
HateCrime Estimate (s.e.) .35 (.90) -.13 (.02) 1.62 (.89)* -.13 (.12) -.46 (.10)**** .02 (.02)
.00 (.00)** .00 (.00) -.00 (.00)**** 6.22 (2.13)*** 17(05)*** -.11 (.07) -110.52
PrimSeat
Estimate (s.e.)
-1.53 (.23)**** .04 (.07) 1.07(1.39) -.13 (.15) -.41 (.24)* -.00 (.01)
.00 (.00)
.00 (.00)** -.00 (.00)
.71 (3.42) -.06 (.09)
.01 (.05) -102.61
KidHelmet Estimate (s.e.)
-1.51 (.59)** .01 (.08) 3.69 (.49)**** .05 (.09) -.65 (.44) -.02 (.01)*** .00 (.00)
.00 (.00) -.00 (.00) 4.89 (.98)**** -.03 (.04) .04 (.03) -84.16


Table 15
Phase 1, Analysis Scenario 6
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
Presence DD .78 (.40)* -.06 (.56)
Time Duration -.04 (.07) -.10 (.02)****
Fixed Region Diff 2.01 (.96)** 3.81 (.62)****
Proxto Gov Election -.16 (.14) -.00 (.20)
Repub Control -.08 (.20) .04 (.15)
Rural Pop -.03 (.01)**** -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 6.29 (2.66)** 1.18(1.47)
Education -.12 (.03)**** -.02 (.09)
Single Party Leg and Gov (CPC) .01 (.05) -.03 (.05)
Log Pseudolikelihood_____________-124.75___________-127.12
Note. ****<0.001, ***<.01, **<0.05, *<0.1
EconDev MedMar
Estimate (s.e.) Estimate (s.e.)
-.45 (.41)
-.20 (.02)****
FamCap Estimate (s.e.)
1.39 (.51)*** -.14 (.12) 4.60(1.86)** -.01 (.24) .24 (.18) -.01 (.03) .00 (.00) -.00 (.00) .00 (.00)
1.36 (2.00) .19 (.04)**** .01 (.04) -82.48
4.70 (.72)**** .22 (.17) -.47 (.27)* .02 (.01)** .00 (.00)**** -.00 (.00) -.00 (.00)
2.66 (2.48) -.04 (.05) -.08 (.03)** -105.35
1.66 (.44)**** -.00 (.06) 3.13(1.32)** -.27 (.15)* -1.05 (.34)*** .02 (.03)
.00 (.00)
.00 (.00)* .00 (.00)** -4.06(1.32)*** .00 (.04) -.13 (.07)* -69.17
HateCrime Estimate (s.e.)
.57 (.72) -.16 (.04)****
3.48 (.64)**** -.12 (.11) -.17 (.08)** .02 (.02)
.00 (.00)**
.00 (.00)
.00 (.00) 7.57(1.98)**** .16 (.05)*** -.08 (.07) -112.57
PrimSeat Estimate (s.e.) -1.67 (.33)**** -.01 (.05)
2.08 (.42)**** -.19 (.20) -.30 (.11)*** -.01 (.00)** .00 (.00)
.00 (.00)*
.00 (.00)
1.49 (2.10) -.08 (.06)
.02 (.05) -110.74
KidHelmet Estimate (s.e.) -1.30 (.29)**** -.01 (.07) 5.21 (2.42)** .01 (.08) -.52 (.30)* -.05 (.04)
.00 (.00) -.00 (.00) -.00 (.00) 5.41 (.77)**** -.03 (.07)
.02 (.01) -88.44


Table 16
Phase 1, Analysis Scenario 7
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.) FamCap Estimate (s.e.) EconDev Estimate (s.e.) MedMar Estimate (s.e.) HateCrime Estimate (s.e.) PrimSeat Estimate (s.e.) KidHelmet Estimate (s.e.)
DD Index -.11 (.08) -.03 (.11) -.28 (.13)** .05 (.08) -.32 (.13)** -.04 (.11) .20 (.04)**** .20 (.08)**
T ime Duration -.01 (.18) -.02 (.05) -.17 (.11) -.03 (.05) .08 (.03)** -.01 (.02) .03 (.07) .01 (.08)
Contiguous State Diff 1.16(1.45) 2.93 (.69)**** 3.07 (2.44) 1.37 (.45)*** 2.21 (.66)*** 1.63 (.88)* 1.06(1.30) 3.65 (.50)****
Proxto Gov Election -.07 (.20) -.03 (.23) .01 (.23) .29 (.17)* -.24 (.18) -.13 (.12) -.12 (.15) .05 (.09)
Repub Control -.11 (.23) .14 (.22) .47 (.24)* -.58 (.28)** -1.16 (.34)*** -.46 (.09)**** -.44 (.25)* -.64 (.43)
Rural Pop -.03 (.02)* -.03 (.03) .00 (.03) .01 (.01) -.02 (.04) .02 (.02) .00 (.00) -.02 (.01)**
Wealth .00 (.00) .00 (.00) .00 (.00) .00 (.00)*** .00 (.00) .00 (.00)** .00 (.00) .00 (.00)
Population -.00 (.00)* -.00 (.00) .00 (.00) -.00 (.00) .00 (.00) .00 (.00) .00 (.00)** .00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)* .00 (.00)*** -.00 (.00)* .00 (.00)** -.00 (.00)**** -.00 (.00) -.00 (.00)
LegLs Prof 6.43 (3.55)* 1.95(1.61) -.15(2.08) 1.92 (2.25) -12.00 (7.00)* 6.28 (2.16)*** .82 (3.33) 4.80 (1.12)****
Education -.08 (.00)**** -.01 (.08) .16 (.05)*** -.01 (.07) ..10 (.07) .17 (.05)*** -.06 (.09) -.22 (.04)
Single Party Leg and Gov (CPC) .01 (.05) -.05 (.05) .04 (.05) -.09 (.07) -.28 (.13)** -.11 (.07) .00 (.05) .04 (.04)
Log Ps eudolikelihood -120.1 -122.46 -82.38 -101.24 -61.1 -110.62 -103.01 -84.61
Note. ****<0.001, ***<.01, **<0.05, *<0.1


Table 17
Phase 1, Analysis Scenario 8
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
DD Index -.11 (.07) .01 (.08)
T ime Duration -.05 (.07) -.10 (.03)****
Fixed Region Diff 2.12 (.96)** 3.81 (.59)****
Prox to Gov Election -.16 (.14) -.00 (.20)
Repub Control -.08 (.20) .04 (.16)
Rural Pop -.03 (.01)**** -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)*
Legjs Prof 6.41 (2.71)** 1.23 (1.54)
Education -.12 (.03)*** -.02 (.09)
Single Party Leg and Gov (CPC) -.12 (.03)*** -.03 (.05)
Log Pseudolikelihood -124.73 -127.11
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.) EconDev Estimate (s.e.) MedMar Estimate (s.e.) HateCrime Estimate (s.e.) PrimSeat Estimate (s.e.) KidHelmet Estimate (s.e.)
-.22 (.07)*** .06 (.06) -.22 (.06)**** -.08 (.09) .24 (.05)**** 17(06)***
-.16 (.12) -.20 (.03)**** -.01 (.06) -.16 (.04)**** -.01 (.05) -.01 (.07)
4.61 (1.82)** 4.68 (.73)**** 3.27(1.40)** 3.58 (.49)**** 2.14 (.36)**** 5.21 (2.39)**
-.01 (.24) .22 (.17) -.28 (.16)* -.12 (.10) -.18 (.20) .01 (.08)
.25 (.17) -.47 (.27)* -1.06 (.34)*** -.17 (.08)** -.35 (.12)*** -.54 (.29)*
-.00 (.03) .02 (.01)** .02 (.03) .02 (.02) -.01 (.01) -.05 (.04)
.00 (.00) .00 (.00)**** .00 (.00) .00 (.00)** .00 (.00) .00 (.00)
-.00 (.00) -.00 (.00) .00 (.00)* .00 (.00) .00 (.00)*** -.00 (.00)
.00 (.00) -.00 (.00) .00 (.00)** .00 (.00) -.00 (.00) -.00 (.00)
1.10(2.09) 2.65 (2.43) -3.03 (1.69)* 7.47 (1.88)**** 1.68 (2.25) 5.30 (.91)****
19 ( 05)**** -.04 (.05) -.01 (.05) .16 (.05)*** -.07 (.07) -.02 (.07)
.02 (.04) -.08 (.03)** -.14 (.070)** -.08 (.007) .02 (.05) .02 (.01)*
-82.01 -105.4 -69.05 -112.53 -110.76 -88.79


Table 18
Phase 1, Analysis Scenario 9
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
Presence DD .97 (.69) .25 (.75)
Time Duration -.01 (.18) -.02 (.05)
Contiguous State Diff 1.29 (1.49) 2.85 (.50)****
Proxto Gov Election -.03 (.21) -.04 (.25)
Repub Control -.10 (.27) .18 (.19)
Single Party Gov (GPC) .12 (.05)** -.04 (.04)
Rural Pop -.04 (.03) -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)* -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 5.28 (3.25) 1.93 (1.89)
Education -.09 (.00)**** -.02 (.08)
Single Party Leg(LMC) -.04 (.03) .01 (.04)
Log Pseudolikelihood_____________-116.29___________-122.39
Note. ****<0.001, ***<.01, **<0.05, *<0.1
HateCrime PrimSeat
Estimate (s.e.) Estimate (s.e.)
FamCap Estimate (s.e.) 1.93 (.97)** -.26 (.16) 2.25 (2.68) .02 (.22)
.91 (.46)** -.00 (.06) .02 (.03)
.00 (.00)
.00 (.00)* .00 (.00)** -1.00 (2.00) .18 (.05)**** .10 (.05)** -79.07
EconDev Estimate (s.e.) -.52 (.52) -.01 (.05) .98 (.72)
.30 (.18)* -.47 (.28)* .11 (.04)
.01 (.01)* .00 (.00)* -.00 (.00) -.00 (.00)
2.67 (2.22) -.02 (.08) .01 (.03) -102.33
MedMar Estimate (s.e.)
1.81 (1.02)* .09 (.02)****
1.37 (.67)** -.13(46) -1.39 (.58)** .04 (.05) -.01 (.02) .00 (.00)
.00 (.00)
.00 (.00)* -5.63 (2.92)* .16 (.04)**** -.06 (.02)*** -66.14
.10 (.92) -.03 (.08) 1.80 (.88)** -.09 (.15) -.29 (.26) -.01(41)
.03 (.02)
.00 (.00)
.00 (.00)* -.00 (.00) 7.26 (1.56)**** .16 (.05)*** .04 (.06) -112.02
-1.64 (.25)****
.03 (.07) 1.45 (1.62) -40(41) -.41 (.28)
.06 (.09) -.00 (.01)
.00 (.00)
.00 (.00)** -.00 (.00)
.85 (3.07) -.05 (.10)
.00 (.01) -101.48
KidHelmet Estimate (s.e.) -1.89 (.60)*** -.04 (.06) 3.76 (.65)**** .03 (.09) -.42 (.31) .00 (.05) -.02 (.01) .00 (.00)
.00 (.00) -.00 (.00) 5.29 (.68)**** -.04 (.05) .08 (.01)**** -81.37


Table 19
Phase 1, Analysis Scenario 10
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.) FamCap Estimate (s.e.) EconDev Estimate (s.e.) MedMar Estimate (s.e.) HateCrime Estimate (s.e.) PrimSeat Estimate (s.e.) KidHelmet Estimate (s.e.)
Presence DD .87 (.44)** -.11 (.63) 1.39 (.53)*** -.45 (.40) 1.88(1.12)* .42 (.66) -1.72 (.33)**** -1.61 (.26)****
Time Duration -.01 (.08) -.09 (.04)** -.18 (.14) -.18 (.06)*** .03 (.04) -.17 (.03)**** -.02 (.05) -.03 (.07)
Fixed Region Diff 1.92 (.98)* 3.62 (.95)**** 3.46 (2.39) 4.45 (.98)**** 2.53 (1.07)** 3.93 (.75)**** 1.88 (.35)**** 4.82(2.18)**
Proxto Gov Election -.10 (.14) -.01 (.22) -.01 (.23) .23 (.18) -.16 (.16) -.08 (.13) -.17 (.18) -.01 (.07)
Repub Control -.09 (.24) .06 (.13) .46 (.27)* -.37 (.25) -1.41 (.63)** -.04 (.13) -.27 (.12)** -.40 (.21)*
Single Party Gov(GPC) .12 (.06)** -.03 (.05) -.02 (.06) -.01 (.04) .05 (.05) .00 (.08) .04 (.07) -.01 (.03)
Rural Pop -.04 (.01)*** -.03 (.02) .01 (.03) .02 (.01) .01 (.02) .03 (.02) -.01 (.00)** -.04 (.03)
Wealth .00 (.00) .00 (.00) .00 (.00) .00 (.00)*** .00 (.00) .00 (.00) .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00) .00 (.00) -.00 (.00) .00 (.00) .00 (.00)* .00 (.00)* .00 (.00)
Econ Conditions -.00 (.00) -.00 (.00) .00 (.00)* -.00 (.00) .00 (.00)*** .00 (.00) .00 (.00) -.00 (.00)*
Legis Prof 5.53 (2.57)** 1.28(1.91) .65 (1.84) 2.90 (2.54) -3.14(1.36)** 8.66(1.89)**** 1.42 (2.03) 5.76 (.47)****
Education -.13 (.04)*** -.02 (.10) .20 (.03)**** -.05 (.06) .09 (.03)*** .14 (.07)** -.07 (.06) -.02 (.09)
Single Party Leg (LMC) -.04 (.04) .02 (.07) .07 (.04) -.01 (.03) -.06 (.02)** .04 (.06) .01 (.00)**** .05 (.02)***
Log Pseudolikelihood -121.14 -127.03 -80.26 -106.47 -70.53 -113.28 -110.08 -86.95
Note. ****<0.001, ***<.01, **<0.05, *<0.1


Table 20
Phase 1, Analysis Scenario 11
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.)
DD Index -.11 (.09) -.02 (.11)
T ime Duration -.00 (.18) -.02 (.06)
Contiguous State Diff 1.25(1.47) 2.81 (.50)****
Prox to Gov Election -.03 (.20) -.04 (.25)
Repub Control -.10 (.29) .18 (.20)
Single Party Gov (GPC) .12 (.05)** -.04 (.04)
Rural Pop -.04 (.02)* -.03 (.03)
Wealth .00 (.00) .00 (.00)
Population -.00 (.00)* -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 5.38(3.19)* 2.06(1.97)
Education -.09 (.01)**** -.02 (.08)
Single Party Leg (LMC) -.04 (.03) .02 (.04)
Log Pseudolikelihood -116.71 -122.44
Note. ****<0.001, ***<.01, **<0.05, *<0.1
FamCap Estimate (s.e.) -.29 (.15)** -.26 (.16) 2.20 (2.77) .02 (.22) .92 (.49)* -.01 (.007) .01 (.04) .00 (.00) .00 (.00)** .00 (.00)** -1.21 (2.26) .18 (.06)*** .10 (.04)** -78.79
EconDev Estimate (s.e.) .06 (.07) -.02 (.05) .98 (.71) .30 (.18)* -.45 (.27)* .01 (.04)
.01 (.01)* .00 (.00)* .00 (.00) -.00 (.00) 2.60 (2.16) -.02 (.08) .01 (.03) -102.41
MedMar Estimate (s.e.)
-.24 (.16) .10 (.02)**** 1.41 (.71)** -.13 (.17) -1.43 (.67)** .04 (.05) -.02 (.03) .00 (.00)
.00 (.00)
.00 (.00) -5.55 (3.48) .15 (.03)**** -.06 (.03)** -66
HateCrime Estimate (s.e.) -.00 (.11) -.03 (.08)
1.81 (.89)** -.09 (.15) -.29 (.26) -.01 (.11)
.03 (.02)
.00 (.00)
.00 (.00)* -.00 (.00) 7.39(1.54)**** .15 (.05)*** .04 (.06) -112.04
PrimSeat Estimate (s.e.) .21 (.04)**** .02 (.07) 1.43 (1.51) -.10 (.11) -.43 (.29) .06 (.09)
.00 (.01)
.00 (.00)
.00 (.00)** -.00 (.00) .97 (2.92) -.05 (.10) .00 (.01) -101.9
KidHelmet Estimate (s.e.) .25 (.07)**** -.05 (.05) 3.71 (.60)**** .03 (.09) -.43 (.31) .00 (.05) -.02 (.01) .00 (.00)
.00 (.00) -.00 (.00) 5.04 (.88)**** -.03 (.05) .08 (.01)**** -81.95


Table 21
Phase 1, Analysis Scenario 12
VARIABLE TELs Estimate (s.e.) EnterZone Estimate (s.e.) FamCap Estimate (s.e.) EconDev Estimate (s.e.) MedMar Estimate (s.e.) HateCrime Estimate (s.e.) PrimSeat Estimate (s.e.) KidHelmet Estimate (s.e.)
DD Index -.12 (.07) .02 (.09) -.22 (.07)*** .06 (.06) -.24 (.15) -.06 (.09) .25 (.05)**** .21 (.06)****
T ime Duration -.20 (.08) -.09 (.05)* -.16 (.14) -.17 (.06)*** .03 (.04) - 17(03)**** -.02 (.04) -.03 (.06)
Fixed Region Diff 2.06 (1.01)** 3.61 (.92)**** 3.43 (2.38) 4.42 (.97)**** 2.67(1.18)** 4.01 (.61)**** 1.96 (.55)**** 4.86 (2.17)**
Proxto Gov Election -.10 (.13) -.01 (.22) -.01 (.23) .23 (.18) -.17(47) -.08 (.12) -.16 (.18) -.00 (.07)
Repub Control -.08 (.25) .06 (.13) .48 (.28)* -.37 (.25) -1.39 (.62)** -.05 (.13) -.30 (.13)** -.44 (.22)**
Single Party Gov (GPC) .11 (.06)** -.03 (.05) -.02 (.06) -.01 (.04) .05 (.05) .00 (.08) .04 (.07) -.00 (.03)
Rural Pop -.04 (.01)*** -.03 (.02) .01 (.03) .02 (.01) .01 (.02) .03 (.02) -.00 (.00) -.04 (.03)
Wealth .00 (.00) .00 (.00) .00 (.00) .00 (.00)*** .00 (.00) .00 (.00) .00 (.00) .00 (.00)
Population -.00 (.00)** -.00 (.00) .00 (.00) -.00 (.00) .00 (.00) .00 (.00)* .00 (.00)*** .00 (.00)
Econ Conditions -.00 (.00) -.00 (.00) .00 (.00)*** -.00 (.00) .00 (.00)*** .00 (.00) -.00 (.00) -.00 (.00)**
Legis Prof 5.69 (2.61)** 1.34(1.99) .35 (1.99) 2.89 (2.45) -2.03 (.84)** 8.60 (1.87)**** 1.65 (2.14) 5 53 (77)****
Education -.13 (.04)*** -.02 (.10) .20 (.04)**** -.04 (.07) .08 (.05)* ,14(.06)** -.06 (.07) -.01 (.09)
Single Party Leg (EMC) -.05 (.04) .02 (.07) .07 (.04)* -.01 (.03) -.06 (.03)** .04 (.07) .01 (.00)**** .05 (.02)
Log Pseudolikelihood -121.15 -127.01 -79.84 -106.52 -70.61 -113.26 -109.97 -87.37
Note. ****<0.001, ***<.01, **<0.05, *<0.1


Table 22
Phase 1, Summary of Statistically Significant Independent Variables by Analysis Scenario
VARIABLE (SCENARIOS ANALYZED) TELs EnterZone FamCap EconDev M edM ar HateCrime PrimSeat KidHelmet
Presence DD 2, 6, 1, 2, 5, 1, 2, 5, 6, 1, 2, 5, 6, 1, 2, 5, 6,
(6) 10 6, 9, 10 9, 10 9, 10 9, 10
DD Index 3,4, 3,4, 3,4, 3,4,
(6) 4 7, 8, 11, 12 7, 8 7, 8, 11, 12 7, 8, 11, 12
Time Duration 2, 4, 6, 2, 4, 6, 1,3, 5, 2, 4, 6,
(12) 8, 10, 12 8, 10, 12 7, 9, 11 8, 10, 12
Contiguous State Diff 1, 3, 5, 7, 1,3, 5, 1,3, 5, 5, 1,3, 5,
(6) 9, 11 7 7, 9, 11 7, 9, 11 7, 9, 11
Fixed Region Diff 4, 6, 2, 4, 6, 6, 2, 4, 6, 2, 4, 6, 2, 4, 6, 2, 4, 6, 2, 4, 6,
(6) 8, 10, 12 8, 10, 12 8 8, 10, 12 8, 10, 12 8, 10, 12 8, 10, 12 8, 10, 12
Proxto Gov Election 1,3, 5, 6,
(12) 7, 9, 11 8
Repub Control 1, 2, 3, 4, 5, 1,3, 5,6, 1, 2, 3, 4, 5, 6, 1,3, 5, 6, 2, 4, 5, 6, 2, 4, 6,
(12) 7, 9, 10, 11, 12 7, 8, 9, 11 7, 8, 9, 10, 11, 12 7, 8 7, 8, 10, 12 8, 10, 12
Single Party Gov (GPC) L) 1, 2, 3, 4, 9, 10, 11, 12 1,3 1,2, 3, 4
Rural Pop 2, 4, 6, 1, 2, 3, 4, 6, 2, 6, 5,
(12) 7, 8, 10, 11, 12 8, 9, 11 10 7
Wealth 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6,
(12) 7, 8, 9, 10, 11, 12 7, 8
Population 1,2, 3, 4, 5, 6, 3, 6, 9, 10, 11, 12 1, 2, 3, 4, 5, 6,
(12) 7, 8, 9, 10, 11, 12 9, 11 8 7, 8, 9, 10, 11, 12
Econ Conditions 1,3, 5, 1, 2, 3, 4, 5, 1,3, 4,5, 1, 2, 3, 4, 5, 6, 5, 2, 4,
(12) 7, 8 7, 9, 10, 11, 12 7 7, 8, 9, 10, 12 7 10, 12
Legis Prof 2, 4, 5, 6, 1,3, 5, 6, 1, 2, 3, 4, 5, 6, 1,2, 3, 4, 5, 6,
(12) 7, 8, 10, 11, 12 7, 8, 9, 10, 12 7, 8, 9, 10, 11, 12 7, 8, 9, 10, 11, 12
Education 1,2, 3, 4, 5, 6, 1,2, 3,4, 5, 6, 1,3, 1, 2, 3, 4, 5, 6,
(12) 7, 8, 9, 10, 11, 12 7, 8, 9, 10, 11, 12 9, 10, 11, 12 7, 8, 9, 10, 11, 12
Single Party Leg (LM C) (S) 1,3, 9, 11, 12 9, 10, 11, 12 1,3 4, 10, 12 1, 2, 3, 4, 9, 10, 11
Single Party Leg and Gov (CPC) (S) 4 1, 2, 3, 4, 6, 8 1, 2, 3, 4, 5, 6, 7, 8 1, 2, 3, 4 1,3, 8
Note. For direct democracy variables, hypotheses are supported for TEL, FamCap, and MedMar policies; hypotheses are not supported for PrimSeat and KidHelment policies


