WHAT HAPPENS IN YOUR STATE DOESN™T STAY IN YOUR STATE: OMISSIONS AND OPPORT UNITIES IN POLICY DI FFUSION By Marty P . Jordan A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Political Science ŠDoctor of Philosophy 2019 ABSTRACT WHAT HAPPENS IN YOUR STATE DOESN™T STAY IN YOUR STATE: OMISSIONS AND OPPORT UNITIES IN POLICY DI FFUSION By Marty P . Jordan Decades of research have offered strong evidence for policy diffusion, whereby one government™s adoption of a new policy influences subsequent governments™ enactment of the same innovation. But most of this rich research has narrowly focused on the spread of statutes in the legislative arena, neglecting the myriad other venues where policy change occurs. And even when scholars have taken note of policies adopted via multiple forums, they have typically employ ed binary models to estimate enact ment without accounting for inter -venue dynamics that might affect policy diffusion. In addition , nearly all diffusion studies fall prey to selection bias, explaining the transfer of innovations that have knowingly diffused, omitting from the models those policies that failed to spread. What is more, most of this research has focused on the transmission of the policy itself, overlooking the potential diffusion of alternative aspects of the policymaking process. This dissertation addresses these omissions a nd capitalizes on existing opportunities in the policy diffusion literature. First, to better understand the spread of policies beyond the legislative context, I mapped the diffusion of a large sample of ballot measures across U.S. states from 1902 Œ 2016, and both anti - and pro -gay marriage policies via multiple venues from 1993 - 2015. I offer evidence of policy diffusion via state legislatures, legislative referenda, citizen initiatives, state courts, and federal courts. While the results reinforc e much of our current understanding of policy diffusion, they also help refine the precise nature of this dynamic process across varying institutional arrangements. Second, I used an established but underutilized modeling strategy Šmultinomial logistic regres sion Što better account for the transfer of innovative ideas via multiple competing arenas. This approach allow s me to simultaneously recognize each factor™s contribution to policy adoption in the respective venues and uncover inter -venue dynamics. Third, to address the persistent selection bias in diffusion studies, I rel ied on the same large sample of ballot measures pursued across U.S. states from 1902 - 2016. I find that nearly half of the ballot measures did not diffuse to other states, and almost thre e-quarters of the measures were enacted by less than a handful of states. Moreover, when I reran the models omitting policies that did not diffuse or only narrowly spread, policy learning™s effect on adoption was twice as large when compared to the full se t. This suggests that policy scholars may be overstating the rate of policy diffusion and inflating fundamental mechanisms™ effect on the process. Finally, fusing the policy -diffusion and venue -shopping literatures, I investigated whether policy actors™ ch oice of venue to press for anti - or pro -gay marriage policies in one state influenced subsequent states™ actors to pick the same forum, a process I term venue diffusion . I posit that policy advocates look to and learn from others, purposively seeking a solution to their shared problem (i.e., policy learning) and how best to achieve that solution (i.e., political learning). By incorporating political learning into my models, I am better able to explain the dynamics of policy diffusion and offer evidence of venue diffusion, at least in the context of a salient morality policy. States are more likely to pick the venue that other, especially similarly -situated, states have cho sen to enact the policy successfully . The interdependence between the American laboratories of democracy appears to go beyond merely the copying of a policy idea to emulating a fundamental input of the policymaking process. iv To my soulmate, Elly, and our marvelous daughters, Gracie and Dot. You fill me with sated joy. v ACKNOWLEDGMENTS This research highlights the importance of interdependence among policy actors in explaining the policy process. Emphasizing this interdependence is fitting because social connectedness to and learning from others have been no less critical to this project and my success in graduate school. I could not have completed this mammoth enterprise without the support of so many people in my life. I relied on the insight and encouragement from numerous faculty mentors, colleagues, friends, and family. If it fitakes a village to raise a child, fl it took a megalopolis to help me complete my dissertation. Of course, any errors in the research presented here are entirely my doing. And u ndoubtedly, because of the scale of this project and my feeble memory, I have likely o mitted the names of some individuals that have positively contributed to this project in some capacity. I extend my apologies for neglecting to recognize their input here and express my heartfelt thanks. My dissertation committee ŠRyan Black, Eric Gonza lez Juenke, Ch arles Shipan (from the University of Michigan), and Sa undra Schneider Šwas invaluable to this research and my progress throughout my graduate studies. Some of the best advice I received in graduate school was to conscript the smartest, most pr olific scholars to form my committee, even if my topic did not perfectly align with their research foci. I am glad I heeded such guidance . Undoubtedly, these four individuals ™ wells of knowledge and unique perspectives greatly enhanced the development and completion of this project , as well as my career. Ryan Black is one of the most prolific, rational, and resourceful researchers I know. He sees empirical questions in all things. In the classroom, Ryan sets expectations high, and students always seem to rise to the occasion. He is accessible, humble, and eschews titles. And he is amiable and humorous, making every interaction with him a joy. I appreciate that Ryan recognized my passion for teaching early on, modeled evidence -based instruction, demonstr ated innovative approaches in vi the classroom, and made me a better teacher. Ryan deserves all the accolades (and more!) that he has earned for his scholarship, teaching, and mentoring because he unduly helped me in all of these . One of my first courses in graduate school Šan introduction to the philosophy of science and research methods Šwas with Eric Gonzalez Juenke. Eric socialized me to the discipline and taught me to prioritize sound research design. His research identifying selection bias in the study of minority political candidates led me to consider the selection bias in the policy diffusion literature. I think he is one of the best writers in the discipline, modeling clear and concise prose. Eric also exemplifies extraordinary teaching; he has an unca nny ability to relate pop culture with research methods, piquing students ™ interest in political science. Eric has been one of the most ardent supporters in my corner. Those who know me know that I place high value on kindness. Eric is as nice as he is int elligent, making him a remarkable scholar, adviser, and person. I long admired Chuck Shipan, a distinguished scholar at the University of Michigan, for his acclaimed research on policy diffusion. As a long shot, I invited him to be on my committee as an external faculty member. Everyone would have understood if he had politely declined the request from a student at another institution. You can imagine my delight when he agreed to serve on my committee. And serve he did! Chuck meticulously read, studied, and annotated my dissertation drafts, even pouring over the footnotes and appendices. He drove to East Lansing on multiple occasions to hear about my research and offer discerning feedback. Chuck followed -up with detailed written bullet points summarizing his advice and charting a path forward. His brilliance is only matched by the charitable way he delivers constructive feedback. Chuck is an uncommon scholar and mentor, and I am forever grateful for his beneficence. I was exceptionally lucky that th e estimable scholar I wanted to work directly with during graduate school, Sandy Schneider, was on the selection committee when I applied. She plucked my application from the pile of superb candidates and recommended that MSU recognize my aptitude vii with a U niversity Fellowship. Sandy ™s support did not stop there. On day one, she invited me to co -author meaningful disaster policy research with her. She thought one of my initial term papers was solid enough for publication and encouraged me to submit it. Sandy treated me to coffees and meals to discuss my papers, gently nudged me to polish my writing, and pushed me to read early seminal works, which led me to identify omissions in the literature for this project . She also graciously introduced me to her extensi ve network of scholars, afforded me opportunities to participate in the ICPSR Summer Program in Quantitative Methods, and wrote numerous letters of recommendation on my behalf. Moreover, Sandy modeled how to consume research critically and run a seminar co urse. She mixes probing, open -ended questions with gourmet popcorn, simultaneously