Table 23
Phase 1, Count of times the independent variable is significantly related to policy adoption
VARIABLE (SCENARIOS ANALYZED) TELs EnterZone FamCap EconDev M edM ar HateCrime PrimSeat KidHelmet
Presence DD (6) 3 6 6 6 6
DD Index (6) 1 6 4 6 6
T ime Duration (12) 6 6 6 6
Contiguous State Diff (6) 6 4 6 4 6
Fixed Region Diff (6) 6 6 2 6 6 6 6 6
Proxto Gov Election
(12) 6
Repub Control (12) 10 8 12 6 8 6
Single Party Gov (GPC) (8) 8 2 4
Rural Pop (12) 8 8 3 2
Wealth (12) 12 8
Population (12) 12 3 2 4 12
Econ Conditions (12) 5 10 5 11 2 4
Legis Prof i 12) 9 9 12 12
Education (12) 12 12 6 12
Single Party Leg (LM C) (8) " 5 4 2 3 7
Single Party Leg and Gov (CPC) ' 4 6 8 4 3
Note. For direct democracy variables, hypotheses are supported for TEL, FamCap, and MedMar policies; hypotheses are not supported for PrimSeat and KidHelment policies


Scenarios 1 and 2
Scenarios 1 and 2 model the independent variables’ impact on policy adoption using the presence of direct democracy, all measures of partisanship, and allowing the measure of diffusion to vary between the contiguous-state measure (scenario 1) and the fixed-rate measure (scenario 2). Across the two scenarios, direct democracy has a statistically significant impact on policy adoption in about half of the policy areas, though there were slight differences in the results for the different measures of diffusion. The log pseudolikelihood measure varies only slightly for each analysis across the scenarios.
In scenario 1, under the contiguous state diffusion measure, the presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 2), again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the presence of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation.
Unexpectedly, under both scenarios 1 and 2, the presence of direct democracy was negatively and significantly related to the adoption of primary seat belt laws and kid helmet laws. Running against the hypothesis stated, the presence of direct democracy is a deterrent to adopting these safety-minded policies. This finding is explored further in the discussion section of this dissertation.
The influence of the partisanship variables on the impact of direct democracy on policy adoption is discussed in the subsequent analyses of the scenario pairs.
75


This research expects to find that both diffusion measures would be positively related to policy adoption. Results show the contiguous state diffusion measure was positively and significantly related to the policy adoption of four policies - enterprise zones, economic development, medical marijuana, and kid helmet laws. The fixed region measure of diffusion was positively and significantly related to the policy adoption of six policies, including the same four measures just noted plus hate crimes and primary seat belt laws.
Regarding partisanship, the research expects to find that prolonged single-party dominance of any branch of government would be negatively related to policy adoption, as lack of political competition would reduce the motivation to adopt innovation policies. Consecutive years of single-party control over the governor’s office, however, was positively and significantly related to the adoption of three policies under scenario 1: TELs, economic development, and medical marijuana. This variable was positively and significantly related to the adoption of two policies under scenario 2: TELs and medical marijuana.
Consecutive years of single-party control of both chambers of a state legislature was also positively and significantly related to the adoption of three policies under scenario 1: family cap expenditures, hate crime legislation, and kid helmet laws. Under scenario 2, this variable was positively and significantly related to only one policy variable, kid helmet laws.
Consecutive years of single-party control over both the governor’s office and both chambers of the legislature was negatively and significantly related to the adoption of four policies under scenario 1: economic development, medical marijuana, hate crime laws, and kid helmet laws. Interestingly, even where the single-party control of either the governor’s office or the legislature was positively related the policy adoption, when a single party showed sustained control of both the governor’s office and the legislature, the relationship to
76


the policy adoption turned negative. This last measure of partisanship supports the expected findings that prolonged single-party control would act as a discouragement to policy innovation. Results also indicate that having single-party control over just one branch of government does not necessarily act as a deterrent to policy innovation, as political competition could still exist from a different branch of government.
Within scenario 1, several of the other independent variables showed a significant relationship multiple policy adoptions. A state’s measure of Republican control was significantly related with four policy variables (one positive and three negative). A state’s economic conditions were also significantly related with four policy variables (two positive and two negative), along with education (one negative and three positive). Other variables showed significance with fewer policy adoptions. Within scenario 2, the Republican measure was again significantly related to four policies (one positive and three negative). Other variables showed a significant relationship with a fewer number of policy variables.
In summary, under scenarios 1 and 2, results show that diffusion measures more frequently had a statistically significant impact on policy adoption (10 of 16 analyses) than measures of direct democracy (5 of 16 analyses as hypothesized), with the fixed region measure specifically being the most accurate predictor of policy adoption (6 of 8 analyses). Scenarios 3 and 4
Scenarios 3 and 4 are the same as scenarios 1 and 2, except that the measure of direct democracy is the difficulty index. As the difficulty to place a question on the ballot increases based on varying state rules, the index score also increases (between 0 and 6, with noninitiative states receiving a 10 score). As such, this research hypothesizes that the direct
77


democracy index score is negatively related to policy adoption. Again, the log pseudolikelihood measures vary only slightly for each analysis across the scenarios.
In scenario 3, coupled with the contiguous state diffusion measure, the direct democracy index was, as hypothesized, negatively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 4), again as hypothesized, the direct democracy index was negatively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the direct democracy index, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation. This mirrors the results from scenarios 1 and 2. Under these scenarios, altering the measure of direct democracy does not change where direct democracy is significantly related to policy adoption in support of the research hypotheses.
Using the direct democracy index variable again showed a significant relationship with the adoption of primary seat belt laws and child helmet laws under bother scenarios, and again in the opposite direction hypothesized (this time a positive relationship). Here, the increasing difficulty to place a question on the ballot acts as a predictor of policy adoption, rather than a deterrent to policy adoption. Because this research scores non-initiative states as a 10 on the difficulty index, this finding indicates that non-initiative states are more likely to adopt these safety measures, which is a conclusion that would be consistent with the findings in scenarios 1 and 2.
The diffusion measures showed the same levels of statistical significance with the adoption of the same policy variables as scenarios 1 and 2 (relative to scenarios 3 and 4,
78


respectively), except that the fixed-region diffusion measure under scenario 4 also showed significance with the adoption of TELs, but only at the <0.1 level. This shows that the measure of direct democracy being used does not substantially affect the findings relative to diffusion’s impact on policy adoption.
The findings regarding the three measures of partisanship were also largely unchanged from scenarios 1 and 2. In scenario 4, single-party control over the governor’s office and the legislature was significantly and negatively related to the adoption of TELs, where it was not under scenario 2. In another difference, the single-party control over the legislature was significantly and positively related to the adoption of primary seat belt laws. The direction of the relationship among the partisanship variables was unchanged from scenarios 1 and 2, so regarding these variables, the same conclusions can be drawn as noted above.
In summarizing the other independent variables, scenario 3 showed three independent variables significantly related to four policies: a state’s Republican control (one positive, three negative), economic conditions (two in each direction), and education (three positive and one negative). In scenario 4, two independent variables showed a significant relationship with four policies: Republican control (one positive, three negative), and economic conditions (two in each direction). Other variables showed a significant relationship with fewer policy adoptions.
In summary, under scenarios 3 and 4, results show that diffusion again proved to be the most accurate predictor of policy adoption, this time for 11 of 16 analyses. The fixed region measure of diffusion again proved to be the most accurate predictor of policy adoption
79


(7 of 8 analyses). Like under scenarios 1 and 2, the index measure of direct democracy was significantly related in the direction hypothesized in 5 of the 16 policy analyses.
Scenarios 5 and 6
Scenarios 5 through 8 mirror scenarios 1 through 4, respectively, except that scenarios 5 through 8 only include the partisanship measure of a single party controlling both the governor’s office and both chambers of the legislature. These scenarios remove the measure of single-party control of only the governor’s office, and the measure of single-party control of only the legislature. Scenarios 5 and 6 use the presence of direct democracy measure coupled with the contiguous-state and fixed-region measures of diffusion. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios.
In scenario 5, under the contiguous state diffusion measure, the presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 6), again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the presence of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation.
Again, these scenarios also find a negative relationship (against the stated hypotheses) with the adoption of primary seat belt laws and kid helmet laws. Overall, relative to the measures of direct democracy from scenarios 1 and 2, there is no change to the policy adoptions that are statistically predicted by the presence of direct democracy, including the primary seat belt law and kid helmet law policy adoptions being negatively related at a
80


statistically significant level. Even with removing the two measures of single-party partisanship, direct democracy’s impact on policy innovation is unchanged.
The contiguous state diffusion measure showed positive significance with five of the eight policy measures (enterprise zones, economic development, medical marijuana, hate crimes and kid helmets). Under scenario 6, the fixed-region measure of diffusion shows a positive and statistically significant relationship with policy adoption in all eight areas at least at the <0.05 level. This supports the expected finding that states learn from other policy adoptions by other states and are not restricted to learning only from bordering states.
Using only the partisan measure of a single-party controlling both the governor’s office and both chambers of the legislature, this measure only shows a negative and significant relationship with one policy adoption (medical marijuana) under the contiguous state scenario, and two policy adoptions (economic development and medical marijuana) under the fixed-region scenario.
In scenario 5, two other independent variables showed a significant relationship with five policy adoptions: Republican control (one positive and four negative), and economic conditions (two positive and three negative). A state’s measure of legislative professionalism was significantly related to the adoption of four policy variables (three positive and one negative). In scenario 6, Republican control again was significantly related to five policy adoptions (all negative), and legislative professionalism with four policy adoptions (three positive and one negative). Other variables showed a significant relationship with a fewer number of policy adoptions.
In mirroring the results from scenarios 1 and 2, the existence of direct democracy played a significant role as hypothesized in five of the 16 policy adoption analyses. The
81


fixed-region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role in all eight policy analyses. The contiguous state measure was a predictor of policy adoption in five of eight analyses. Overall, diffusion played a significant role in 13 of 18 analyses.
Scenarios 7 and 8
Scenarios 7 and 8 use the direct democracy index measure using the contiguous state and fixed-region measures of diffusion. Like scenarios 5 and 6, scenarios 7 and 8 only include the partisanship measure of a single party controlling both the governor’s office and both chambers of the legislature. These scenarios remove the measure of single-party control of only the governor’s office, and the measure of single-party control of only the legislature. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios.
In scenario 7, under the contiguous state diffusion measure, the direct democracy index was, as hypothesized, negatively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 8), again as hypothesized, the direct democracy index was again negatively and significantly related to the adoption of family cap expenditure policies and medical marijuana.
There was no change to the index’s impact on the adoption of primary seat belt laws and kid helmet laws. The index showed a significant and positive relationship with the adoptions of these two policies against the stated hypotheses.
When paired with the fixed-region measure of diffusion, and in contrast to scenario 4, scenario 8 does not show that the direct democracy index has a significant relationship with
82


the adoption of TEL policies. By removing two measures of single-party partisanship, the index’s impact on TEL policies is no longer significant.
Linder scenario 7, the contiguous state measure of diffusion is significantly and positively related to five policy adoptions - enterprise zones, economic development, medical marijuana, hate crime legislation and kid helmet laws. Like scenario 6, the fixed region diffusion measure in scenario 8 showed a positive and statistically significant relationship with policy adoption for all eight policy variables at no less than the <0.05 level.
In these scenarios, there is only one measure of partisanship used - the single party control of both the legislature and the governor’s office. This measure under both scenarios is significantly related to five of the 16 policy analyses. As expected, this partisanship measure showed a negative relationship with medical marijuana under both scenarios, along with TELs and economic development under scenario 8. However, against expected findings under scenario 8, the partisanship measure was positively and significantly related to the adoption of kid helmet laws.
In scenario 7, two other independent variables show a significant relationship with the adoption of five policy areas: Republican control (one positive and four negative), and economic conditions (two positive and three negative). Legislative professionalism was significantly related to four policy adoptions (three positive and one negative). Under scenario 8, Republican control predicts the adoption of five policy areas (all negative), and legislative professionalism four policy areas (three positive and one negative). Other variables showed a significant relationship with a fewer number of policy variables.
Under scenarios 7 and 8, the direct democracy index plays a significant role as hypothesized in four of the 16 policy adoption analyses. This is one fewer compared to
83


scenarios 3 and 4, as the index is no longer significantly related to the adoption of TEL policies. The fixed-region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role in all eight policy analyses. The contiguous state measure was a predictor of policy adoption in five of eight analyses.
Overall, diffusion played a significant role in 13 of 18 analyses.
Scenarios 9 and 10
Scenarios 9 through 12 remove the partisanship variable used in scenarios 5 through 8 (single-party control over both the legislature and governor’s office) and reinstate the separate measures of single-party control of only the governor’s office and only the legislature. Scenarios 9 and 10 used the existence of direct democracy measure paired with the contiguous-state and fixed-region measures of diffusion. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios.
In scenario 9, under the contiguous state diffusion measure, the presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 10), again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the presence of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation.
There is no change regarding direct democracy’s negative relationship with primary seat belt laws and kid helmet laws, significantly related but in the opposite direction hypothesized.
84


By again altering the partisanship variables used, direct democracy’s impact on policy innovation is unchanged. Throughout these findings, the presence of direct democracy is insensitive to the changing combinations of partisanship variables.
The contiguous state diffusion measure in scenario 9 showed positive significance with four of the eight policy measures (enterprise zones, medical marijuana, hate crimes and kid helmets). Under scenario 10, the fixed-region measure of diffusion shows a positive and statistically significant relationship with seven of the eight policy variables, excluding only family cap expenditures.
The single-party control of the governor’s office is positively and significantly related to the adoption of TELs under both scenarios, against expected results. The measure of single-party control of the legislature holds a significant and negative relationship with medical marijuana under both scenarios. However, it holds a significant and positive relationship with family cap expenditures (scenario 9) primary seat belt laws (scenario 10), and kid helmet laws (both scenarios). To reiterate, the research expected the partisanship variables to be negatively related to policy innovation.
In scenario 9, two other independent variables show a significant relationship with four policy adoptions, population and education (three positive and one negative for each). Under scenario 10, three independent variables show a significant relationship with four policy adoptions: Republican control (one positive and three negative), legislative professionalism (three positive and one negative), and education (three positive and one negative). Other variables showed a significant relationship with a fewer number of policy variables.
85


In mirroring the results from scenarios 1, 2, 5 and 6, the existence of direct democracy played a significant role as hypothesized in five of the 16 policy adoption analyses. The fixed-region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role in seven of the eight policy areas. The contiguous state measure was a predictor of policy adoption in four of eight analyses. Overall, diffusion played a significant role in 11 of 18 analyses.
Scenarios 11 and 12
Scenarios 11 and 12 mirror scenarios 9 and 10 in using the single-party control individually of the legislature and governor’s office, except that these scenarios use the index measure of direct democracy. Like all of the other analysis pairs, these scenarios use the contiguous-state and fixed-region measures of diffusion. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios.
When paired with the either measure of diffusion, the direct democracy index is significantly related to policy adoption in the hypothesized direction for only one policy area, family cap expenditures. Under the previous scenarios using the direct democracy index, there was a significant and negative relationship with medical marijuana policy with both diffusion measures. That is no longer the case in scenarios 11 and 12. Further, in contrast to scenario 4 but mirroring scenario 8, the direct democracy index is not a significant predictor of TEL policy adoptions. These findings suggest that while the existence of direct democracy is not sensitive to the partisanship variables used, the direct democracy index’s predictive value of policy innovation is impacted by the changing measures of partisanship.
86


Under scenarios 11 and 12, the direct democracy index still remains significantly and positively related (against the hypotheses) to the adoption of primary seat belt laws and kid helmet laws.
Under scenario 11, the contiguous-state measure of diffusion is positively related to the adoption of four policies. The fixed-region measure of diffusion remains the driving force of policy innovation in these analyses, being statistically significant with the policy adoption of seven of the eight policy areas.
Single-party control of the governor’s office is only statistically related to TEL adoptions under both scenarios, and in a positive direction. The measure of single-party control of the legislature holds a significant and negative relationship with medical marijuana under both scenarios. However, it holds a significant and positive relationship with family cap expenditures (both scenarios) primary seat belt laws (scenario 12), and kid helmet laws (scenario 11).
In scenario 11, two independent variables show a significant relationship with four policy adoptions: population and education (three positive and one negative for both). In scenario 12, three independent variables show a significant relationship with four policy adoptions: Republican control (one positive and three negative), legislative professionalism, and education (three positive and one negative for each). Other variables showed a significant relationship with a fewer number of policy variables.
In summary, under scenarios 11 and 12, the direct democracy index holds a significant relationship with policy adoption in one policy area in the direction hypothesized. Diffusion again is the most significant predictor of policy adoption, showing significance in
87


11 of the 16 policy analyses (four for the contiguous-state measure, and seven for the fixed-region measure).
Summary of Phase 1 Results
This dissertation hypothesized that the mere presence of the initiative would be positively and significantly related to policy innovation, though this is only true for two policy variables in every scenario, and true for the adoption of TELs only when viewing diffusion through the fixed-region measure. Regarding the direct democracy index, this research hypothesized a negative relationship with policy innovation. This held true for adoption of family cap expenditure policies under all six scenarios, for medical marijuana policies under four of six scenarios, and for TELs under one scenario. These findings are discussed in greater detail below.
Overall, results in these analyses show that policy diffusion is the main driving force behind policy innovation, especially the fixed-region measure of diffusion. The various measures of partisanship, overall, had less impact on policy innovation, and the direction of the relationship across policy adoptions was inconsistent. These findings suggest that diffusion plays a greater role in policy innovation than does the presence of an initiative process in the states.
However, the two measures of direct democracy do significantly impact policy innovation in certain policy areas, partially supporting the hypotheses. Tables 24 and 25 show the p-values, when significant, resulting from the measures of direct democracy on each policy innovation in this research. These tables show that direct democracy measures do impact policy adoption in the hypothesized direction for TELs, family cap expenditures,
88


and medical marijuana policies. The significant relationship with primary seat belt laws and kid helmet laws, as noted before, are opposite the direction hypothesized.
Table 24
Phase 1, P-Values, Presence of Direct Democracy
Scenario TELs ENTERZONE FAMCAP ECON DEV MEDMAR HATECRIME PRIMSEAT KIDHELMET
1 0.045 0.000 0.000 0.003
2 0.035 0.009 0.000 0.000 0.000
5 0.027 0.015 0.000 0.011
6 0.053 0.006 0.000 0.000 0.000
9 0.047 0.076 0.000 0.001
10 0.049 0.009 0.092 0.000 0.000
Note . PRIMSEAT and KIDHELMET show significance but opposite the direction hypothesized
Table 25 Phase 1, P-Values, Direct Democracy Index
Scenario TELs ENTERZONE FAMCAP ECON DEV MEDMAR HATECRIME PRIMSEAT KIDHELMET
3 0.048 0.001 0.000 0.001
4 0.078 0.002 0.001 0.000 0.000
7 0.033 0.012 0.000 0.013
8 0.002 0.000 0.000 0.006
11 0.047 0.000 0.000
12 0.002 0.000 0.000
Note . PRIMSEAT and KIDHELMET show significance but opposite the direction hypothesized
Findings support, at least for three of the eight policy areas, that the initiative process does play some role in policy adoption in the hypothesized direction. For TELs, the impacts are only apparent when viewing the models with the fixed-region diffusion measure.
This research intentionally focused on four economic policies and four social policies to potentially allow one to draw distinctions between different types of policy. Yet, the results fail to show consistent patterns within policy types.
In the economic variables, the measures of direct democracy impact policy adoptions for TEL and family cap expenditure policies, but show no significant relationship with enterprise zones or the adoption of economic development measures. Regarding social policies, direct democracy plays a role in the adoption of medical marijuana policies, along with an adverse role in the adoption of primary seat belt laws and child helmet laws. The
89


lack of consistent findings makes it hard to draw any definitive conclusions from this research, but is discussed in further detail in Chapter VI.
While Phase 1 explores single-adoption policy events modeled after Berry’s and Berry’s (1990) research on lottery adoption, Jones and Branton (2005) found statistically inconsistent findings when analyzing the policy adoption of obscenity legislation in the states while using both single-event and repeating-event models. As such, Phase 2 of this research provides a special focus on TEL policy and medical marijuana policy by analyzing these dependent variables through a repeating-events model.
Each of these policy variables, as detailed above, shows multiple policy adoptions in multiple states. Where the Phase 1 research drops a state from the analysis once a given policy is adopted, Phase 2 takes into account the first and every subsequent policy adoption related to TELs and medical marijuana. The same 12 scenarios from Phase 1 are duplicated in the Phase 2 analysis. Jones and Branton (2005) found conflicting results when viewing policy adoption through single-event and repeating-event models; this research finds similar inconsistencies when exploring repeating policy adoptions of TEL and medical marijuana policy.
Phase 2
In Phase 2’s repeating event model, both measures of direct democracy, when modeled alongside the fixed-region measure of diffusion, were significantly related to the adoption of TEL policies in the direction hypothesized. There was no statistically significant relationship with TEL policy when using the contiguous-state measure of diffusion. Furthermore, neither measure of direct democracy was related to repeating-event policy
90


adoptions of medical marijuana in a statistically significant way regardless of the diffusion measure used.
The measures of fixed-region diffusion were statistically significant in every model of TEL adoption, but no measure of contiguous state diffusion was statistically significant. Regarding the variables of partisanship, no discernible patterns emerged to assess its impact on policy innovation.
More discussion follows regarding the hypothesis testing of direct democracy’s impact on policy innovation under Phase 2’s repeating events model.
Unlike in Phase 1, because no states drop out of the analysis after policy adoption, the descriptive statistics remain static under each of the 12 analysis scenarios. Appendix C provides for a table of descriptive statistics for each dependent and independent variable.
See Tables 26-37 for analysis results showing estimates, standard errors, and log pseudolikelihood measures for each of the 12 analysis scenarios. Table 38 shows the count of times each independent variable showed significance for repeating policy adoptions, giving a visual representation of the relative statistical significance of each independent variable on policy adoption.
Given that the partisanship variables did not consistently demonstrate a pattern of significance, the analysis is best focused on the scenarios where measures of direct democracy and measures of diffusion are common. This allows for better identification of any hidden patterns.
91


Table 26
Phase 2, Analysis Scenario 1
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
Presence DD .60 (.40) .32 (.65)
Contiguous State Diff .19 (.72) -.06(1.08)
Proxto Gov Election .02(42) -.13(44)
Repub Control -.46 (.22)** -.54 (.20)***
Single Party Gov (GPC) .07 (.04)* .01 (.04)
Rural Pop -.06 (.01)**** .01 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00)**** .00 (.00)
Legis Prof 7.25 (1.55)**** .98 (2.75)
Education .02 (.05) .11 (.08)
Single Party Leg(LMC) -.03 (.03) .01 (.02)
Single Party Leg and Gov (CPC) -.05 (.06) -.18 (.06)***
Log Pseudolikelihood -202.22 -158.8
Note . ****<0.001, ***<01, **<0.05, *<0.1
Table 27 Phase 2, Analysis Scenario 2
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
Presence DD .85 (.41)** .21 (.57)
F ixed Region Diff -1.86 (.89)** -.43 (.92)
Proxto Gov Election .01 (.1) -.08 (.10)
Repub Control -.30 (.25) - 70 ( 19)****
Single Party Gov (GPC) .08 (.04)* .02 (.03)
Rural Pop -.07 (.01)**** .02 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)*** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00
Legis Prof 4.67(1.76)*** 2.17(2.63)
Education .04 (.05) .15 (.07)**
Single Party Leg(LMC) -02 (.02) .01 (.02)
Single Party Leg and Gov (CPC) -.08 (.06) -.09 (.07)
Log Pseudolikelihood -217.59 -184.47
Note . ****<0.001, ***<01, **<0.05, *<0.1
92


Table 28
Phase 2, Analysis Scenario 3
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
DD Index -.08 (.06) -.01 (.08)
Contiguous State Diff .12 (.71) -.11 (1.06)
Proxto Gov Election .02 (.12) -.12(44)
Repub Control -.47 (.23)** -.54 (.20)***
Single Party Gov (GPC) .07 (.04)* .01 (.04)
Rural Pop -.06 (.01)**** .01 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00)*** .00 (.00)
Legis Prof 7.07(1.67)**** 1.42(2.66)
Education .01 (.05) .11 (.08)
Single Party Leg(LMC) -.03 (.03) .01 (.02)
Single Party Leg and Gov (CPC) -.05 (.06) -.17 (.06)***
Log Pseudolikelihood -202.19 -159.02
Note . ****<0.001, ***<01, **<0.05, *<0.1
Table 29 Phase 2, Analysis Scenario 4
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
DD Index -.12 (.06)** .00 (.07)
F ixed Region Diff -1.82 (.89)** -.43 (.90)
Proxto Gov Election .02(41) -.07 (.11)
Repub Control -.32 (.25) -.69 (.18)****
Single Party Gov (GPC) .08 (.04)** .02 (.03)
Rural Pop -.07 (.01)**** .02 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 4.60(1.80)** 2.49 (2.63)
Education .04 (.05) .15 (.07)**
Single Party Leg(LMC) -.02 (.02) .01 (.02)
Single Party Leg and Gov (CPC) -.08 (.06) -.09 (.07)
Log Pseudolikelihood -217.5 -184.6
Note . ****<0.001, ***<01, **<0.05, *<0.1
93


Table 30
Phase 2, Analysis Scenario 5
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
Presence DD .67 (.42) .37 (.61)
Contiguous State Diff .17 (.71) .02 (.90)
Proxto Gov Election .00(42) -.13(42)
Repub Control -.39 (.20)** -.55 (.20)***
Rural Pop -.06 (.01)**** .00 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00)**** .00 (.00)
Legis Prof 7.26(1.54)**** .76 (2.67)
Education .01 (.06) .11 (.08)
Single Party Leg and Gov (CPC) -.02 (.06) -.16 (.05)***
Log Pseudolikelihood -205.07 -159.02
Note . ****<0.001, ***<01, **<0.05, *<0.1
Table 31 Phase 2, Analysis Scenario 6
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
Presence DD .87 (.41)** .26 (.54)
F ixed Region Diff -1.70 (.88)* -.42 (.90)
Proxto Gov Election -.01 (.11) -.09(40)
Repub Control -.24 (.22) -.72 (.18)****
Rural Pop -.06 (.01)**** .02 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 4.93 (1.75)*** 1.91 (2.61)
Education .03 (.06) .15 (.07)**
Single Party Leg and Gov (CPC) -.05 (.06) -.07 (.05)
Log Pseudolikelihood -220.54 -184.72
Note . ****<0.001, ***<01, **<0.05, *<0.1
94


Table 32
Phase 2, Analysis Scenario 7
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
DD Index -.09 (.06) -.02 (.08)
Contiguous State Diff .10 (.71) -.01 (.91)
Proxto Gov Election .00 (.11) -.12 (.12)
Repub Control -.39 (.20)** -.55 (.20)***
Rural Pop -.06 (.01)**** .00 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00)**** .00 (.00)
Legis Prof 7.10(1.66)**** 1.17(2.63)
Education .00 (.06) .11 (.08)
Single Party Leg and Gov (CPC) -.02 (.06) -.15 (.05)***
Log Pseudolikelihood -205.15 -159.31
Note . ****<0.001, ***<01, **<0.05, *<0.1
Table 33 Phase 2, Analysis Scenario 8
VARIABLE TELs Estimate (s.e.) MedMar Estimate (s.e.)
DD Index -.12 (.06)* -.00 (.07)
F ixed Region Diff -1.66 (.88)* -.42 (.89)
Proxto Gov Election -.01 (.11) -.08(40)
Repub Control -.25 (.23) - 71 (17)****
Rural Pop -.066 (.01)**** .02 (.03)
Wealth -.00 (.00)** -.00 (.00)
Population -.00 (.00)**** -.00 (.00)
Econ Conditions -.00 (.00) -.00 (.00)
Legis Prof 4.90(1.80)*** 2.17(2.61)
Education .03 (.06) .15 (.06)
Single Party Leg and Gov (CPC) -.05 (.06) -.06 (.05)
Log Pseudolikelihood -220.63 -184.92
Note . ****<0.001, ***<01, **<0.05, *<0.1
95


Full Text

PAGE 1

DIRECT DEMOCRACY AND POLICY INNOVATION IN THE AMERICAN STATES: THROUGH THE LENS OF POLITICAL COMPETITION b y K. W. HARP B.S., University of Colorado Boulder, 2000 M. A ., University of Colorado Denver, 2006 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Public Affairs Program 2017

PAGE 2

ii This thesis for the Doctor of Philosophy degree by K. W. Harp has been approved for the Public Affairs Program by Paul Teske , Chair Christine Martell, Advisor Brian Gerber Tony Robinson Date: December 16 , 2017

PAGE 3

iii Harp, K . W. (PhD, Public Affairs Program) Direct Democracy and Policy Innovation in the American States: Exploring the Citizen Thesis directed by Associate Professor Christine Martell ABSTRACT This dissertation explores the role played by direct democracy in the states regarding the state level adoption of various social and economic policy innovations. Using a modified event history analysis, thi s dissertation explores two independent variables of interest the simple presence of direct democracy in the states, and the difficulty of ballot placement for each state against the single adoption of eight policy variables. It then explores the same independent variables in a repeating events model, exploring the variable impact s on the multiple adoptions of t ax and e xpenditure l imitations and m edical m arijuana policies. In the modified event history analysis, both measures of direct democracy play statistically significant roles to increase the adoption of three policy variables family cap exemptions, medical marijuana, and tax and expenditure limitations. Further, mandatory bicycle helmets for children and primary seatbelt laws are significantly impacted by both measures of direct democracy, but in these case s direct democracy acts as a deterrent to policy adoption. In the repeating events model, both measures of direct democracy are significantly related to the adoption of repeating tax and exp enditure limitation policy adoptions, but are not significant for the repeated adoptions of medical marijuana policies. The form and content of this abstract are approved. I recommend its publication. Approved: Christine Martell

PAGE 4

iv ACKNOWLEDGEMENTS Thanks to my wife Victoria for her support and patience (mostly) during my completion of this dissertation. Thanks to my two sons Henry and Owen for being awesome. Thanks to my parents for their infinite support. Thanks to Dr. Christine Martell for her per sistence. Thanks to my committee members Drs. Paul Teske, Brian Gerber, and Tony Robinson for years of support during my academic journey. Thanks to my colleague Kegan Reiswig for STATA help.

PAGE 5

v TABLE OF CONTENTS CHAPTER I. II. ..6 Direct Democracy Political Competition Policy Innovation Review of Empirical Literature . 2 Literature Review Conclusion III. RESEARCH QUESTI IV. RESEARC Phase Phase Explanation of 39 V. RES Phase 1 Ana Phase 2 Ana VI. DISCUSSION & Limitations and Fut . 118 APPENDI 126

PAGE 6

1 CHAPTER I INTRODUCTION I n his seminal work on direct democracy in the American states, Matsusaka (2004 ) argues that the days of questioning the value of direct democracy whether it is good or bad F uture research in direct democracy will turn toward understanding how it interacts with other democratic institutions and how these various institutions can effectively work together. To address these questions, new frameworks and theories must be developed to study the function of direct democracy in the states within . One such the American republic is the competition framework . As outlined by Matsusaka (2004 ), the competition framework views the citizen initiative as an avenue for the creation of public policy, and it co mpetes against the traditional political p arties (Republicans and Democrats) in the creation of new laws . Just as Republicans and Democrats compete in the arena of public policy, with votes and public support as the prize, the citizen initiative must also compete for votes at the ballot box. Ultimately, the citizen initiative eliminates the monopoly that elected officials otherwise have over the lawmaking process. In the private sector, product innovation is a necessary tool to gain or keep a competitive advantage over other firms in the marketplace. The focus of product innovation is to meet the evolving needs and expectations of consumers, the markets, and other stakeholders in the business world ( Rainey, 2005 ). In a decentralized marketplace, individua l firms and people experiment with new innovations to satisfy consumer demand

PAGE 7

2 that is not possible in a monopoly. . Applying this basic economic principle to policymaking, it follows that increased competition in the lawmaking process will promote greater levels of policy inn ovation (Matsusaka, 2004 ) as policy entrepreneurs are allowed entry into the lawmaking process. Where direct democracy does not exist, a policy entrepreneur must either run for elected office or successfully lobby an elected official to give attention to his policy innovation or idea. In states where the citizen initiative is available, access to t he policymaking marketplace is open to any individual with the means to place a ballot question before the voters. While this may not include every citizen in a state, it no doubt opens the doors to allow significantly more policy entrepreneurs to partici pate in the creation of new laws. At the time of publication, Matsusaka (2004 ) wrote that there exists no research viewing the citizen initiative through the competition perspective . This research is intended to contribute to the first empirical explorati the policymaking process through the lens of a competition framework. Throughout this dissertation, state 1 . D are used interc hangeably to indicate that a state allows citizens to propose a new statutory law or constitutional amendment through a ballot measure or referendum 2 . Putting an initiative on the ballot is generally accomplished through a petition, requiring a certain number of valid signatures under different rules , which var y by state, to place a question on a ballot. This dissertation does account for varying state rules 1 See the Literature Review section for a more detailed definition. 2 Popular referendum, which allows citizens to approve or reject legislation passed by the legislature, are allowed in 24 states though compared to the initiative are rarely used (Initiative & Referendum Institute, 2011).

PAGE 8

3 and regulations regarding the initiative process difficulty of ballot plac . Currently, there are 24 initiative states 18 allow constitutional amendments and 21 allow statutory changes (Initiative & Referendum Institute, 2011) (see Appendix A) . While Oregon was the first state to see an initia tive on the ballot in 1904, use of the initiative increased significantly after California passed the tax cutting Proposition 13 in 1978 ( Initiative Use, 2010). The 1990s saw a record 377 statewide initiatives throughout the country, of which 177 were app roved. There were 420 initiatives in the 2000s (2000 20 10 ) , of which 1 7 8 were approved by voters ( Initiative & Referendum Institute, 2017 ). Since 1904, almost 45 percent of all citizen initiatives in the United States were placed on the ballot in the last three decades alone (Initiative Use, 2010) , which indicates a sustained popularity to use the initiative to drive public policy in the American states. Additionally, Coloradans have s een 2 15 statewide initiatives (through 20 10 ), which lags behind only Oregon (35 5 ) and California (3 40 ) ( Initiative & Referendum Institute, 2017 ). Further , more than 70 percent of the American population now lives in a state or city where direct democracy is available. No state has ever repealed a law allowing for direct democracy, and states are adding the initiative at a rate of about one st ate per decade since WWII (Lupi a & Matsusaka, 2004). It is evident that use of the citizen initiative in policy formation is a permanent reality for almost half of the American states , but the role the initiative plays in policy innovation is largely untested. control variable to be included in studies of state politics and policy but also a subject worthy For policy makers, policy

PAGE 9

4 entrepreneurs, academics , and civic minded citizens, understanding what role the initiative plays in policy formation and how it interacts with other democratic instit utions is fast becoming a prerequisite to fully understanding the American policy process. The overriding resea rch question for this dissertation is: Does the citizen initiative promote policy innovation at the state level? To address this question, two hypotheses are tested: 1) direct democracy is positively related to policy innovation in the states, and 2) the difficulty of placing a question on a statewide ballot will be negatively related with higher levels of policy innovation. The test these hypotheses, this research tests multiple variables across eight various policy adoptions in 48 states. There are tw o models in this research. 1) Policies are treated as single event policy adoptions (the first policy adoption in each state). Once a state has adopted the policy in question, it drops from the data set. 2) For two policy variables, the second phase exp lores repeating policy adoptions (analyzing each policy adoption event even if a prior policy adoption has already occurred). The simple presence of direct democracy as a dichotomous variable is used to test the first hypothesis. The second hypothesis is tested by employing a state by state Qualification Difficulty Index (Bowler & Donovan, 2004), which considers various factors required to place a question on the ballot, ( e.g., signature requirements). This dissertation proceeds as follows. The next chapte r explore s existing literature in direct democracy, political competition, and policy innovation , and provides an overview of the relevant empirical literature . Chapter III details the research questions and hypotheses. Chapter IV covers the research des ign for Phase 1 and Phase 2 of this research and provides an explanation of all variables used. Chapter V details results of both statistical analyses

PAGE 10

5 implications for th eory and policy practice, while also detailing research limitations and areas for future research.

PAGE 11

6 CHAPTER II LITERATIVE REVIEW an government, the federal government was intentionally set up as a republic a system of government where people elect representatives. Madison (2012/1787) wrote about the vid Calvinistic sense of Convention it was clear that this distrust of man was first and foremost a distrust of the While not ignoring the democratic elements with in dem ocracy was intentionally avoided. 3 While direct democracy has been avoide d at the the citizen initiative at the state level beginning more than a century ago . Direct democracy in the states ome tax, just to name two accepted American institutions . After South Dakota became the first state to adopt the initiative in 1898, 19 more states followed suit by 1918 as part of the Progressive movement. To this day , opinion polls consistently show st rong support for the ini ti ative process at the state and local level (Matsusaka, 2004). 3 See Hofstadter (1989) for a more complete historical overview.

PAGE 12

7 The insertion of the citizen initiative at the state level plays a vital role in the creation of state policy and law s. T hree relevant themes are relevant to support the general hypothesis that greater competition in the policy arena through the inclusion of direct democracy will result in greater policy innovation at the state level. The following sections explore the literature in direct democracy , politi cal competition, and policy innovation. This is followed by a review of the relevant empirical literature. Direct Democracy literature on direct democracy has just recently come of age in the last 25 years (Smith & Tolber t , 2007) , and much of the early literature was purely descriptive or normative (Lupia & Matsusaka, 2004) . This literature has focused on a variety of topics related to direct democracy, including it s impact on voter preferences, minority interests, policy implementation, civic engagement, voter competencies, interest groups, and more (Smith & Tolbert, 2007) . 4 This review focuses on the junction between direct democracy and public policy. While most research impact on public policy are almost exclusively focused on individual or a limited number of policy areas, it is clear that the initiative does impact public policy in the states. There is, for example, a la rge amount of literature available on tax and expenditure limitations (TELs) and the initiative, which blossomed after California voters passed Proposition 13 in 1978. Wallin (2004) found that access to the ballot box through citizen initiative was a bett er predictor of TEL passage in a state than any other variable. This was especially true where legislatures were dominated by 4 See Smith & Tolbert, 2007; and Lupia & Matsusaka, 2004, for a more complete picture of all literature on direct democracy.

PAGE 13

8 a single party. Mullins and Wallin (2004) found that TEL expansion was more rapid in states where the citizen initiative was ava ilable. Not only did access to direct democracy play a TELs. New (2001) and Krol (2007) both found that TELs passed by initiative were more effective at li In other fiscal policy areas, Blume, Muller and Voigt (2009) found welfare spending to be lower in countries with mandatory referendums. Wagschal (1997) found direct democracy to be an effective constraint on expansive spending and taxation as countries with some form of direct democracy have significantly lower levels of public expenditure and taxation. iscal arena . For example, Gerber (1996) found that initiative states are more likely to pass parental consent laws (concerning abortion) closer to the median voter preference. But the boundaries in California (Gerber & Phillips, 2005). Initiative states are also more likely to adopt term limit policies for legislators (Tolbert, 1998; Bowler & Donovan, 1995), campaign finance restrictions (Pippen, Bowler & Donovan, 200 2 ), and laws conc erning English as the official language (Hero & Tolbert, 1996). Matsusaka (2000) concluded that the initiative process is not inherently a government reducing institution, but rather it is a tool that brings the legislature in line with voter preferences of the time through increased information. Matsusaka (1995, 2000, 2004) , the lead scholar in this policy area , found that in the latter half of the 20 th century, state spending was about 4 percent lower on average in initiative states (1995) . To the contrary,

PAGE 14

9 however, when looking at the first half of the 1900s, initiative states had higher state and local expenditures and revenues (Matsusaka, 2000). Much of the literature on direct democracy has focused on fiscal policy, since 1950, with most literature finding that initiative states had lower overall spending than non initiative states. The literature on direct democracy clearly indicates that the initiative does impact those that use it frequently, tend to have different policies than non p. 422). But other question s then emerge how does public policy differ across states and ergence ? Literature shows (Lupia & Matsusaka, 2004) that the initiative can impact policy in both a direct and indirect fashion. The direct model simply implies that the initiative impacts policy when voters approve a measure. The indirect effect is a measure of a change in legislative behavior resulting from the potential or the threat of an initiative (Lupia & Matsusaka, 2004 ; Gerber, 1996 ). The impact on policy adoption from the direct effect is many times larger than the indirect effect (Matsusaka, 2014). Bowler and Donovan (2004) also show that direct democracy should empirically be measured as more than a simple dummy variable, which assumes that all initiatives states are the same. In reality, each initiative state has different rules for placing a question on the ballot . While many of the procedural activities are similar, differences do exist among the states. Notably, the number of signatures required to place a question on the ballot varies by state. Also, some states have geographic re quirements ( e.g. , a certain number of signatures must be collected in each congressional district ) . Does the difficulty of placing a question on the ballot impact the use of the initiative process? Bowler and Donovan (2004) conclude it

PAGE 15

10 does matter, as tho se states with an easier process (less restrictions) had more initiatives on the ballot than in states where the process is more difficult. Laws concerning the initiative in Colorado make it the second easiest state for citizens to place a question on the statewide ballot (during the review period of this research ) , tied with California and North Dakota (Bowler & Donovan, 2004) . In 2016, however, Colorado voters passed Amendment 71, which require s , for constitutional measures, a certain number of signatur on the ballot and a supermajority requirement (55%) to approve the measure . Amendment 71 increases the difficulty for a citizen to place a question on the ballot. Until Amendm ent 71 passed, o allow ed for easier access to the ballot (Bowler & Donovan, 2004). This literature review shows that the presence of a citizen initiative process in a state does impact public policy. The literature fails, however, to show from a macro perspective exactly how and why the initiative impacts policy. Additionally, the cupboard is empty in viewing the phenomenon through a theoretical perspective beyond the direct and indirect impacts of the initiative . This di ssertation begins to fill the cupboard by exploring the Political Competition While the Progressive movement is generally credited with the expansion of direct democracy in the states, Smith and Fridkin (2008) argue that political competition was the catalyst for the adoption of the very policy innovation being explored in this research the adoption of the initiative in the American states. They found that int erparty legislative

PAGE 16

11 competition, weaker party organizational strength, and the presence of third parties were the most powerful predictors of a legislatively enacted ballot measure to allow direct democracy. Downs (1957) applied economic theory to the p olitical world to explain party politics, ideologies, and voter behavior. Just as firms seek to maximize profits and consumers their utility, parties seek to maximize political support while voters seek to maximize benefits from the government. By polit ical support, Downs explicitly intends vote maximization are apt to cease to be firms , so politicians who do not pursue votes in a rational manner are apt to ) . Downs (1957) assumes that both parties and voters are rational beings. He also notes that policies are cr eated to win elections, and denies that candidates seek office to pursue virtuous policy endeavors. To gain election (or re election), a political party must seek to maximize votes by moving party ideology to the middle of voter preference (Downs, 1957) , which is the root of the median voter theorem 5 (Llavador, 2000) . This often results in two party platforms converging toward the middle and ultimately coming to resemble one another (Downs, 1957) . However, where uncertainty and/or polarization exist, party platforms may evolve in opposite directions. Regardless of where party ideology ultimately res ts, the motivation is derived from the rational goal of gaining or keeping power through the election process. This economic model is predicated on competition between political parties. Stigler (1972) further applied the principles of economic competit ion to the political world. He asserted that the more competition over repeated election cycles that exists between the parties, the more responsive the political system becomes to the desires of the 5 Looking at policy and how it relates to median voter preferences in not explored in this disserta tion, but is further discussed in the section on future research.

PAGE 17

12 majority voter. But he also found, however, that elect oral results (how large or small the electoral victory) impact the responsiveness . If one party achieves, for example, only 49 percent of the seats in a legislative chamber, then the probability that a single party will achieve its entire platform before the next election is reduced. If a party has a strong majority over a long period of time (think of a monopoly of power) , however, then that party is better able to control the policy agenda where there is less risk in departing from median voter preferen ces . In either case a monopoly of power or electoral competition Stigler (1972) found evidence that political competition (or lack thereof) impacts how near or far a Crain and Tollison (1976) found additi onal findings . Bueno de Mesquita, Morrow, Siverson, and Smith (2001) found that prolonged control and dominance by a single party stifles competition and leads to cronyism and corruption. Conversely, they also found political competition leads to economic growth and prosperity. Kessler (2005) inserted information asymmetries into political competition. Where information asymmetries exist, policies formed in a representative system were fu rther away from the median voter than those policies cr eated through direct democracy. This indicates that direct democracy helps reduce information asymmetries the initiative process signals voter preferences to lawmakers. This allows for competing la wmakers to produce more innovative policy ideas based on voter preference. Berry and Berry (1990) found that states are more likely to adopt innovative policies in an election year. This indicates that election year competition does produce an incentive to promote innovative policy ideas. This idea was anecdotally supported from the Colorado 2012 legislative session, which preceded a presidential election. Two separate democratic lawmakers, who ultimately ran for Congress and lost in the 2012 election, ran bills that

PAGE 18

13 pressing need on the part of the citizens of Colorado ( Bad Bills , 2012, p. 10B ). One bill would have prevented discrimination of motorcycle riders based on their dress. The other would have withheld pay from lawmakers if the legislature failed to pass a budget on time. Bad Bills , 2012, p. 10B). Other scholars have expanded the political competiti on literature with more complex additions to the theory. For example, Tavits (2007) finds that welfare maximizing policy leads to vote increases at the polls, but policy shifts in core belief areas (typically social or value based issues such as minority rights or social justice) lead to feelings of betrayal repercussions on Election Day . Persico, Pueblita, and Si lverman (2011) explored intraparty competition through factions within each party and found that elected officials from non competitive districts are more successful in maintaining policies (even if inefficient) and capturing monetary advantages (pork barr el spending) . Rogers and Rogers ( 2000 ) applied political competition to the executive branch and found that a close race in a gubernatorial election acts as a check against bigger government ( measured as revenue and expenditure relative to personal income and population) . Additionally, the median preference of voters must be adjusted to take into account voter abstention, which could lead to a winning policy bearing little resemblance to the preferences of the median voter (Llavador, 2000). The literatu re thus far does indicate that parties do compete for votes, and that the level of competition does impact party platform, and thus public policy. While the literature

PAGE 19

14 on political competition is expanding its theoretical base, the inclusion of direct dem ocracy in the states as a variable of interest is largely absent in this literature. The exception is Bowler and Donovan (2006) study on the impact of direct democracy on political parties imp act on policy formation . They found that the presence of direct democracy in a state leads to a more restrictive legal framework and institutional constraints for political parties such as the implementation of term limits, civil service regulation, limitations of campaign contributions, and the ability to recall elected officials thus regulati ng and weakening their autonomy . This weakening effect gives direct democracy more relative power at the very least. Direct democracy was also associated with weaker party organizations as measured by elements of hierarchy, level of control over candidates, and patronage, among other measures. They also found that political parties are beginning to ad a pt to the presence of direct democracy by indirectly funding initiative campaigns, notably those that may drive a marriage bans, for example). This finding indicates that political parties may be competing with, or possibly leveraging , the citizen ini tiative in some way. To further bolster this claim, Boehmke (2002) found that the presence of direct initiative states. Polsby (1984) suggested that interest groups typically operate through If one accepts that interest groups are policy entrepreneurs who do impact the public policy agen da, then it implies that an increase in interest groups will impact the public policy agenda, and thus create more competition for elected officials. In non initiative states, a policy

PAGE 20

15 entrepreneur must run for office or lobby an elected official. In ini tiative states, however, where the legislative policy making monopoly is broken, citizens and interest groups have direct access to the ballot. creators of public policy, and in some states they (Matsusaka, 2005, pp. 157 158) when forced to react to successful initiatives. While they found that imminent legislati ve action minimized the role played by policy entrepreneurs, they did find entrepreneurs to be effective in cases where existing policy change systems are inadequate to address real or perceived challenges. In short, policy entrepreneurs are effective whe While these policy entrepreneurs must be willing to invest time, energy, reputation and money, allowing direct access to the ballot lessens the investment and risk to challenge the policy status quo ( Minstrom & Norman , 2009) . Mintrom and Vergari (1996) argue the exploration of policy entrepreneurship helps explain periods of dynamic policy change. The entrepreneurs do so by identifying policy needs and innovative solutions, bearing financial and emotional risk, and coordinating policymaking process what business entrepreneurs do for th e marketplace. That is to say, changes at the margins of current policy settings. Rather, they seek to change radically

PAGE 21

16 policymaking process, policy entrepreneurs often work outside the policy subsystem, which is to say that e ntrepreneurs look for way s to push their policy agenda outside the traditional legislative scope. In states where more policymaking avenues exist outside the legislative system, clearly the more opportunity exists for entrepreneurs to push an agenda and create dynamic and innovati ve policy changes. While the literature generally support s the conclusion that increased competition leads to an increase in policy ideas and policy adoption, within the context of direct democracy there are at least two conceivable ways in which the indir ect effects of direct democracy (still increasing competition) could act as deterrents to policy adoption. First, it is conceivable that the threat of a reactionary initiative to a legislatively adopted policy would dissuade a legislature from adopting th e policy in question. The literature on the indirect effects of direct democracy, however, tend to focus on policies that are legislatively adopted based on the threat of an initiative (Matsusaka, 2014), rather than policies that are not adopted due to th e threat of a reactionary ballot measure. Second, Matsusaka (2014) defines a second ary indirect effect of direct democracy as one that provides new information (a communication or signaling effect) to a legislature. Under this measure of the indirect eff ect , he found that in initiative states where a conservative ballot measure was defeated at the polls, that state was 21 percent less likely than a non initiative state to adopt the policy in question in future years. This provides a bit of evidence that political competition through direct democracy can decrease policy innovation. There is otherwise little to no evidence, scientific or anecdotal, that policymaking competition resulting from direct democracy actually decreases the potential for policy ad option. As such, this research continues under the notion that

PAGE 22

17 political competition would, at least generally if not overwhelmingly, support an increase in policy innovation. The literature in this section indicates three main themes: 1) political parti es do compete for votes through policy platforms, 2) direct democracy does have a direct impact on traditional political parties, and 3) the initiative significantly expands both the number of policy entrepreneurs who are able to impact the formulation of public policy and the avenues available to them . I t logically follows from these three findings that direct democracy in the states does play a competitive role in the American policy process. Policy Innovation In the literature, the generally accepted definition of a policy innovation i s a program or policy that is new to the state adopting it without regard for whether the policy has already been adopted in other states (Walker, 1969; Berry & Berry, 2007; Boehmke & Witmer, 2004). T his demands a distinction be made between policy innovation and policy invention , which is defined as the adoption of a policy never seen before in any state. Further, policy innovation is not a value statement, meaning that innovation , as used throughout this dissertation , in no way implies a given policy is good or bad only that it is a new policy for the state. Matsusaka (2005) holds that direct democracy is increasingly responsible for shaping the direction of state policy. He identified a bevy of h igh profile state policy issues that emerged through the initiative process; including affirmative action, illegal immigration, medical marijuana, TELs , and school vouchers. important policy developments that emanated from legislatures; it is difficult to identify more

PAGE 23

18 being driven as much by voter i 162). While this particular article was descriptive in nature, assertion exemplifies his conceptual idea (2004) that the citizen initiative creates competition in the lawmaking process lead ing to policy innovation in the states. There is limited empirical work elsewhere to support nascent competition framework. Kotsogiannis and Schwager (2006) found that in a federal system with political competition for office, there is a stro innovative policy positions. Candidates use innovative policy ideas to signal their ability to the electorate. For example, a legislator with aspirations for higher office will want to promote innovative policy ideas to be well received in future elections. This provides evidence that a federal system will generate more innovation than a unitary system. Adding to the policy innovation la boratorie ( Kotsogiannis & Schwager, 2006) . This paper ) competition framework in that it signals an elected owledgment of voter preferences . Further support for the competition framework can be drawn from Williams (2003), who found that policy innovation in criminal justice policy was predicted more by the presence of and advocacy by policy entrepreneur s than by state characteristics. In considering Boehmke (20 02) finding that initiative states have a 17 percent high er interest group population, i t follows that initiative states with more interest groups (policy entrepreneurs) and avenues for policy change could have more policy innovation. Obinger (1998), whil e concluding that the effects of federalism and direct democracy in Switzerland reduce political power related to the expansion of the Swiss welfare state , acknowledged

PAGE 24

19 down of decisionmaking does not necessarily affect the innovation capacity a nd Most of the academic discussion on policy innovation has taken place in the literature on policy diffus ion (Berry & Berry, 2007; Boehmk e & Witmer, 2004; Karch, 2006). Outside of the diffusion literature, policy innovation is little explored . But the literature that does exist provides a foundation from which to build this research . diffusion (Welch & Thompson, 1980), policy innovation was the dependent variable, with various sociodemographic factor s, political characteristics, and other factors acting as the independent variables. To operationalize his dependent variable, Walker looked at the rate of He found that wealthy, higher populated , and industrial state s had the highest innovation scores, but he did not account for states with direct democracy (the presence of the citizen initiative was one of the 88 policies considered for the innovation score, but was not considered as an independent variable that may impact the rate of policy innovation). He did measure political competition by the competitiveness of gubernatorial elections and found that it did A better measurement of party competition within the legislat ure can be attained for this study following the ideas of Stigler (1972), Crain and Tollison (1976), and Bueno de Mesquita et al. (2001). Walker (1969) does s would try to out do each other by embracing the newest, most progressive programs and this would naturally encourage the rapid adoption of innovations (p. 885). While he did not find

PAGE 25

20 empirical support for this assertion, he seems to suggest that competi tion is not fully captured by party competition Walker (1969) also found that legislative competition based on a measure between urban and rural representation impacted innovation scores from 1930 1966 . While he did not measure legislative competition as a function of party, the finding still indicates that legislative competition leads to policy innovation. Th is finding by Walker is supported by historical evidence. In the Civil War years, more than 80 percent of Americans lived in rural areas (Matsusaka, 2004). By 1920, a majority lived in urban areas, thus creating a strong demand for new and different serv ices such as water sanitation, sewage treatment, and garbage collection. This increased demand from a growing urban population for government services is a unique measure of legislative competition appropriate for that time . While it pits two competing a nd diverse interests (urban versus rural demands), there are parallels to t he concept of two competing interests in the form of party competition leading to policy innovation. ates Democrat, while representation of rural Colorado is dominated by Republican legislators. It is not a stretch to speculate rural and urban competition in the earl y part of the 20 th century is similar to contemporary times when legislative competition is measured as a function of partisanship. Boehmke and Skinner (2012 a ) study and increased the number of policy innovations to 189 , all treated as non repeating events . They found that states with higher populations, greater per capita income, and higher rates of urbanization

PAGE 26

21 were associated with higher innovation scores. This study did include the presence of the citizen initiative and f The authors, however, did not focus on the specific time frame that could have isolated the period from 1978 to present when in itiative use reached its peak (their study explored policy adopt ions beginning in 1912). Further, they treated the initiative variable as dichotomous, which did not take into account the varying institutional rules across the states. In their study of innovation in e government, Tolbert, Mossberger, and McNeal (2008) also found that wealthier states were more likely to engage in policy innovation, as were states with higher levels of education and legislative professionalization. Political, economic and social characteristics are also important factors leading to pol icy innovation in internal determinants models (Berry & Berry, 2007). Welch and Thompson (1980) also found that policies with federal monetary incentives lead to higher rates of policy innovation in the states. In the policy innovation literature, t he concept of party competition is only briefly explored, most notably by Walker (1969) and Boehmke and Sk inner (2012a) . Further, while internal characteristics play a role in policy innovation, external influences from other states must also be considered . Past research on policy innovation has traditionally explored either internal characteristics within a state or diffusion effects ( external influences) from other states to explain a policy change, but rarely were both captured in a single study (Berry & Berry, 1990). As explored further in the next section, diffusion literature is the leading arena in wh ich to discuss policy innovation in the states, though none of the literature expressly considers the potential role of direct democracy.

PAGE 27

22 This research incorporates ideas presented in the diffusion literature and adds the element of the citizen initiativ e as a possible determinant of policy innovation. While there exists a hint of direct democracy in the innovation literature, this dissertation puts the initiative on center stage in exploring policy innovation in the states. Review of Relevant Empirical Literature To remedy a methodology shortfall of failing to capture both internal characteristics of a state and external diffusion influences , Berry and Berry (1990) developed event history analysis (EHA) for the social sciences, which a ccounts for both internal and external influences on state policy adoption . Additionally, this approach captures influences that may impact policy adoption in each state in each year to account for changing circumstances in each individual state. Berry an d Berry (2007) proposed the following unified model of state government innovation that captures both internal determinants and the external influence of policy diffusion. Their proposed model is as follows: ADOPT i,t = f (MOTIVATION i,t , RESOURCES/OBSTACLE S i,t , OTHER POLICIES i,t , EXTERNAL i,t .) In their model (Berry & Berry, 2007), the unit of analysis is the individual state that will potentially adopt a policy. ADOPT, the dependent variable, is the probability that an individual state ( i ) will adopt a given policy in a given year ( t ). ADOPT is a function of internal and external variables (see Table 1 ). The first three subsets of variables relate to internal determinants. MOTIVATION contains variab les that indicate the motivation of public officials, such as the proximity to an election, public opinion, and electoral competition. RESOURCES/OBSTACLES is concerned with obstacles to innovation and resources that are available to overcome those obstacl economic development, legislative professionalism, and the pr esence of policy entrepreneurs.

PAGE 28

23

PAGE 29

24 The presence of direct democracy and difficulty of ballot placement variables fall in this category. OTH ER POLICIES would indicate whether or not other policies exist that might increase or decrease the likelihood of the policy in question be ing adopted in any given year. EXTERNAL is concerned with diffusion effects across state boundaries and is not concer ned with any internal determinants within a state. Berry and Berry (1990) analyzed pooled cross sectional time series data via event history analysis to test their model of state lottery adoptions. They used a discrete time model, meaning the time elemen t was divided into years (any fixed period of distinct time units would be a discrete time model). As such, the data are one observation per state for each year the state might possibly adopt the given policy. Event history analysis is explicitly structur ed to look at trends over time, while a simple regression or probit analysis would not policy was adopted. nce a lottery was adopted, a state is no it is a non repeating event. Further, because the variable is dichotomous, Berry and Berry (1990) used probit maximum likelihood estimation. The estimates of the coefficients for the independent variables measure the predicted impacts of these independent variables on the likelihood a state will adopt the specific policy innovation (Berry & Berry, 2007 generate predictions of the probability that a state with any specified combination of values 244). Further, the model esti mates the probability of adoption when any of the independent variables change while holding other variables constant.

PAGE 30

25 Since their 1990 study on state lottery adoptions, some refinements to event history analysis have been made. In the early literature, t he most common measure of diffusion (external influences) was to quantify the number or percentage of contiguous states adopting a policy (Berry & Berry, 2007). More recently, fixed region diffusion has been utilized. It first defines regions of the count ry and measures the number or percentage of states from a region that have adopted a policy to account for diffusion across non contiguous states (Mooney & Lee, 1995; Andrews, 2000; Allen, Pettus, and Haider Markel, 2004) , as states can learn from other st ates, even when there is no common border. Allen, Pettus, and Haider Markel (2004) also measured vertical influence to capture any effects of federal policies. The OTHER POLICIES variables, which capture influence from previously adopted policies within a state, were not included in early event history analysis studies (Berry & Berry, 2007). More recently, Balla (2001) measures the impact of previously adopted HMO measures that we re complementary to the policy in question (HMO Model Act). And Soule and Earl (2001) studied whether a state was more likely to adopt hate crime laws if other hate crime legislation had previously been approved. Further, Seljan and Weller (2011) found th at states were less likely to propose a TEL through the legislature or citizen initiative if a state in close geographic proximity had already rejected a TEL. Buckley and Westerland (2004) offered three fundamental criticisms of Berry and ) event history analysis model, but also offered recommendations to overcome those problems. The three major concerns over the EHA model are duration dependence, functional form, and corrected standard errors.

PAGE 31

26 Duration Dependence At a very minimum, scholars must account for duration dependence that is, scholars must account for the impact that the passage of time has on the likelihood of policy adoption in EHA models (Buckley & Westerland, 2004). The Berry and Berry (1990) EHA model, as with othe r EHA mode ls, assumes a flat hazard rate; EHA models assume that the probability of policy adoption does not change over time. To correct this problem, one must relax the flat hazard rate assumption. The most basic solution to this problem is to simply add a linear time counter to the model, adding one year to the measurement for each year in duration. The result is the following revised model: ADOPT i,t = f (MOTIVATION i,t , RESOURCES/OBSTACLES i,t , OTHER POLICIES i,t , EXTERNAL i,t , TIME COUNTER i,t .) This appro ach still assumes that the risk of policy adoption as a function of time changes at a constant rate, but this revised model is superior to the previous model that did Functional Form While logit and probit analyse s are still appropriate, they should not be the preferred functional form for EHA modeling (Buckley & Westerland, 2004). Buckley and Westerland (2004) assert that a complementary log log function (cloglog) can remedy this problem . In EHA, states are removed from the dataset once a given policy has been adopted , therefore creat ing a dataset with many more 0s than 1s. Using the cloglog function instead of probit or logit, Buckley and Westerland (2004) find that the hazard rate approaches 1 (p olicy adoption)

PAGE 32

27 for EHA because it accounts for the disproportionately high number of 0s (no policy . Corrected Standard Errors The final criticism of traditional EHA modeling is that it remains possible that reported standard errors are incorrect (Buckley & Westerland, 2004) . In general linear regression models (cloglog included), a requirement is an assumption that data are indepe ndent across time and in space; (p. 104) as EHA depends on correlation over both time and spac e. The solution is to use clustered standard errors, which relaxes the assumption of independent observations in the data (Buckley & Westerland, 2004). One fix for this problem is to cluster standard error based on region, as defined by the U.S. Census B ureau , to account for regional similarities among the states. This corrects for bias in EHA standard errors caused by spatial autocorrelation. Repeating Events was non repeating. In other words, once a state adopted the policy in question, there was no further exploration in that particular state . In the Berry and Berry model, the only notable event is the first event , which makes an unnecessarily (Jones & Branton, 2005, p. 430). To accoun t for policy events that can be repeated after the initial adoption, Jones and Branton (2005) propose a Cox Duration Model with a conditional gap time model modification they call it the stratified Cox model of

PAGE 33

28 repeated events. In this model, once a sta te passes a policy, instead of disappearing from the data set, it becomes at risk for a second event. The duration time counter resets to 1 after each event. This model is explored further in the research design section below. Literature Review Conclusion From this literature review, three major themes must be recognized. The first is the finding that direct democracy does impact public policy in various capacities at the state level but limited empirical evidence shows how. The second important theme i s party competition the literature clearly indicates that political parties compete against each other for votes and electoral victory. Within this theme it should also be recognized that the use of the initiative, gaining in popularity, has increasingl y broken the policy making monopoly impacting both parties and policy directly and indirectly. Finally, the literature on policy innovation suggests that political competition is one of many possible explanatory factors leading to policy innovation. Com bined, these themes indicate that direct democracy in the states could be a competitive factor in the formulation of public policy. Coupled with basic economic ideas on product innovation, it logically follows that direct democracy, through increased comp etition against the traditional parties, will enhance policy innovation in the American states. cutting Proposition 13 in 1978. This proposal was considered an outlandish idea at the time of its passage (Matsusaka, 2004) and was dismissed by the established parties and political experts, yet it was approved at the polls by a 65 Proposition 2½ in 1980, is largely credited as setting off

PAGE 34

29 1980s (Resnick, 2004). Currently, 46 states have some form of tax and expenditure limitation (TEL) in place at the statewide level (Amiel, Deller, & Stallman, 2009). While it cannot be known if the TEL movement would hav e taken off without the citizen initiative, it is clear that the initiative brought TELs to the forefront of public policy in the states for multiple decades. Finally, the empirical models to evaluate policy innovation and diffusion have advanced signif icantly since Berry and Berry (1990) laid the foundation for using event history analysis in the study of policy innovation . Newer models and modifications account ls over probit and logit. Other refinements account for spatial autocorrelation that was present in early EHA models and allow for the modeling of multiple event policies. These more advanced models and methods continue to be refined, but the relevant em pirical models in existing literature are capable of producing results for this research that are more empirically rigorous than the original Berry and Berry model. These more advanced models also allow for testing theory in a new way, allowing for the in sertion of various measures of direct democracy to test a competition theory alongside well established variables, such as external influences (diffusion), internal factors within a state, and the impacts of time passage. Having established that direct dem ocracy affects public policy at the state level, that political competition exists and shapes policy agendas , and that political competition is one possible explanation for policy innovation, this research now moves to the exploration of impact on policy innovation. As direct democracy increases the number of players available to affect change policy, there is more competition in the policy arena. This research tes

PAGE 35

30 state level. As such, this research hypothesizes that the existence of direct democracy in the states, via increased political competition, will impact policy inn ovation at the state level.

PAGE 36

31 CHAPTER III RESEARCH QUESTION AND HYPOTHESES The overarching research question is: D oes the citizen initiative promote policy innovation at the state level ? Two hypotheses (see Table 2 ) will be tested. Hypothesis 1: D irect democracy , through increased political competition, is positively related to policy innovation in the states. Hypothesis 2: The difficulty of placing a question on a statewide ballot , with the magnitude of competition decreasing as access to the ballot i s more difficult, will be negatively related with higher levels of policy innovation. Each hypothesis is tested using a variety of policy innovations from two policy subsets at the state level : 1) matters of social policy, and 2) matters of fiscal policy. In exploring two policy subsets, this research allows one to potentially draw distinctions between different types of policies and the independent variables that may or may not influence poli cies within the subsets. Within the social policy subset, this dissertation uses medical marijuana policies, mandat ory bicycle helmets for minors , hate crime laws, and primary seat belt laws. For fiscal policies, this research explores strategic planning for economic development, state enterprise zones, family cap exemptions for welfare policy, and tax and expe nditure limitations (see Table 3 ).

PAGE 37

32 Each policy was selected b ) study on policy innovation, which looked at more than 180 policy innovations. In select ing these eight specific policies, several criteria were used. First, Proposition 13 in California, so policy innovations from 1978 to present day were isolated to captur e the period in American history when initiative use has been at its peak (Initiative Use, 2010). TEL policy (first adopted in 1976) fails this criterion, but TEL literature first hate crime and medical marijuana laws were adopted in 1978, so the first adoptions would not have been influenced during the peak days of initiative use (not to say that the policy was not influenced by the initiative at all) . B ut again , most of the se laws were adopted after 1978. For the remaining five policy variables, the first adoption took place in 1981 or later. Second , it was preferred that policy adoption occurred over at least a 10 year span so as to exclude trendy punctuations in policy ad option. This allows the research to look at more sustained policy adoptions over time. Adoptions of family cap exemptions over 21

PAGE 38

33 states occurred only during a seven year period, and it is the only policy selection outside of this criterion. To satisfy other concerns as noted above , it was necessary to include this exception. Third , it was also preferred that at least 15 states had adopted any given policy so that the sample size was sufficient. Fourth, f ringe or extreme policies were not selected as t o avoid extremely partisan and contentious issues. These might include gay marriage bans or parental notification laws for abortions by minors . This variety of policies over two policy subsets allows for the hypotheses to be evaluated across different po licy areas in the specified time frame of interest. This study also controls for political bias and other factors. This dissertation fills several research gaps. As detailed in the literature review, a majority of EHA and policy innovation research ha s focused on single policies. Further, the two studies (Walker, 1969; Boehmke & Skinner, 2012 a ) that took a holistic approach to studying innovation in the states failed to account for several areas of interest : There was no detailed look at how policy in novation may have varied in different time frames or in different policy subsets; neither looked at the citizen initiative specifically during the time frame when initiative use reached its peak; Walker (1969) did not account for the citizen initiative, wh ile Boehmke and Skinner (2012 a ) treated the initiative as a simple dichotomous variable; and both studies treated all policy innovations in a similar dichotomous fashion without accounting for repeating events. This research contributes to the literature by : Look ing framework;

PAGE 39

34 Provid ing competition was maximized through higher use of the initiative; Us ing a modified event history analysis that controls for the passage of time; E xplor ing multiple policies over two separate policy subsets to differentiate the effects the initiative may have in different policy areas; Exclud ing punctuated diffusion by looking only at policies adopted over an extended period of time; Treat ing the initiative as a dichotomous variable indicating the presence of direct democracy, but also taking a more in depth account into the varying rules and laws of ction; P rovid ing an in depth look at two policies where repeated adoptions were the norm across the states and over multiple decades. And exploring explanatory variables to policy innovation through the lens of political competition created by the presence and form of direct democracy in the states. The research in this dissertation relies heavily on the work of previous scholars in the policy innovation field, but also fills in many research gaps until now left unexplored. As scholars, lawmakers and prac titioners work to fully understand policy innovation in the states perspective from new angles.

PAGE 40

35 CHAPTER IV RESEARCH DESIGN This dissertation answers the resea rch question by exploring four policies each within two policy subfields fiscal policy and social policy for a total of eight policies . STATA is the primary tool used for analysis. The individual state is the unit of analysis with a fixed time period of one year, which is the accepted standard for research o n policy innovation in the states . There are two main phases of analysis : first, an event history analysis is completed for each of the eight policy variables; second , a re peating event analysis is completed for the tax and expenditure limitations variable and the medical marijuana variable. For each phase, data were collected for 48 states from 1976 through 2015 and does not consider the District of Columbia or any U.S. te rritory . Illinois is excluded because the initiative process there can only be employed to modify the organization of the state legislature, and fiscal initiatives are not allowed (Matsusaka, 2004). Wyoming is excluded for several reasons, most notably bec restrictions on revenue and appropriation measures (Matsusaka, 2004). Hawaii and Alaska are excluded from this study when exploring diffusion effects using the conti guous state measure of diffusion , but are included in the western region when using the fixed region diffusion measure . The contiguous state measure was used in early event history analyses to account for external influences on policy innovation. Later m easures used the fixed region diffusion measure , again to account for external influences while accounting for the idea that policy learning is n o t limited to only neighboring states.

PAGE 41

36 This section first explores the two phases of this study and explains the model used for each phase. This is followed by an in depth discussion of all the variables used for each phase. Phase 1 For phase one of this dissertation, e vent history analysis is employed on each policy individual ly to determine if the presence of a citizen initiative and the ease of placing a question on the ballot are explanatory factors for each policy innovation. This phase employs the Berry and Berry (1990) model of event history analysis with several modifica tions derived from Buckley and Westerland (2004) , as discussed in detail in the empirical section of the literature review . Specifically, these modifications include adding a linear time counter, replacing logit and probit with a cloglog function to accou nt for the disproportionately high number of 0s in the dichotomous dependent variables, and the clustering of standard errors by region. By using EHA , this study accounts for both internal and external i nfluences on policy innovation . T his approach captu res influences that may impact policy adoption in each state in each year to account for changing circumstances in each individual state. Adding a linear time counter to the Berry and Berry model (Buckley & Westerland, 2004) , the model for phase 1 is: ADOPT i,t = f (MOTIVATION i,t , RESOURCES/OBSTACLES i,t , EXTERNAL i,t , TIME COUNTER i,t .) , not included in early EHA models, was removed from this phase as identifying potential policies that may impact the policy variable in quest ion would be somewhat arbitrary and subjective. As utilized in previous studies, for example, Balla (2001) measured the impact of previously adopted HMO measures on the adoption of the HMO Model Act. Also, Soule and Earl (2001) studied hate crime policy

PAGE 42

37 adoption as a function of whether or not some form of hate crime law already existed in the state. Identifying and researching every potential policy in every state over 40 years that may impact the adoption of the eight policy variables researched here w ould have been time prohibitive , while being somewhat arbitrary and providing limited value in the context of this research given the number of independent variables already utilized. This minor limitation is overcome in the Phase 2 repeating events model , as it by definition account s for similar policies already in place for two policy variables. Following Buckley and Westerland ( 2004) , a complementary log l og function (cloglog) is employed for analysis rather than standard probit or logit, which were used in early EHA models given the dichotomous dependent policy variables. Further, P hase 1 corrects for bias in EHA standard errors caused by spatial autocorrelation (Buckley & Westerland, 2004) by cluster ing standard error based on region, as defined b y the U.S. Census Bureau , to account for regional similarities among the states. Phase 2 Jones and Branton (2005) focus on a failure of event history analysis treating all policies as non repeating events. In their model on lottery adoption, Berry and Be rry (1990) did focus on a policy that generally occurred only one time. H owever, some policy events are considered to be repeating. Notably, as analyzed in phase 1, Boehmke and Skinner (2012 a ) treat tax and expenditure limitations and medical marijuana l aws as non repeating events. In other words, a state drops out of the data set once the first policy adoption occurs. However, some states have adopt ed multiple TELs and medical marijuana laws over time. For example, Colorado alone has four TELs currently in place the 1977 Kadlecek Amendment, which imposes a 7 percent annual spending limit; the 1982 Gallagher

PAGE 43

38 Amendment, which caps residential property taxes; the 1991 Arveschoug Bird Amendment, a limit on general fund expenditures; and the Taxpayer Bill of Rights, approved by voters in 1992 (Martell & Teske, 2007). Also, some states have passed multiple marijuana laws over the years , with many states first approving a medical marijuana law and then refining the law to add restrictions and/o r expansions . When a policy innovation study only looks at the first policy adoption, it ignores information and provides potentially faulty conclusions (Jones and Branton, 2005). In looking at obscenity legislation across the states in both single eve nt and repeat ing event models, Jones and Branton (2005) found statistically inconsistent findings. In this light, P hase 2 of this dissertation looks at TEL policy and medical marijuana policy using This provides impact not only on policy adoption, but on repeated adoptions in the same policy area. This allows examination of the role of direct democracy in ongoing policy change. Jones and Branton (2005) use a stratified Cox model to study policy innovation, where the baseline hazard rate is left unspecified. Further, unlike other models of EHA, the Cox model is easily extended to handle repeating events. The Cox model is as follows: h i (t) = h 0 where h 0 (t) is the baseline hazard function and are the covariates and regression parameters. Jones & Branton, 2005, p. 424 ). To modify the model to account for repeating events, no state drop s out of the dataset onc e the first policy is adopted. multiple TEL laws as an example (see Table 4) , the conditional time counter measures the number of years since the last policy

PAGE 44

39 event. It starts at 1 in the first year and increases by one each year until another policy adoption. Then, the conditional time counter resets with a 1 in the year following another policy adoption. T he risk sequence indicates the ordering of policy events. It starts at 1. To model repeating events, the modified Cox model is stratified by the event number (risk sequence column) to preserve the ordering of events, and it defines time in the duration d ata set as the conditional gap time (conditional time column). Combined, these modifications create a conditional gap time Cox model (Jones & Branton, 2005). This is the model used for P hase 2 . Explanation of Variables For both P hase 1 and P hase 2 , a ll d ependent and independent variables are measured in each state in each year from 197 6 through 2015 . They are summarized in Table 5.

PAGE 45

40

PAGE 46

41

PAGE 47

42 Dependent Variables Policy innovation is captured in the eight policy innovations. These dependent variables are tested as a function of the presence of the citizen initiative, the difficulty of the literature (see Table 5 ) . The eight dependent variables are captured in ADOPT i,t . T hey are coded in a policy has been adopted in any given year (see Table 6) . Tax and Expenditure Limitations Tax and expenditure limitations (TELs) come in many forms spending limits, revenue limits, requiring voter approval or legislative supermajorities for tax increases but all attempt to limit government spending and/or government growth. Using updated f igures regarding states with a tax and expenditure limitation, 33 states have these policies in place (Wa isanen, 2010; New, 2001). About half of TELs are constitutional, with the other half statutory provisions. They have been passed through initiative, referenda, or legislative approval , and can be either statutory or constitutional (Waisanen, 2010). Four states passed TELs prior to 1976. These TELs are not considered here.

PAGE 48

43

PAGE 49

44 State Enterprise Zones State initiated enterprise zones (EnterZone s) are specific geographic areas that provide tax subsidies to businesses and are meant to spur development through tax incentives and regulator y relief (Wilder & Rubin, 1996) . This research collected policy adoption data from Wilder and Rubin (1996). Wh ile 39 states have adopted enterprise zones 6 , not all enterprise zones are currently active. Family Cap Exemptions Family cap exemption policies (FamCap) exempt a certain number of children from welfare family caps, which deny welfare aid to children born into families already receiving money through a TANF supported welfare program ( Gutiérrez, 2013). Many states exempt a certain number of children up to eight from this cap based on certain criteria. For example, many states allow exemptions in cases of rape or incest, while California also allows an exemption in cases of failed contraception (Karch, 2007). This research analyzes 19 states that have adopted this policy, using Karch (2007) as the source for states with policy adoptions. Strategic Plan ning (Economic Development) Strategic planning for economic development (EconDev) is defined ). First, a clear statement . Second, identification and rec ognition of external constituencies and their assessment of the agency . Third, an d objectives in a 3 5 year plan. Fourth, strategic development to achieve those goals and objectives. If these features were met, a stra tegic planning policy was considered to be 6 Including the District of Columbia

PAGE 50

45 adopted. Berry found 24 states had policy adoptions fitting these criteria. Due to the survey methodology, this research uses the policy adoption data set as provided by Berry (1994) . While the survey identifi ed nine areas of strategic development, this research focuses on strategic planning for economic development. Medical Marijuana Medical marijuana (MedMar) is defined as a policy that allows marijuana for medical use based on data from the Marijuana Policy Project Twenty two states have adopted a qualifying policy. This is more restrictive than the medical marijuana definition use d by Boehmke and Skinner (2012b) , which included state policies that were only a simple recognition of the benefits of medical marijuana. Policy adoption data were derived from the Marijuana Policy Project This policy variable, along with tax and expenditure limitations, is used in the repeating events policy analysis. Please see below for more details regarding medical marijuana and repeating policy adoptions. Hate Crime Laws Simply put, hate crime laws ( HateCrime) motivated crime laws derived from the perception of increased racial, ethnic, and religious forms of conflict, among other categories, in the 1970s. Through the late 1980s and 1990s, states began to push legislating hate crime laws (Grattet, Jenness & Curry, 1998). Data were collected from Grattet, Jenness and Curry (1998). California was the first state to pass a hate crime law , and this research analyzes 32 states with hate crime policies . Across the

PAGE 51

46 states, these laws take the form of criminal penalties to penalty enhancements on existing law. Primary Seat Belt Laws A primary seat belt law (PrimSeat) allows law enforcement to give a driver a citation for not wearing a seat belt. Updating the Boehmke and Skinner (2012b) statistics used in their paper, there are 25 states plus Puerto Rico and the District of Columbia with primary seat belt laws (NHSTA, 2006). Dat a are collected from the National Highway Traffic Safety Administration (NHSTA, 2006). Twenty four states carry a secondary enforcement provision, which allows a citation to a driver for not wearing a seat belt only if another violation occurred. These s econdary laws are not included in this research . Mandatory Bicycle Helmets for Minors Mandatory bicycle helmet laws for minors (KidHelmet) are laws that require helmet use for minors while riding a bicycle. From the same source as Boehmke and Skinner (20 12b) but updated, there are 22 states with mandatory bicycle helmet laws for minors (Bicycle Helmet Safety Institute, 2016). The age limit for mandatory helmet use varies from those under 12 years of age to under 18, though some local laws cover all ages. All of these laws are included regardless of the age considerations. Some states also have passenger only helmet laws , which are not included . The year of policy passage here includes laws for minor riders, not for minor passengers. Three states with passenger laws California, Massachusetts, and New York also have laws covering riders. Independent variables The TIME COUNTER variable (TD) in the modified Berry and Berry model for P hase 1

PAGE 52

47 adding 1 unit for each year of observation. In P hase 2 , this variable is captured again by starting at 1 and increasing one unit for each year passed. However, the co ndit ional time counter used in P hase 2 reset s after every policy adoption or policy change. External V ariables These variables capture the diffusion effects on policy innovation. This research captures both contiguous state diffusion (CS) and fixed region diff usion (FR) . For the former, the variable is captured by counting the percentage of contiguous states that have adopted the policy in question , necessarily excluding Hawaii and Alaska. U sing a series of formulas in Excel , a percentage of states with the re levant policy were calculated for each state . For fixed region diffusion , the states are divided into regions as defined by U.S. Census Bureau ( Census Regions and Divisions of the United States , 2016) . Using Excel formulas, the percentage of states in each region that have adopted the policy w as calculated . Motivation V ariables These variables capture internal characteristics of a state; specifically they indicate the motivation of public officials to pass a policy innovatio n. In this study, MOTIVATION is captured by several partisan and election related factors. Walker (1969) found an increase in legislative rural urban competition was positively related to innovation. He also concedes that parties that face tight elections would be more apt to promote policy innovation. Single party dominance of both chambers in the legislature make it more likely that a party is able to control the policy agenda (Stigler, 1972; Crain & Tollison, 1976), creating an easier path to policy in novation. However, prolonged political dominance has also been found to reduce political competition (Bueno de Mesquita, Morrow, and Smith, 2001) and therefore policy innovation would be expected to decrease.

PAGE 53

48 To capture these phenomena, this research meas ures the proximity to gubernatorial elections (GE) . These data were collected from The Green Papers ( All Up Chart of G overnors by Election Cycle, 2012) . It also captures partisanship using a 0 3 scale for Republican control (PS) , with 1 point each for con legislative chamber, and the upper legislative chamber. These data were collected from the National Conference of State Legislatures ( State Partisan Composition, 2016 ) and from Wikipedia data regarding the state h istory of governors by state ( List of Current Un ited States Governors, 2016). This study also measures prolonged party dominance by capturing the consecutive number of years a single party has controlled the governorship (GPC) , both chambers of a legislat ure (LMC) , and both at the same time (CPC) using the same data sources . P rolonged dominance is expected to reduce competition. Resources/Obstacles V ariables The independent variables of interest the presen ce of direct democracy (DD) and the difficulty of ballot placement (DDi) are captured in RESOURCES/OBSTACLES . The simple presence of direct democracy in a state is a dichotomous variable. These data are collected from the Initiative and Referendum Institute (2011) . The measure for difficulty of ballot placement is derived from an index in Bowler and Donovan (2004) between 0 and 6, with 0 representing the least difficult requirements. Non initiative states are scored a 10, following Bowler and Donovan (2004) . Each state has different signature requirements for placing a question on the ballot, and some states require a certain number of signatures to be gathered in specific geographic locations within the state. Furth er, some states limit the time available to collect signatures (Bowler & Donovan, 2004). Banducci

PAGE 54

49 (1998) and Matsusaka and McCarty (2001) found that stricter signature requirements reduce the number of initiatives in a state. It makes sense, then, that s tricter signature requirements will reduce competition and therefore policy innovation. These variables also look at other internal characteristics of a state. As legislative competition not only exists between the parties, but also between urban and rur al (RP) is captured as a percentage of total state population . Rural population data were collected from the U.S. Census Bureau via Iowa State University ( Urban Percentage of the P opulation for States, Historial, 2010 ). Data were available by decade (1970, 1980, 1990, 2000, and 2010), so, for example, 2000 figures were used in the data from 2000 through 2009. Wealth by median household income (W) is a measure of median household income for each state. Data were collected from the U.S. Census Bureau ( Historical Income Tables: Households , 2016) . Data for all years were not available. As such, data figures from 1969 were used for 1976 through 1978. Data figures from 1979 were us ed from 1979 through 1983. Annual figures were available from 1983 forward. Total state population (P) was again captured from data from the U.S. Census Bureau ( Population and Housing Unit Estimate , ( 2016 ). Population figures were available for each yea r in this research. Economic conditions (EC) in the states were captured by tax revenue per capita. First, U.S. Census Bureau data regarding state historical tax collections were captured ( State Government Tax Collections , 2016 ). These data were availab le for all years. The tax revenue figures were then divided by the population figures to attain a measure of tax revenue per capita.

PAGE 55

50 I t has been found that states with higher levels of legislative professionalism contributed to policy innovation in e gov ernment ( Tolbert, Mossberger, & McNeal , 2008) . Therefore, t his dissertation captures each level of professionalism (LP) using an index by Squire (2007) , which is based on measures of salary and benefits, length of legislative sessions , and a vailability of staff and resources . Squire (2007) provides ranks for 1979, 1986, 1996, and 2003 for each state. The 1979 score is used for the data years 1976 through 1985, the 1986 sc ore from 1986 through 1995, the 1996 score from 1996 through 2002, and the 2003 score from 2003 through 2015. Finally, this research uses a measure of education in each state based as a percentage (EDU) . Data were available by decade from the U.S. Census Bureau ( A H alf Century of Learning: Historical Census Statistics on Educational Attainment in the United States, 1940 to 2000: Detailed Tables, 2016; Am erican Community Survey, 2010 1 Yea r Estimates, Table S. 1501, 2016). Data from 1970 were used from 1976 through 1 979 , data from 1980 were used from 1980 through 1989 , data from 1990 were used from 1990 through 1999 , data from 2000 were used from 2000 through 2009, and data from 2010 were used from 2010 through 2015. Dependent V ariables for Repeating Events Model Agai n, Phase 2 of this dissertation explores policy adoptions of tax and expenditure limitations and medical marijuana laws using a repeating events model. Unlike Phase 1, where all policy adoptions are treated as single events and a state drops from analysis once the first policy adoption occurs, the repeating events model considers all policy adoptions of an issue from 1976 to 2015. Table 7 details the number of policy adoptions by policy and by state. Table 8 provides a vertical timeline of policy adoptio ns by year.

PAGE 56

51

PAGE 57

52

PAGE 58

53 Fifteen states have adopted more than one TEL , which constitutes 19 repeating events. Colorado and Oregon each have four TELs . detailed above. Oregon first passed a TEL in 1979, passed by the state legislature, which required immediate refunds of surplus revenue to taxpayers. The refund provision was established in the state Constitution in 2000, and the next year was modified by the legislature (New, 2010) . Further, Oregon , by referendum in 1996, approved the requirement of a three fifths legislative supermajority to raise taxes (Wais anen, 2010) . Thirteen other states have adopted two TEL policies. Arizona passed a constitutional spending limit in 1978 and a legislative supermajority requiremen t to raise taxes in 1992. Connecticut passed a statutory spending limit in 1991 and a constitutional limit in 1992. Delaware in 1978 passed a constitutional spending limit, adding a legislative supermajority to raise taxes in 1980. Louisiana enacted its first TEL legislatively in 1979 and added a constitutional spending limit in 1993. Massachusetts passe d a spending limit in 1986 , which raise taxes was passed in 1994 regarding state property taxes. Mississippi passed two spending limits in 1982 and 1992. Missouri passed a constitutional revenue limit in 1980, and added a constitutional measure requiring voter approval for certain tax increases. Nevada passed a stat utory spending limit in 1979, and through the initiative process added a first TEL in 1976, which expired in 1983. The legislature approved a new spending limit in 1990. Oklahoma in 1985 legislatively approved constitutional spending and appropriations limits (approved through the referenda procedures), and later approved an initiative in 1992 requiring a legislative supermajority to raise taxes. In 1980 and 19 84, South Carolina

PAGE 59

54 approved constitutional spending limits. And finally, Washington approved a TEL through the initiative process in both 1979 and 1993 (New, 2010 ; Waisanen, 2010 ). Eighteen states have adopted one TEL, while 15 states do not have any TELs. All states , regardless of the number of policy adoptions , are analyzed in this research. Regarding medical marijuana, 13 states have adopted multiple policies constituting 27 repeating events . The initial wave of medical marijuana laws were approve d by initiative, but currently 12 of the 2 2 laws were legislatively enacted laws with Hawaii being the first in 2000 by . Of the 13 states with repeating events, Main e has the most repeating events with a total of five policy adoptions by In Maine, the policy was first adopted at the ballot in 1999 and has the most repeating events with five total policy adoptions. In 2001 , the legislature clarified protections for patients and caregivers while also increasing the limit for marijuana possession. Another successful ballot question in 2009 established a patient and caregiver registry while also establishing dispensaries. Th ere were two additional legislative changes in 2010 and 2011 by Four states have four policy adoptions. Nevada, though initiative, passed constitutional ballot measures in 1998 and 2000 to require the legislat ure to implement medical marijuana laws. Two legislative enactments in 2001 and 2013 separately removed criminal penalties and allowed for dispensaries. enacted by the legislature in 2006, 2007, 2009, and 2012. The initial provision removed state level criminal penalties for medical marijuana. Subsequent amendments created

PAGE 60

55 legislature, medical marijuana registration was only av ailable to people with certain diseases, including AIDS and cancer. Subsequent amendments softened eligibility requirements and was passed by initiative. Subsequen t legislative changes in 2010, 2011, and 2015 allowed patients and caregivers to collectively grow marijuana, allowed professionals to legally recommend marijuana for treatment, and created a voluntary registry by State 5) . Three states have three policy adoptions. California, by ballot initiative, was the first state to allow for medical marijuana use in 1996, effectively removing criminal penalties for patients who receive a doctor recommendation. In 2003, the California legislature expanded by was legislatively adopted (to be phased in over several years) replacing the collective/cooperative model with a dispensary licensing system. New Jersey has one of the most restrictive laws in the country, requiring mandatory ID cards for patients and caregivers . The initial law, passed legis latively in 2010, also developed a state regulated marijuana law was first approved at the ballot in 1998 and removed state level criminal penalties. Legislative amendmen ts in 2005 and 2013 increased the plant and possession limits, and also established a registry system by 2015) . F ive states have two policy adoptions . Colorado approved Amendment 20 in 2000, with the medical marijuana law taking effect in 2001. In 2010 the legislature adopted

PAGE 61

56 dispensary licensing procedures. Hawaii, the first state to adopt a medical marijuana law in 2000 through the legislature , modif ied the law in 2015 allowing for civil protections and regulations for patient access to marijuana . Alaska approved a ballot measure in 1998 removing state level criminal penalties, and a legislative bill in 1999 implemented a mandatory registry program. Maryland legislatively removed state level criminal penalties in 2014, and was amended in 2015. Finally, Montana approved medical marijuana by initiative in 2004. In 2011, the legislature made it more difficult to receive recommendations and placed othe r regulatory policies in place ( by 2015). Nine states have adopted one medical marijuana , while 28 states do not have any medical marijuana policy. All states, regardless of the number of policy adoptions, are analyzed in this research.

PAGE 62

57 CHAPTER V RESULTS The results overview in this section cover s both phases of analysis. Phase 1 analyzes single policy adoptions for 8 policy variables, while Phase 2 analyzes repeating events for two policy variables (TELs and medical marijuana). Both phases of this dissertation focus on two independent variables of interest . To reiterate, the simple presence of direct difficulty index established by Bowler & Donovan (2004), giving states a higher score as it becomes more difficult to place a question on the ballot. Hypothesis 1: Direct democracy is positively related to policy innovation in the states. Hypothesis 2: The difficulty of placing a q uestion on a statewide ballot will be negatively related with higher levels of policy innovation. Twelve analysi s scenarios (see Table 9) with different combinations of the independent variables were run for each phase six combinations using the dichot omous variable for the existence of direct democracy, and the same six combinations using the direct democracy index variable . For the diffusion component, the contiguous states measure ( percentage of bordering states having already adopted the policy in question) and fixed region measure ( percentage of states within the same region having already adopted the policy in question) were necessarily run separately . Finally, there are three similar variables for partisanship: 1) Consecutive years of single par (GPC), 2) c onsecutive years a single party controls both chambers of a legislature (LMC), and 3)

PAGE 63

58 c onsecutive years office (CPC). For this dissertation, three combinations of the partisanship variables were considered: including all three variables, including only CPC, and a third excluding CPC . These scenarios were developed since CPC is similar to the combination of LMC and GPC. The first four scenarios capturing all three variables are able to capture single party control of indiv idually , while the CPC variable informs whether the same party or different parties controlled both . Phase 1 Phase 1 looked at each of the 8 policy variables as single event adoptions. Once a state adopted the policy in question, the state dropped out of analysis. In other words, once As such, for each of the 8 dependent variable s , the range of years in the analysis varied for different states , so the descriptive statistics for all variables (dependent and independent) also

PAGE 64

59 varied depending on the dependent variable . For a full table of the descriptive statistics for all variable s for each of the 8 analyses, please see Appendix B. Tables 10 t hrough 21 present the results for all 8 dependent variables for each of the 12 scenarios (for a total of 96 sets of results) , showing the estimates , standard errors , and log pseudolik e lihood values . Independent variables found to be related to policy adop tion at a statistically significant level are highlighted with asterisks. There are four measures of statistical significance: p values of <0.001, <0.01, < 0.05, and <0.1. The log likelihood value is not used as an index of fit as it is a function of sa mple size, but can be used to compare the fit of various models relative to each other, where a value closer to zero is desired ( Interpreting Log Likelihood, 2017 ). When clustering standard errors, STATA provides a log pseudolikelihood value ( Log Pseudo L ikelihood, 2004 ). Table 22 summarizes by scenario the independent variables that show any level of statistical significance for each of the eight dependent variables. Table 23 summarizes the number of times each independent variable shows a significant relationship to each policy adoption. These tables give a visual represen tation of the relative statistical significant of the independent variables on policy adoption . This section highlight s the main findings from the Phase 1 analysis by reviewing findings from the 8 dependent variable policies within scenario pairs, pairing the scenarios with the same measures of direct democracy and partisanship, but differing diff usion m easures . It makes sense to report analysis findings by scenario pair in order to begin parsing out the relevant findings by variable type. For example, by analyzing scenarios 1 and 2 together, the dependent variables, the direct democracy variable , and all other independent

PAGE 65

60 variables except the measure of diffusion remain common to the two scenarios. In this way, the impact of the diffusion measures on policy can be viewed in isolation. Overall, findings are mixed regarding the role played by either measure of direct democracy on policy adoption. Throughout the various scenarios, the impact of direct democracy was little influenced by the partisanship measure used. Switching up the partisanship measure alter ed the outcome only for the index measure of direct democracy and only for four total scenarios relating to medical marijuana and TELs. The impact of both measures of direct democracy was influenced by the diffusion measures only when using the fixed region mea sure of diffusion and only when analyzing TEL policy, as the direct democracy measures were only statistically significant for TELs when paired with the fixed region measure. In total, 26 policy adoptions are statistically predicted by either measure of d irect democracy in the direction hypothesized, out of a total of 96 possibilities (12 scenarios each for 8 policies). More details on these follow the summary results by scenario. Further, overall findings show that diffusion plays the largest role in predicting the passage of policy innovation. Out of the 96 analyses, 69 show that diffusion has a statistically significant impact on policy adoption. More specifically, the fixed re gion measure of diffusion is able to predict policy innovation in 43 of the 48 analyses on a statistically significant level.

PAGE 66

61

PAGE 67

62

PAGE 68

63

PAGE 69

64

PAGE 70

65

PAGE 71

66

PAGE 72

67

PAGE 73

68

PAGE 74

69

PAGE 75

70

PAGE 76

71

PAGE 77

72

PAGE 78

73

PAGE 79

74

PAGE 80

75 Scenarios 1 and 2 Scenarios 1 and 2 model the independent the presence of direct democracy, all measures of partisanship, and allowing the measure of diffusion to vary between the contiguous state measure (scenario 1) and the fixed rate measure (scenario 2). Across the two scenarios, direct democracy has a statistically significant impact on policy adoption in about half of the policy areas, though there were slight differences in the results for the different measures of diffusion. The log pseudolikelihood measure vari es only slightly for each analysis across the scenarios. In scenario 1, under the contiguous state diffusion measure, the presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure poli cies and med ical marijuana policies . Under the fixed region diffusion measure (scenario 2) , again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs , family cap expenditure policies, and medi cal marijuana . Under the fixed region diffusion measure, one additional policy was statistically related to the presence of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation. Unexpectedly, under both scenarios 1 and 2, the presence of direct democracy was negatively and significantly related to the adoption of primary seat belt laws and kid helmet laws. Running against the hypothesis stated, the presence of direct democracy is a deterrent to adopting these safety minded policies. This finding is explored further in the discussion section of this dissertation. The influence of the partisanship variables on the impact of direct democracy on policy adoption is discussed in the subsequent analyses of the scenario pairs.

PAGE 81

76 This research expects to find that both diffusion measures would be positively related to policy adoption. Results show the contiguous state diffusion measure was positively and significantly related to the policy adoption of four poli cies enterprise zones , economic development, medical marijuana, and kid helmet laws . The fixed region measure of diffusion was positively and significantly related to the policy adoption of six policies, including the same fo ur measures just noted plus hate crimes and primary seat belt laws . Regarding partisanship, the rese arch expects to find that prolonged single party dominance of any branch of government would be negative ly related to policy adoption, as lack of politic al competition would reduce the motivation to adopt innovation policies. Consecutive years of single , however, was positively and significantly related to the adoption of three pol icies under scenario 1: TELs, econ omic development, and medical marijuana . This variable was positively and significantly related to the adoption of two policie s under scenario 2: TELs and medical marijuana . Consecutive years of single party control of both chambers of a state legislat ure was also positively and significantly related to the adoption of three policies under scenario 1: family cap expenditures, hate crime legislation, and kid helmet laws . Under scenario 2, this variable was positively and significantly related to only on e policy va riable, kid helmet laws. Consecutive years of single party control over both chambers of the legislature was negatively and significantly related to the adoption of four policies under scenario 1: economic devel opment, medical marijuana, hate crime laws, and kid helmet laws . Interestingly, even where the single party control of either office or the legislature was positively related the policy adoption, when a single party showed sustained control of both

PAGE 82

77 the policy adoption turned negative . This last measure of partisanship supports the expected findings that prolonged single party control would act as a discouragement to policy innov ation. Results also indicate that having single party control over just one branch of government does not necessarily act as a deterrent to policy innovation, as political competition could still exist from a different branch of government. Within scenario 1, several of the other independent variables showed a significant economic conditions were also significantly related with four policy variables (two positive and two negative), alon g with education (one negative and three positive). Other variables showed significance with fewer policy adoptions. Within scenario 2, th e Republican measure was again significantly related to four policies (one positive and three negative). Other variables showed a significant relationship with a few er number of policy variables. In summary, u nder scenarios 1 and 2, results show that dif fusion measures more frequently had a statistically significant impact on policy adoption (10 of 16 analyses) than measures of direc t democracy ( 5 of 16 analyses as hypothesized ), with the fixed region measure specifically being the most accurate predictor of policy adoption (6 o f 8 analyses). Scenarios 3 and 4 Scenarios 3 and 4 are the same as scenarios 1 and 2, except that the measure of direct democracy is the difficulty index. As the difficulty to place a question on the ballot increases based on va rying state rules, the index score also increases (between 0 and 6, with non initiative states receiving a 10 score) . As such, this research hypothesizes that the direct

PAGE 83

78 democracy index score is negatively related to policy adoption. Again, the log pseud olikelihood measure s var y only slightly for each analysis across the scenarios. In scenario 3, coupled with the contiguous state diffusion measure, the direct democracy index was , as hypothesized, negatively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 4) , again as hypothesized, the direct democracy index was negatively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the direct democracy index , but in total only five of the 16 measures under thes e two scenarios support the hypothesis in this dissertation. This mirrors the results from scenarios 1 and 2. Under these scenarios, altering the measure of direct democracy does not change where direct democracy is significantly related to policy adopti on in support of the research hypotheses. Using the direct democracy index variable again showed a significant relationship with the adoption of primary seat belt laws and child helmet laws under bother scenarios, and again in the opposite direction hypoth esized (this time a positive relationship) . Here, the increasing difficulty to place a question on the ballot acts as a predictor of policy adoption, rather than a deterrent to policy adoption. Because this research scores non initiative states as a 10 o n the difficulty index, this finding indicates that non initiative states are more likely to adopt these safety measures, which is a conclusion that would be consistent with the findings in scenarios 1 and 2. The diffusion measures showed the same levels o f statistical significance with the adoption of the same policy variables as scenarios 1 and 2 (relative to scenarios 3 and 4,

PAGE 84

79 respectively), except th at the fixed region diffusion measure under scenario 4 also showed significance with the adoption of TELs , but only at the <0.1 level. This shows that the measure of direct democracy being used does not substantially affect the findings relative to The findings regarding the three measures of partisan ship were also larg ely unchanged from scenarios 1 and 2 . In scenario 4, single office and the legislature was significantly and negatively related to the adoption of TELs, where it was not under scenario 2. In an other difference, the singl e party control over the legislature was significantly and positively related to the adoption of primary seat belt laws. The direction of the relationship among the partisanship variables was unchanged from scenarios 1 and 2, so regarding these variables, the same conclusions can be drawn as noted above. In summarizing th e other independent variables, scenario 3 showed three independent three negative), economic c onditions (two in each direction), and education (three positive and one negative). In scenario 4, two independent variables showed a significant relationship with four policies: Republican control (one positive, three negative), and economic conditions ( two in each direction). Other variables showed a significant relationship with fewer policy adoptions. In summary, under scenarios 3 and 4, results show that diffusion again proved to be the most accurate predictor of policy adoption, this time for 11 of 16 analyses. The fixed region measure of diffusion again proved to be the most accurate predictor of policy adoption

PAGE 85

80 (7 of 8 analyses). Like under scenarios 1 and 2, the index measure of direct democracy was significantly related in the direction hypothe sized in 5 of the 16 policy analyses. Scenarios 5 and 6 Scenarios 5 through 8 mirror scenarios 1 through 4, respectively, except that scenarios 5 through 8 only include the partisanship measure of a single party controlling both measure of single party control of only party control of only the legislature. Scenarios 5 and 6 use the presence of direct democracy measure coupled with the contiguous state and fixed region measure s of diffusion. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios. In scenario 5, under the contiguous state diffusion measure, th e presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 6), again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs, family cap expenditure policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the presenc e of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis in this dissertation. Again, these scenarios also find a negative relationship (against the stated hypotheses) with the adoption of primary seat belt laws and kid helmet laws. Overall, r elative to the measures of direct democracy from scenarios 1 and 2, there is no change to the policy adoptions that are statistically predicted by the presence of direct democracy, including the primary seat b elt law and kid helmet law policy adoptions being negatively related at a

PAGE 86

81 statistically significant level. Even with removing the two measures of single party The contiguous state diffusion measure showed positive significance with five of the eight policy measures (enterprise zones, economic development, medical marijuana, hate crimes and kid helmets). Under scenario 6, the fixed region measure of diffusion sh ows a positive and statistically significant relationship with policy adoption in all eight areas at least at the <0.05 level. This supports the expected finding that states learn from other policy adoptions by other states and are not restricted to learn ing only from bordering states. Using only the partisan measure of a single office and both chambers of the legislature, this measure only shows a negative and significant relationship with one policy adoption (medi cal marijuana) under the contiguous state scenario, and two policy adoptions (economic development and medical marijuana) under the fixed region scenario. In scenario 5, two other independent variables showed a significant relationship with five policy adoptions: Republican control (one positive and four negative), and economic was significantly related to the adoption of four policy variables (three positive and one negative). In scenario 6, Republican control again was significantly related to five policy adoptions (all negative), and legislative professionalism with four policy adoptions (three positive and one negative). Other variables showed a significa nt relationship with a fewer number of policy adoptions. In mirroring the results from scenarios 1 and 2, the existence of direct democracy played a significant role as hypothesized in five of the 16 policy adoption analyses. The

PAGE 87

82 fixed region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role in all eight policy analyses. The contiguous state measure was a predictor of policy adoption in five of eight analyses. Overall, diffusion played a significant role in 13 of 18 analyses. Scenarios 7 and 8 Scenarios 7 and 8 use the direct democracy index measure using the contiguous state and fixed region measures of diffusion. Like scenarios 5 and 6, scenarios 7 and 8 only include the partisanship measure of a single party controlling both both chambers of the legislature. These scenarios remove the measure of single party control of only party cont rol of only the legislature. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios. In scenario 7, under the contiguous state diffusion measure, the direct democracy index was, as hypothesized, negatively and signifi cantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 8), again as hypothesized, the direct democracy index was again negatively and significantly related to th e adoption of family cap expenditure policies and medical marijuana. and kid helmet laws. The index showed a significant and positive relationship with the adoptions of these two policies against the stated hypotheses. When paired with the fixed region measure of diffusion, and i n contrast to scenario 4 , scenario 8 does not show that the direct democracy index has a significant relationship with

PAGE 88

83 the adoption of TEL poli cies. By removing two measures of single party partisanship, the Under scenario 7, the contiguous state measure of diffusion is significantly and positively related to five policy adoptions enter prise zones, economic development, medical marijuana, hate crime legislation and kid helmet laws. Like scenario 6, the fixed region diffusion measure in scenario 8 showed a positive and statistically significant relationship with policy adoption for all e ight policy variables at no less than the <0.05 level. In these scenarios, there is only one measure of partisanship used the single party is significantly rel ated to five of the 16 policy analyses. As expected, this partisanship measure showed a negative relationship with medical marijuana under both scenarios, along with TELs and economic development under scenario 8. However, a gainst expected findings under scenario 8, the partisanship measure was positively and significantly related to t he adoption of kid helmet laws . In scenario 7, two other independent variables show a significant relationship with the adoption of five policy areas: Republican control (one positive and four negative), and economic conditions (two positive and three negative). Legislative professionalism was significantly related to four policy adoptions (three positive and one negative). Under scenario 8, Republican control predicts t he adoption of five policy areas (all negative), and legislative professionalism four policy areas (three positive and one negative). Other variables showed a significant relationship with a fewer number of policy variables. Under scenarios 7 and 8, the direct democracy index plays a significant role as hypothesized in four of the 16 policy adoption analyses. This is one fewer compared to

PAGE 89

84 scenarios 3 and 4, as the index is no longer significantly related to the adoption of TEL policies. The fixed region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role in all eight policy analyses. The contiguous state measure was a predictor of policy adoption in five of eight analyses. Overall , diffusion played a significant role in 13 of 18 analyses. Scenarios 9 and 10 Scenarios 9 through 12 remove the partisanship variable used in scenarios 5 through 8 (single and reinstate the separate measures of single legislature. Scenarios 9 and 10 used the existence of direct democracy measure paired with the contiguous state and fixed region measures of diffusion. The log pseudolik elihood measures vary only slightly for each analysis across the scenarios. In scenario 9, under the contiguous state diffusion measure, the presence of direct democracy was, as hypothesized, positively and significantly related to the adoption of family cap expenditure policies and medical marijuana policies. Under the fixed region diffusion measure (scenario 10), again as hypothesized, the presence of direct democracy was positively and significantly related to the adoption of TELs, family cap expenditu re policies, and medical marijuana. Under the fixed region diffusion measure, one additional policy was statistically related to the presence of direct democracy, but in total only five of the 16 measures under these two scenarios support the hypothesis i n this dissertation. There is no change regarding direct negative relationship with primary seat belt laws and kid helmet laws, significantly related but in the opposite direction hypothesized.

PAGE 90

85 By again altering the partisanship variables use innovation is unchanged. Throughout these findings, the presence of direct democracy is insensitive to the changing combinations of partisanship variables. The contiguous state diffusion measure in scenario 9 showed positive significance with four of the eight policy measures (enterprise zones, medical marijuana, hate crimes and kid helmets). Under scenario 10 , the fixed region measure of diffusion shows a positive and statistically significant relationship with seve n of the eight policy variables, excluding only family cap expenditures. The single to the adoption of TELs under both scenarios , against expected results . The measure of sing le party control of the legislature holds a significant and negative relationship with medical marijuana under both scenarios. However, it holds a significant and positive relationship with family cap expenditures (scenario 9) primary seat belt laws (scen ario 10), and kid helmet laws (both scenarios). To reiterate, the research expected the partisanship variables to be negatively related to policy innovation. In scenario 9, two other independent variables show a significant relationship with four polic y adoptions, population and education (three positive and one negative for each). Under scenario 10, three independent variables show a significant relationship with four policy adoptions: Republican control (one positive and three negative), legislative professionalism (three positive and one negative), and education (three positive and one negative). Other variables showed a significant relationship with a fewer number of policy variables.

PAGE 91

86 In mirroring the results from scenarios 1, 2, 5 and 6, the exis tence of direct democracy played a significant role as hypothesized in five of the 16 policy adoption analyses. The fixed region measure of diffusion continues to be the most accurate predictor of policy adoption, playing a significant and positive role i n seven of the eight policy areas. The contiguous state measure was a predictor of policy adoption in four of eight analyses. Overall, diffusion played a significant role in 11 of 18 analyses. Scenarios 11 and 12 Scenarios 11 and 12 mirror scenarios 9 and 10 in using the single party control measure of direct democracy. Like all of the other analysis pairs, these scenarios use the contiguous state and fixed region measures of diffusion. The log pseudolikelihood measures vary only slightly for each analysis across the scenarios. When paired with the either measure of diffusion, the direct democracy index is significantly related to policy adoption in the hy pothesized direction for only one policy area, family cap expenditures. Under the previous scenarios using the direct democracy index, there was a significant and negative relationship with medical marijuana policy with both diffusion measures . That is n o longer the case in scenarios 11 and 12. Further, in contrast to scenario 4 but mirroring scenario 8, the direct democracy index is not a significant predictor of TEL policy adoptions. These findings suggest that while the existence of direct democracy predictive value of policy innovation is impacted by the changing measures of partisanship.

PAGE 92

87 Under scenarios 11 and 12, the direct democracy index still remains significantly and positively related (against the hypotheses) to the adoption of primary seat belt laws and kid helmet laws. Under scenario 11, the contiguous state measure of diffusion is positively related to the adoption of four policies. The fixed region measure of diffusion remains the driving force of policy innovation in these analyses, being statistically significant with the policy adoption of seven of the eight policy areas. Single adoptions under both scenario s, and in a positive direction . The measure of single party control of the legislature holds a significant and negative relationship with medical marijuana under both scenarios. However, it holds a significant and positive relationship with family cap expenditures (both scenarios) prima ry seat belt laws (scenario 12), and kid helmet laws (scenario 11). In scenario 11, two independent variables show a significant relationship with four policy adoptions: population and education (three positive and one negative for both). In scenario 1 2, three independent variables show a significant relationship with four policy adoptions: Republican control (one positive and three negative), legislative professionalism, and education (three positive and one negative for each). Other variables showed a significant relationship with a fewer number of policy variables. In summary, under scenarios 11 and 12, the direct democracy index holds a significant relationship with policy adoption in one policy area in the direction hypothesized. Diffusion again is the most significant predictor of policy adoption, showing significance in

PAGE 93

88 11 of the 16 policy analyses ( four for the contiguous state measure, and seven for the fixed region measure). Summary of Phase 1 Results This dissertation hypothesized that the mere presence of the initiative would be positively and significantly related to policy innovation, though this is only true for two policy variables in every scenario, and true for the adoption of TELs only when viewing diffusion through the fixed region measure . Regarding the direct democracy index, this research hypothesized a negative relationship with policy innovation. This held true for adoption of family cap expenditure policies under all six scenarios, for medical marijuana policies under four of six scenarios, and for TELs under one scenario. These findings are discussed in greater detail below. Overall, results in these analyses show that policy diffusion is the main driving force behind policy innovation, especially the fixed region measure of diffusion. The various measures of partisanship, overall, had less impact on policy innovation, and the direction of the relationship across policy adoptions was i nconsistent. These findings suggest that diffusion plays a greater role in policy innova tion than does the presence of an initiative process in the states. H owever, the two measures of direct democracy do significantly impact policy innovation in certain policy areas, partially supporting the hypotheses . Tables 2 4 and 2 5 show the p values, when significant, resulting from the measures of direct democracy on each policy innovation in this research. These tables show that direct democracy measures do impact policy adoption in the hypothesized direction for TELs, family cap expenditures,

PAGE 94

89 and medical marijuana policies. The significant relationship with primary seat belt laws and kid helmet laws, as noted before, are opposite the direction hypothesized. Findings support, at least for three of the eight policy areas , that the initiative process does play some role in poli cy adoption in the hypothesized direction. For TELs, the impacts are only apparent when viewing the models with the fixed region diffusion measure. This research intentionally focused on four economic po licies and four social policies to potentially allow one to draw distinctions between different types of policy. Yet, the results fail to show consistent patterns within policy types. In the economic variables, the measures of direct democracy impact poli cy adoptions for TEL and family cap expenditure policies, but show no significant relationship with enterprise zones o r the adoption of economic development measures. Regarding social policies, direct democracy plays a role in the adoption of medical mari juana policies , along with an adverse role in the adoption of primary seat belt laws and child helmet laws. The

PAGE 95

90 lack of consistent findings makes it hard to draw any definitive conclusions from this research, but is discussed in further detail in Chapter VI. While Phase 1 explores single a , Jones and Branton (2005) found statistically inconsistent findings when analyzing the policy adoption of obscenity legislation i n the states while using both single event and repeating event models. As such, Phase 2 of this research provides a special focus on TEL policy and medical marijuana policy by analyzing these depend ent variables through a repeating events model. Each of these policy variables, as detailed above, show s multiple policy adoptions in multiple states. Where the Phase 1 research drops a state from the analysis once a given policy is adopted, Phase 2 takes into account the first and every subsequent policy adoption related to TELs and medical marijuana. The same 12 scenarios from Phase 1 are duplicated in the Phase 2 analysis. Jones and Branton (2005) found conflicting results when viewing policy adoption through single event and repeating event models; this research finds similar inconsistencies when explo ring repeating policy adoptions of TEL and medical marijuana policy. Phase 2 In P when modeled alongside the fixed region measure of diffusion , were significantly related to the adoption of TEL polic ies in the direction hypothesized. There was no statistically significant relationship with TEL policy when using the contiguous state measure of diffusion. Furthermore, neither measure of direct democracy was related to repeating event poli cy

PAGE 96

91 adoptions of medical marijuana in a statistically significant way regardless of the diffusion measure used. The measures of fixed region diffusion were statistically significant in every model of TEL adoption, but no measure of contiguous state diffusi on was statistically significant. Regarding the variables of partisanship, no discernible patterns emerged to assess its impact on policy innovation. More discussion follows impact on policy innovati Unlike in Phase 1, because no states drop out of the analysis after policy adoption, the descriptive statistics remain static under each of the 12 analysis scenarios. Appendix C provides for a table of descripti ve statistics for each dependent and independent variable. See T ables 2 6 3 7 for analysis results showing estimates , standard errors , and log pseudo lik e lihood measures for each of the 12 analysis scenarios. Table 38 shows the count of times each independent variable showed significance for repeating policy adoptions , giving a visual representation of the relative statistical significance of each independent variable on policy adoption . Given that the partis anship variables did not consistently demonstrate a pattern of significance, the analysis is best focused on the scenarios where measures of direct democracy and measures of diffusion are common. This allows for better identification of any hidden pattern s.

PAGE 97

92

PAGE 98

93

PAGE 99

94

PAGE 100

95

PAGE 101

96

PAGE 102

97

PAGE 103

98 Scenarios 1, 5, and 9 These scenarios use the presence of direct democracy and the contiguous state diffusion variables. Scenario 1 includes all partisanship variables; scenario 5 includes only the partisanship measure of single party control of both office; and scenario 9 includes the single party control of the legislature variable and the single

PAGE 104

99 In each of these scenarios, nearly the same independent variables show a statistically significant relationship with the repeating policy adoption of both TEL and medical marijuana policies. The presence of direct democracy is not among those independent variables showing statistical significa nce. For TEL adoption 3, negatively associated with the adoption of TELs. Given that TELs are inherently anti tax and/or anti spending measures, this finding is counterintuitive. This independent variable did not measure any statistical significance in any Phase 1 scenario. Also showing statistical significance across all three scenarios for TEL policy median household income as measured by state revenue by population professionalis m (positive). were significant at a <0.05 level, while the other independent variables were significant at a best predictors for the adoption of repeating TEL policies. In scenario 1, the measure of consecutive years of single significant, but only at the <0.1 level. For medical marijuana, only one independent variable was significantly related to the repeating policy adoptions negative relationship at the <0.01 level for all three scenarios. This finding is consistent with all 12 Phase 1 scenarios, where this independent variable also showed a negative and statistically significant relationship with single event adoptions of medical marijuana policy.

PAGE 105

100 In scenarios 1 and 5, years of single party control over both the legislature and governo as predicted . This variable was excluded from scenario 9. Together, these findings indicate that some Republican influence is needed, as Democratic dominance of the whole state gover nment is negatively related to policy adoption. But then again, not too much Republican influence. Scenarios 3, 7 and 11 These scenarios again use the contiguous state measure of diffusion, but now with the difficulty index measure of direct democracy. Like the previous scenarios, this measure of direct democracy is not significantly related to either category of policy adoptions. With only a slight variance to the level of significance for the economic conditions variable, the same internal characteris tics variables are shown to be statistically significant predictors of TEL policies at the same levels of significance as seen in scenarios 1, 5, and 9. Regarding medical marijuana, the results are unchanged from scenarios 1, 5 and 9. T he measure of di rect democracy used has little influence (TELs) or no influence (medical marijuana) on policy adoption. Scenarios 2, 6, and 10 These scenarios use the presence of direct democracy variable with the fixed region measure of diffusion . In these scenarios, t he presence of direct democracy is positi vely and significantly related, as hypothesized , to the repeating adoption of TEL policies at the <0.05 level . The variable shows no significance with the adoption of medical marijuana policies. For TELs, the fixe d region measure of diffusion is also significantly but negatively related to policy adoptions, at the <0.05 level under scenario 2, but only at the <0.1 level under scenarios 6 and 10. The negative relationship is unexpected and is inconsistent with

PAGE 106

101 Phas e 1, where 5 of the 6 fixed region scenarios showed a positive relationship with single policy adoptions. This indicates that first TEL policy adoptions are influenced by regional states also having a TEL policy in place, but that repeated policy adoption s are more common when a TEL state is in relative geographic isolation from other TEL states. This could also indicate policy learning, meaning that states are not only influenced by As policy learning increases, the need for policy adjustments or additional policy adoptions is minimized. Like the scenarios already discussed using a contiguous state measure of diffusion , TEL policy adoptions are also predicted by rural populations (negative), median household incomes (negative), population (negative), and legislative professionalism (positive). A ificant level. Unlike the scenarios already discussed, where the predicting factors were only a region measure of diffusion are influenced by the presence direct democracy, diffusion , and internal state characteristics . again negatively related to policy adoption, but at a <0.001 level under these scenarios. Under scenarios 2 and 6, at a <0.05 to repeating policy adoptions. There are no other statistically significant predictors of medical marijuana repeating events, including measures of diffusion and direct democracy. Scenarios 4, 8, and 12 region measure of diffusion. Like scenarios 2, 6, and 10, the direct democracy index is

PAGE 107

102 significantly related to TEL policy adoptions but with a negati ve relations hip as hypothesized, at a <0.05 level under scenario 4, but only at a <0.1 level under scenarios 8 and 12. This measure of direct democracy had no statistically significant relationship with medical marijuana policy adoptions. For TEL policy adoptions, the fixed region measure of diffusion was again significantly but negatively related under all three scenarios, again at a <0.05 level under scenario 4, and only at a <0.1 level under scenarios 8 and 12. Like before, rural population (negative), median h ousehold income (negative), population (negative), and legislative professionalism (positive) were also significantly related to TEL policy adoption. Under scenario 4, single related to policy adoptions, and under scenario 12 the single party control of the legislature was negatively and significantly related. Republican score is negatively and signific antly related at a <0.001 level. Only under independent variables act as a significant predictor of TEL policy adoptions under the Phase 2 repeating events model. Summary of Phase 2 Results The hypotheses made regarding direct democracy were the same for both Phase 1 and Phase 2 of the analysis. Again, this research hypothesized that the presence of direct democracy would be positively and significantly related to policy innovation, and that the direct democracy index would be negatively and significantly related to policy innovation. In Phase 2, the hypotheses held true for TEL policy adoptions, but only when paired with the

PAGE 108

103 fixed region measure of policy diffusion. T ables 3 9 and 40 show the p values resulting from the measures of direct democracy in the Phase 2 analysis for the 6 scenarios where the findings were significant. As can be seen in the tables, for medical marijuana policy adoptions, the hypothesized impac t of either measure of direct democracy was not supported in any scenario. For Phase 2 TEL policies, findings for the measures of direct democracy are consistent with Phase 1 , though results did not duplicate exactly from Phase 1 to Phase 2 . In Phase 2, under all six scenarios with fixed region diffusion, the measure of direct democracy was a predictor of repeating policy adoptions. In Phase 1, the measures of direct democracy were an accurate predictor of single event TEL policies when paired with the fixed region measure of diffusion, but only in 4 of the 6 scenarios. In both phases, the measure of direct democracy was not a significant predictor of TEL policy adoption(s) in any scenario using the contiguous state measure of diffusion.

PAGE 109

104 In Phase 2, the measure of direct democracy did not play a significant role in the adoption of medical marijuana policies under any scenario. This contrasts sharply with the findings from Phase 1, where the measure of direct democracy played a significant role in the hypothesized direction in 10 of the 12 analysis scenarios. For repeating adoptions of TEL policies, across all 12 scenarios, the main driving force seems to be internal characteristics of a state. As can be seen in tables 10 through 21, the internal c haracteristics to a lesser extent were also prominent in Phase 1 in regards to the single adoption of TEL policies, but mostly when modeled with the fixed region model. In Phase 2, internal characteristics play a significant role across all 12 scenarios. Also in Phase 1, the fixed region measure of diffusion was significant in 5 of

PAGE 110

105 the 6 scenarios with the fixed region measure. In Phase 2, the fixed region measure was significant in all 6 scenarios. For repeating adoptions Republican score played a prominent role, with a negative relationship, in predicting policy adoptions. For medical marijuana, neither direct democracy nor diffusion, based on these results, are accurate predictors of repeating policy adoption s . I n Phase 1, Republican score was also significant and negative in all 12 scenarios. However, in Phase 1, it was just one of several variables that played a significant role in single policy adoptions. In Phase 2, it was the only significant independent variable in 5 of 12 scenarios. In the other education (positive) or single party control of both the (negative).

PAGE 111

106 CHAPTER VI DISCUSSION & CONCLUSION The dissertation hypothesized that direct democracy establishes an alternate form of governmental competition on the established legislative and executive branches in the states. This increased competition would lead to higher rates of policy innovation i n states where direct citizen access to the ballot is granted. This dissertation provides several co ntributions to the literature by enhancing the event history analysis model relative to the use of direct democracy as an explanatory variable to policy a doption. T his research focuses on direct democracy as a predictor of policy adoption in both a direct and indirect fashion, uses a more robust measure of a direct democracy index rather than treating the citizen initiative as a simple dichotomous variable , focuses on policy adoptions only during the prime use of the citizen initiative after 1978, researches multiple policy adoptions in the same study, and explores policy adoptions in two policy subgroups. In exploring the results in further detail , this discussion concentrates on the fixe d region diffusion measures couple d with the use of all partisanship variables to mirror the more recent literature preferring the fixed region measure of diffusion over the contiguous state measure (Mooney & Lee, 1995; A ndrews, 2000; Allen, Pettus, and Haider Markel, 2004) . These combinations represent scenarios 2 and 4 . In the early literature, the most common measure of diffusion (external influences) was to quantify the number or percentage of contiguous states adop ting a policy (Berry & Berry, 2007). More recently, fixed region diffusion has been utilized. It first defines regions of the country and measures the number or percentage of states from a region that have adopted a policy to account for diffusion across non contiguous states (Mooney & Lee, 1995; Andrews, 2000; Allen, Pettus, and Haider -

PAGE 112

107 Markel, 2004). Using the three scenarios of partisanship captures the most robust measures of single party control for both the legislative and executive branches separate ly and together. Phase 1 In scenarios 2 and 4, both the presence of direct democracy and the direct democracy difficulty index played a significant role in the single adoption of three policy variables in the direction hypothesized TELs, family cap expe nditures, and medical marijuana. From these findings, it would be difficult to conclude that direct democracy, through the lens of a political competition theory, promotes policy innovation. So the question becomes, is there something unique about the th ree policies whose adoptions do conform to the tested hypotheses relative to the other five policies? TELs, while complex in detail, can be easily understood on a basic level (at least on the surface and at least perceived to be) as restrictions on gover nment revenue and/or expenditures. Either way, TELs are simple to grasp policies that restrict the growth of government and reduce tax burdens on the population. Medical marijuana, sans the complex details of implementation and regulation, is also a stra ight forward policy topic that is easily digestible. At its core, medical marijuana policy is really about a simple yes or no question should marijuana possession and consumption be allowed for medical purposes? If political competition in the form o f direct democracy should be expected to promote policy innovation , as hypothesized, then it seems reasonable that to be true policymaking process. That is, the general pub lic as a whole needs to b e engaged and opinionated on the policy topic. I n this light, the topic of taxes and marijuana are more

PAGE 113

108 common topics of conversation among the general public than say enterprise zones or economic development. I f one accepts that TELs and medical marijuana are, on a basic level, more popular topics of conversation than other policy variables, these results may show that it is not merely the existence of direct democracy in isolation that spurs additional politica l competition and therefore more policy innovation. Rather, the existence of direct democracy coupled with a high level of salience and strong public opinion is the actual catalyst for increased competition and therefore increased policy innovation. Ther e is at least some empirical support for this reasoning. Lax and Phillips (2009) found that gay rights policy adoptions are responsive to policy specific public opinion, even when controlling for voter ideology, ideology of elected officials, interest gro up pressures. salient issue is one that affects a large number of people . He it can adjust over time. Salience determines what pla yers will participate in the policymaking process on a given topic. When there are higher levels of salience, more policymakers and citizen groups are motivated to participate in the policy debate or policymaking process, particularly when the same issue is low in complexity. Consistent with the idea of salience , Matsusaka (2014) found that initiatives states are more likely to adopt policies favored by the majority of the public than non initiatives states. He also found that the gap in policy adoptions between the two types of states increases as public opinion becomes more one sided (a stronger majority opinion) . This would suggest that both salience and level of public opinion matter. From a theoretical standpoint, this would imply that increased pol itical competition through direct democracy, under only certain conditions, can promote policy innovation for

PAGE 114

109 specific types of policy. From a practical viewpoint, then, if one is attempting to promote a new policy in a state, the presence of direct democ racy in that state alone is not sufficient to increase competition . Rather, the presence of direct democracy and a high level of salience would be needed to truly represent an increase i n policymaking competition. In this light, the hypotheses stated in this research may not be false, but rather the hypotheses simply lacked a necessary condition. Further research could explore this premise in more detail. With that said, that direct dem ocracy is also a predictor of the adoption of family cap expenditure policies seems contradictory to the above statements. It is unclear as to how or why policies regarding family cap expenditures were predicted by the two measures of direct democracy giv en the other findings in this research. Said another way, there are no clear factors that set apart family cap expenditure policies from other policies where the hypotheses were not supported. The most unexpected finding from these analyses is the relati onship of both measures of direct democracy with the bicycle helmet and mandatory seatbelt law policies. While significant under both measures of direct democracy, the findings indicate that states without direct democracy are more likely to adopt these policies at the <0.001 level. These findings indicate the any competition created though the direct democracy process runs counter to policy adoption for these safety motivated measures. In the absence of direct democracy, it seems that l egislature s take a more nurturing hand toward their citizens. On the surface, one might reason from these findings, coupled with the findings for TELs, medical marijuana, and family cap exemptions, that dir ect democracy in the states lead s to restrictions of governmental i nstitutions while maximizing personal liberties. TELs , at least as intended , restrict government, while medical marijuana and family cap exemptions

PAGE 115

110 reduce government al regulations Conversely , mandatory bicycle hel mets and mandatory seatbelt laws could be seen as a government mandate restricting personal freedoms and personal choice. It would be a stretch to make this interpretation based only on these findings, but the notion is supported in previous literature. Bowler and Donovan (2006) found that direct democracy in a state leads to a more restrictive legal framework and institutional constraints for political parties, such as th e implementation of term limits, civil service regulation, limitations of campaign contributions, and the ability to recall elected officials , thus regulating and weakening their autonomy. Initiative states are also more likely to adopt term limit policie s for legislators (Tolbert , 1998; Bowler & Donovan, 1995) and have campaign finance restrictions (Pippen, Bowler & Donovan, 2002). Further, New (2001) and Krol (2007) both found that TELs passed by initiative were more effective at limiting government gro While this research focuses on policy adoption at the state level, recent local election results are inconsistent with this notion , at least anecdotally . As an example, voters in Denver in 2017 passed the c (Murray, 2017a) and a $937 million bond package funding transportation, cultural facilities, and libraries, among other areas (Murray, 2017b) . While these local election results are inconsistent with the above stated premise, it could also be argued that these local results are supportive of the notion that direct democracy lea ds to personal liberties for local governments relative to mandates or direction from the state. D oes direct democracy increase personal liberties and/or decrease at least perceived at the state level ? The results suggest this,

PAGE 116

111 and it is a logical extension for further r esearch. If this premise can be supported through further rese arch, it would imply that political competition, as hypothesized in this research, is not the means through which direct democracy promotes policy adoption. Rather, it would be the underlying premise of the policy either restricting or enhancing perceiv ed personal liberties that gains more traction via the presence of direct democracy. From a practical standpoint, this would mean that policy entrepreneurs would be wise to avoid the advocating of policy that enhances governmental power or reduce s perso nal choice and freedom. Phase 2 statistically significant predictors of TEL policy adoptions in the direction hypothesized in scenarios 2 and 4 (and in all scenarios using the fi xed region measure of diffusion). However, neither measure of direct democracy influenced repeating adoptions of medical marijuana policies. Like above, one must ask if there is an inherent difference between these two policies in the context of repeatin g events that would result in contradictory conclusions, such as substantive policy adoption versus regulatory policy adoption. In looking at repeated TEL policy adoptions, each adoption is often a policy of substance; that is to say that repeating adopti ons are not simple regulatory measure s meant to provide a more sustainable framework to the original policy adoption. Further, b ecause the definition of a TEL is relatively ambiguous, a TEL policy can mean restrictions on government revenue, government expenditures, and various forms of taxation limitations. serve as a prime example. It makes sense, then, that the influence of direct democracy can and is felt even beyond the first policy adoption in a state.

PAGE 117

112 In looking at repeated medical marijuana policy adoptions, in further review the results are perhaps not surprising . Though this dissertation hypothesized that there would be a significant relationship, and while this relationship was found in the single adoption model, a closer look into the context of repeating events for medical marijuana may explain the lack of signi ficance. In this policy topic, t he degree that repeating events represent measures aimed at regulation or implementation, versus substantive policy change, may help to explain the empirical results In this context, a substantive policy change would be co nsidered a policy that stands on its own; it is not simply a tinkering or adjustment of existing law. Conversely, a regulatory policy adoption modifies existing law and would otherwise be unnecessary if not for an already existing policy adoption. As not ed above, California, while initially adopting the policy through a ballot measure, has passed several updates to the medical marijuana law through the legislature. These updates pertain to collective marijuana cultivation and a new regulatory framework o f the existing law. Colorado also initially passed medical marijuana through citizen initiative, but the legislature later adopted dispensary licensing procedures. Maine, with the most repeating events, first passed the policy through the ballot, with an other ballot measure later establishing registries and dispensaries. In addition , Maine also has three legislative policy by As a final example, y initiative. However, s ubsequent legislative changes in 2010, 2011, and 2015 allowed patients and caregivers to collectively grow marijuana, allowed professionals to legally recommend marijuana for treatment, and by S

PAGE 118

113 Because the repeating events in this policy area tend to focus on the regulatory framework for the implementation of medical marijuana, it follows that the presence of direct democracy would lack significant influenc e over the adoptions of repeating events. It would logically follow that while direct democracy is a good tool for getting a policy in the books, as supported by the findings for medical marijuana from Phase 1, it is not useful as a tool for regulatory up dates. T he repeating events model understates the explanatory power of direct democracy for subsequent policy adoptions if those adoptions lack innovative substance. If this logic holds, it implies that political competition theory only applies to policy adoptions of a substantive nature, but does not apply to regulatory measures or practical viewpoint, this would not necessarily harm the ambitious policy entrepreneur who is presumably more interested in the substantive policy adoption than the regulatory frameworks subsequently put in place. Limitations and Future Research There are several limitat ions to this research. First regarding data collection, some proxies were n ecessarily used based on limitations of data availability. For example, the education level in the states and the rural population in the states were not found on an annual basis, but rather were only available by decade. In the data analyses, the most r ecent data available were used in each year. Regarding the measures of policy adoption for the eight dependent variables, these analyses do not consider or allow for nuances in the policy. For example, it does not account for whether a TEL policy is const itutional or statutory, or for how restrictive the TEL policy is. This research does not consider the age requirement of the bicycle helmet laws, does not

PAGE 119

114 consider the penalty for breaking a primary seat belt law, and does not consider the level of regula tion put in place with the legalization of medical marijuana. As noted early in this As noted above, the findings from this research coul d benefit from two primary 1) Does the presence of direct democracy need to be coupled with another factor, such as high levels of salience in the policy variable, in order to better fit the hypothesis that increas ed competition leads to greater policy innovation? 2) Is direct democracy, rather than an avenue to competition and increased policy innovation, more of a tool to promote fewer governmental restrictions and enhance personal liberties? Regarding the diffus ion component in both phases of the analysis, this research used two different measures: 1) contiguous state diffusion effects (measured as the percentage of bordering states already having adopted the policy in question), and 2) fixed region diffusion eff ects (measures as the percentage of states within the same geographic region already having adopted the policy in question). These measures of diffusion do not capture other s of exter nal advocacy groups and the out of state financing that often comes from these advocacy groups are not captured in any measure of diffusion focusing on geographic jurisdictional boundaries. Future models could account for these non jurisdictional external influences, even if this alternative diffusion measure is largely absent in the existing diffusion literature. The effects of political cultures and subcultures on political processes and policy adoption have been well explored (Lieske, 2010). While the geographic measures of diffusion used here capture the geographic elements of regional subcultures, new

PAGE 120

115 multidimensional measures are being used to refine the subculture measure (Lieske, 2010). F urther, some literature has expanded the measure of diffusion, notably the idea of politically similar states (Grossback, Nicholson Crotty, & Peterson, 2004). Rather than focus on the geographic location of a state, diffusion could be measured by how politically simi lar (or dissimilar) the at risk state is relative to states that have already adopted the policy in question. Grossback, Nicholson Crotty, and Peterson (2004) measure diffusion effects based on ideological similarity using a measure of liberalism for each state. While their findings did not contradict results from studies that use the geographic measures of diffusion, they did find that the measure of political similarity is the most robust measure of diffusion in three policy areas. Regarding the eight p olicy variables selected for analysis, this research focused on two policy subgroups fiscal and social. The findings do not indicate any clear pattern of policy adoption related to the policy subgroups. Future research could expand on the use of policy subgroups. For example, fiscal policy could be further broken down into revenue and expenditure policy types, or policies that expand or constrict governmental authority. Additional policy subgroups could include virtually any area, such as regulatory p olicies, business related policies, and public safety related policies . Regarding the theoretical foundation of this research that political competition increases policy innovation it is worth noting that political competition derived from direct dem ocracy, under specific conditions, could lead to a decrease in policy innovation (see Chapter 2). Future research could attempt to account for this phenomenon. Finally , i n defining a policy innovation as any policy new to the adopting state, it does not c first adoption in the country regardless of state . In other

PAGE 121

116 words, policy invention rather than policy innovation. This avenue of research could support the idea that direct democracy promotes the states as policy laboratories. Final Thoughts This research shows, as hypothesized, that direct democracy measured as both the simple existence of the initiative and based on an index measuring the difficulty of ballot placement can play a statistically signific ant role in policy innovation. Under the single event policy adoption model, the existence of direct democracy was a statisti cally significant predictor of policy adoption for two policies (family cap expenditures and medical marijuana) under all analysis scenarios, and a statistically significant predictor of policy adoption for TELs when paired with a fixed region measure of d iffusion. The existence of direct democracy was also shown as a statistically significant deterrent to primary seat belt and minor bicycle helmet laws (opposite the direction hypothesized). The direct democracy index was a statistically significant predi ctor of policy adoption of family cap expenditure laws under all analysis scenarios, medical marijuana policies under four of six scenarios, and of TEL policies under a single scenario. Again, the index was a deterrent to the adoption of primary seat belt and minor bicycle helmet laws (opposite the direction hypothesized) . Under the repeating events model, both measures of direct democracy were statistically significant predictors of TEL policy in the direction hypothesized when paired with the fixed re gion meas ure of diffusion, but showed no explanatory power over the repeating adoptions of medical marijuana policy. While the hypotheses were not supported for all policies analyzed, this research explored potential explanations for the lack of support an d suggest ed avenues for further

PAGE 122

117 impact on policy innovation. Further, this research reinforces a foundation set for the exploration of policy innovation from both a single event and repeating event perspective, citing specific areas of potential improvement for distinguishing between the types of repeating policy adoptions in future research. In practice, this research sheds some light on the impacts direct democracy has on policy innovation through the lens of increased political competition. Policy practitioners, based on this research, would be wise to consider the existence of direct democracy in the states as one factor when exploring the feasibility of policy i nnovation at the state level.

PAGE 123

118 LIST OF REFERENCES A bipartisan bill that got away: We hope legislators will try again next year to pass a bill making it harder to amend the state constitution [Editorial]. (2012, May 5). The Denver Post , p. 15A. A Hal f Century of Learning: Historical Census Statistics on Educational Attainment in the United States, 1940 to 2000: Detailed Tables (2016). U.S. Census Bureau. Retrieved from U.S. Census Bureau All Up Chart of Governors by Election Cycle (2012). The Green Papers. Retrieved from http://www.thegreenpapers.com/G12/GovernorsByElectionCycle.phtml Allen, M.D., Pettus, C., & Haider Markel, D. (2004). Making the national local: Specifying the conditions for national government influence on state policyma king. State Politics and Policy Quarterly, 4 (3), 318 344. American Community Survey, 2010 1 Year Estimates, Table S. 1501 (2016). U.S. Census Bureau. Retrieved from https://www.census.gov/newsroom/releases/xls/cb12 33table1states.xls Amiel, L., Deller , S., & Stallmann, J. (2009). The construction of a tax and expenditure limitation index for the U.S. University of Wisconsin Madison Department of Agriculture & Applied Economics Staff Paper No. 536 . Retrieved from http://www.aae.wisc.edu/pubs/sps/ Andre ws, C.J. (2000). Diffusion pathways for electricity deregulation. Publius, 30 (3), 17 34. Bad bills driven by bids for high office: Legislation from two Democratic state lawmakers shows blatant vote seeking, not good public policy [Editorial]. (2012, Marc h 9). The Denver Post , p. 10B. Balla, S.J. (2001). Interstate professional associations and the diffusion of policy innovations. American Politics Research, 29 (3), 221 245. Banducci, S. (1998). Direct legislation: When is it used and when does it pass? In S. Bowler, T. Donovan, & C. Tolbert (Eds.), Citizens as Legislators: Direct Democracy in the United States. Columbus, OH: Ohio State University Press. Berry, F.S. (1994). I nnovation in public management: The adoption of strategic planning. Public Administration Review, 54 (4), 322 330. Berry, F.S., & Berry, W.D. (1990). State lottery adoptions as policy innovations: An event history analysis. The American Political Science R eview, 84 (2), 395 415.

PAGE 124

119 Berry, F.S., & Berry, W.D. (2007) Innovation and diffusion models in policy research. In P.A. Sabatier (Ed.), Theories of the Policy Process 2 nd ed. (pp. 223 260). Boulder, CO: Westview Press. Bicycle Helmet Safety Institute. (201 6). Helmet laws for bicycle riders. Retrieved from: http://www.helmets.org/mandator.htm Blume, L., Muller, J., & Voigt, S. (2009). The economic effects of direct democracy a first global assessment. Public Choice, 140 (3), 431 461. Boehmke, F.J. (2002). T he effect of direct democracy on the size and diversity of state interest group populations. The Journal of Politics, 64 (3), 827 844. Boehmke, F.J., & Skinner, P. (2012 a ). S tate policy innovativeness revisited. State Politics & Policy Quarterly, 12 (3), 303 329. Boehmke, F.J., & Skinner, P. (2012b). State policy innovativeness revisited. Retrieved from: http://myweb.uiowa.edu/fboehmke/Papers/boehmke skinner2012preprint.pdf Boehmke, F.J., & Witmer, R. (2004). Disentangling diffusio n: The effects of social learning and economic competition on state policy innovation and expansion. Political Research Quarterly, 57 (1), 39 51. Bowler, S., & Donovan, T. (1995). Popular responsiveness to taxation. Political Research Quarterly, 48 (1), 79 100. Bowler, S., & Donovan, T. (2004). Measuring the effect of direct democracy on state policy: Not all initiatives are created equal. State politics & Policy Quarterly, 4 (3), 345 363. Bowler, S., & Donovan, T. (2006). Direct democracy and political pa rties in America. Party Politics, 12 (5), 649 669. Buckley, J., & Westerland, C. (2004). Duration dependence, functional form, and corrected standard errors: Improving EHA models of state policy diffusion. State Politics & Policy Quarterly, 4 (1), 94 113. Bueno de Mesquita, B., Morrow, J.D., Siverson, R.M., & Smith, A. (2001). Political competition and economic growth. Journal of Democracy, 12 (1), 58 72. Census Regions and Divisions of the United States (2016). U.S. Census Bureau. Retrieved from https://www2.census.gov/geo/pdfs/maps data/maps/reference/us_regdiv.pdf Cloglog Complementary log log Regression (2016). Retrieved from http://www.stata.com/manuals13/rcloglog.pdf

PAGE 125

120 Crain, W.M., & Tolli son, R.D. (1976). Campaign expenditures and political competition. Journal of Law and Economics, 19 (1), 177 188. Downs, A. (1957). An economic theory of democracy . New York: Harpe r & Row. Gormley, W.T. Jr. (1986). Regulatory issue networks in a federal system. Polity, 18 (4), 595 620. Grattet, R., Jenness V., & Curry, T.R. (1998). The homogenization and differentiation of hate crime law in the United State, 1978 to 1995: Innovati on and diffusion in the criminalization of bigotry. American Sociological Review, 63 (2), 286 307. Gerber, E.R. (1996). Legislative response to the threat of popular initiatives. American Journal of Political Science, 40 (1), 99 128. Gerber, E.R., & Philli ps, J.H. (2005). Evaluating the effects of direct democracy on public American Politics Research, 33 (2), 310 330. Grattet, R., Jenness, V., & Curry, T.R. (1998). The homogenization and differentiation of hate crime law in the United States, 1978 to 1995: Innovation and diffusion in the criminalization of bigotry. American Sociological Review, 63 (2), 286 307. Grossback, L.J., Nicholson Crotty, S., & Peterson, D.A.M. (2004). Ideology and learnings in policy diff usion. American Politics Research, 32 (5), 521 545. CalWORKS maximum family grant rule. University of California Berkeley Law Center on Reproductive Rights and Justice Issue Brief, April, 1 6. Haider Markel, D.P. (2001). Policy diffusion as a geographical expansion of the scope of political conflict: Same sex marriage bans in the 1990s. Stat e Politics & Policy Quarterly, 1 (1), 5 26. Hero, R., & Tolbert, C. (1996). A racial/ethnic diversity interpretation of politics and policy in the states of the U.S. American Journal of Political Science, 40 (3), 851 871. Historical Income Tables: Households (2016). U.S. Census Bureau. Retrieved from http://www.census.gov/data/tables/time series/demo/income poverty/historical income households.html Historical Income Tables: States (2016). U.S. Census Bureau. Retrieved from https://www.census.go v/data/tables/time series/dec/historical income states.html Hofstadter, R. (1989). The American political tradition and the men who made it. New York: Vintage Books.

PAGE 126

121 Initiative & Referendum Institute. (2011). University of Southern California. Retrieved from http://www.iandrinstitute.org/ Initiative & Referendum Institute. (201 7 ). University of Southern California. Retrieved from http://www.iandrinstitute.org/data.cfm Initiative Use. (2010). Initiative & Referendum Institute . Retrieved from http://www. iandrinstitute.org/IRI%20Initiative%20Use%20%282010 1%29.pdf Interpreting Log Likelihood (2017, 3 24). Cross Validated. Retrieved from https://stats.stackexchange.com/questions/214556/interpreting log likelihood Jones, B.S., & Branton, R.P. (2005). Beyond logit and probit: Cox duration models of single, repeating, and competing events for state policy adoption. State Politics & Policy Quarterly, 5 (4), 420 443. Karch, A. (2006). National intervention and the diffusion of polic y innovations. American Politics Research, 34 (4), 403 426. Karch, A. (2007). Democratic Laboratories: Policy diffusion among the American States . Ann Arbor: The University of Michigan Press. Kelley, S. Jr. (1965). Foreward. In Downs, A., An economic theo ry of democracy. New York: Harper & Row. Kessler, A.S. (2005). Representative versus direct democracy: The role of informational asymmetries. Public Choice, 122 (1/2), 9 38. Key Aspects of State and D.C. Medical Marijuana Laws. (2015, September). Marijua na Policy Project . Retrieved from https://www.mpp.org/issues/medical marijuana/state by state medical marijuana laws/key aspects of state and d c medical marijuana laws/ Kotsogiannis, C., & Schwager, R. (2006). On the incentives to experiment in federations. Journal of Urban Economics, 60 (3), 484 497. Krol, R. (2007). The role of fiscal and political institutions in limiting the size of state government. Cato Journal, 27 (3), 431 445. Lax, J.R., & Phillips, J.H . (2009). Gay rights in the states: Public opinion and policy responsiveness. The American Political Science Review, 103 (3), 367 386. Lieske, J. (2010). The changing regional subcultures of the American states and the utility of a new cultural measure. Po litical Research Quarterly, 63 (3), 538 552. List of Current United States Governors (2016). Wikipedia.org. Retrieved from https://en.wikipedia.org/wiki/List_of_current_United_States_governors

PAGE 127

122 Llavador, H.G. (2000). Abstention and political competition. Review of Economic Design, 5 (4), 411 432. Log Pseudo Likelihood (2004, 10 19). STATA Services. Retrieved from https://www.stata.com/statalist/archive/2004 10/msg00558.html . Lupia, A., & Matsusaka, J.G. (2004). Direct democracy: New approaches to old questions. Annual Review of Political Science, 7 , 463 482. Madison, J. (2012/1787). Federalist No. 10. In J. Madison, A. Hamilton, & J. Jay, The Federalist Papers: The Classic Original Edition . Lexington, KY: SoHo B ooks. Marijuana Policy Project (2016). Medical Marijuana. Retrieved from https://www.mpp.org/issues/medical marijuana/ Martel l , C.R., & Teske, P. (2007). Fiscal management implications of the TABOR bind. Public Administration Review, 67 (4), 673 687. Ma tsusaka, J.G. (1995). Fiscal effects of the voter initiative: Evidence from the last 30 years. Journal of Political Economy, 105 (3), 587 623. Matsusaka, J.G. (2000). Fiscal effects of the voter initiative in the first half of the twentieth century. Journal of Law and Economics, 43 (2), 619 650. Matsusaka, J.G. (2004). For the many or the few: The initiative, public policy, and American Democracy . Chicago: The University of Chicago Press. Matsusaka, J.G. (2005). The eclipse of legislature: Direct democracy in the 21 st century. Public Choice, 124 , 157 177. Matsusaka, J.G. (2014). Disentangling the direct and indirect effects of the initiative process. Public Choice, 160 , 345 366. Matsusaka, J.G., & McCarty, N.M. (2001). Political resource allocation: Benefits and costs of voter initiatives. Journal of Law, Economics, & Organization, 17(2), 413 448. Medical Marijuana Overview. (2015). Marijuana Policy Project . Retrieved from https://www.mpp.org/issues/medical marijuana/medical marijuana overview/ Mi ntrom, M. (2001). Policy design for local innovation: The effects of competition in public schooling. State Politics & Policy Quarterly, 1(4), 343 363. Mintrom, M., & Vergari, S. (1996). Advocacy coalitions, policy entrepreneurs, and policy change. Policy Studies Journal, 24 (3), 420 434.

PAGE 128

123 Mooney, C.Z., & Lee, M. (1995). Legislating morality in the American states: The case of pre Roe abortion regulation reform. American Journal of Political Science, 39 (3), 599 627. Mullins, D.R., & Wallin, B.A. (2004). T ax and expenditure limitations: Introduction and overview. Public Budgeting & Finance, 24 (4), 2 15. Murray, J. (2017, November 8 a mayor opponent turns to implementation. The Denver Post , re trieved from http://www.denverpost.com Murray, J. (2017, November 8 b ). Denver voters strongly approve $937 million bond package for road libraries and cultural facilities. The Denver Post , retrieved from http://www.denverpost.com National Highway Traff ic Safety Administration (2006). Primary enforcement saves lives: The case for upgrading secondary safety belt laws. Retrieved from: http://www.ems.gov/pdf/PrimaryEnforcement.pdf New, M.J. (2001, Dec. 13). Limiting government through direct democracy: The case of state tax and expenditure limitations. Cato Institute. Retrieved from: http://www.cato.org/publications/policy analysis/limiting government through direct democracy case s tate tax expenditure limitations New, M.J. (2010). U.S. state tax and expenditure limitations: A comparative political analysis. State Politics & Policy Quarterly, 10 (1), 25 50. Obinger, H. (1998). Federalism , direct democracy, and welfare state developm ent in Switzerland. Journal of Public Policy, 18 (3), 241 263. Persico, N., Pueblita, J.C.R., & Silverman, D. (2011). Factions and political competition. Journal of Political Economy, 119 (2), 242 288. Pippen, J., Bowler, S., & Donovan, T. (2002). Election reform and direct democracy: Campaign finance regulations in the American States. American Politics Research, 30 (6), 559 582. Polsby, N.W. (1984). Political innovation in America: The politics of policy initiation . New Haven, CT: Yale University Press. Population and Housing Unit Estimate (2016). U.S. Census Bureau. Retrieved from https://www.census.gov/programs surveys/popest.html Rainey, D.L. (2005). Product innovation: Leading change through integrated product development . New York: Cambridge University Press.

PAGE 129

124 Resnick, P. (2004). The effect of tax and expenditure limitation measures on tax reform efforts in the states (Doctoral dissertation). Retrieved from ProQuest LLC. (UMI No. 3312864) Rogers, D.L., & Roger s, J.H. (2000). Political competition and state government size: Do tighter election produce looser budgets? Public Choice, 105 (1/2), 1 21. Seljan, E.C., & Weller, N. (2011). Diffusion in direct democracy: The effect of political information on proposal f or tax and expenditure limits in the U.S. states. State Politics & Policy Quarterly, 11 (3), 348 368. Smith, D.A., & Fridkin, D. (2008). Delegating direct democracy: Interparty legislative competition and the adoption of the initiative in the American Stat es. American Political Science Review, 102 (3), 333 350. Smith, D.A., & Tolbert, C. (2007). The instrumental and educative effective of ballot measures: Research on direct democracy in the American states. State Politics & Policy Quarterly, 7 (4), 416 445. Soule , S.A., & Earl, J. (2001). The enactment of state level hate crime law in the United States: Intrastate and interstate factors. Sociological Perspectives, 44 (3), 281 305. Stata Application Tutorial 7: Repeated Events (2005). Retrieved from psfacul ty.ucdavis.edu/bsjjones/icpsrtutorial7.doc State Government Tax Collections (2016). U.S. Census Bureau. Retrieved from https://www.census.gov/govs/statetax/historical_data.html State Partisan Composition (2016). NCSL. Retrieved from http://www.ncsl.o rg/research/about state legislatures/partisan composition.aspx#Timelines State by State Medical Marijuana Laws: How to Remove the Threat of Arrest. (2015). Marijuana Policy Project . Retrieved from: https://www.mpp.org/issues/medical marijuana/state by s tate medical marijuana laws/state by state medical marijuana laws report/ Stcox Cox Proportional Hazards Model (2016). Retrieved from http://www.stata.com/manuals13/ststcox.pdf Stigler, G.J. (1972). Economic competition and political competition. Public Choice, 13 , 91 106. Squire, P. (2007). Measuring state legislative professionalism: The Squire Index revisited. State Politics & Policy Quarterly, 7 (2), 211 227.

PAGE 130

125 Tavits, M. (20 07). Principle vs. pragmatism: Policy shifts and political competition. American Journal of Political Science, 51 (1), 151 165. Tolbert, C. (1998). Changing rules for state legislatures: Direct democracy and governance policies. In S. Bowler, T. Donovan, & C. Tolbert (Eds.), Citizens as Legislators: Direct Democracy in the United States. Columbus, OH: Ohio State University Press. Tolbert, C.J., Mossberger, K., & McNeal, R. (2008). Institutions, policy innovation, and e government in the American states. Pu blic Administration Review, 68 (3), 549 563. Urban Percentage of the Population for States, Historial (2010). Iowa State University. Retrieved from http://www.icip.iastate.edu/tables/population/urban pct states Valcore, L. (2012). Sexual orientation and state hate crime laws . Unpublished dissertation research proposal, School of Public Affairs, University of Colorado Denver, Denver, Colorado. Wagschal, U. (1997). Direct democracy and public policymaking. Journal of Public Policy, 17 (3), 223 245. Waisan en, B. (2010). State tax and expenditure limits 2010. Retrieved from: http://www.ncsl.org/research/fiscal policy/state tax and expenditure limits 2010.aspx#typesoflimts Walker, J.L. (1969). The diffusion of innovations among the American states. The Amer ican Political Science Review, 63 (3), 880 899. Wallin, B.A. (2004). The tax revolt in Massachusetts: Revolution and reason. Public Budgeting & Finance, 24 (4), 34 50. Welch, S., & Thompson, K. (1980). The impact of federal incentives on state policy innovation. American Journal of Political Science, 24 (4), 715 729. Wilder, M.S. & Rubin, B.M. (1996). Rhetoric versus reality: A review of studied on state enterprise zone programs. Jo urnal of the American Planning Association, 62 (4), 472 491. Williams, J. (2003). Criminal justice policy innovations in the states. Criminal Justice Policy Review, 14 (3), 401 122.

PAGE 131

126 APPENDIX A

PAGE 132

127 APPENDIX B

PAGE 133

128

PAGE 134

129

PAGE 135

130

PAGE 136

131

PAGE 137

132

PAGE 138

133

PAGE 139

134

PAGE 140

135 APPENDIX C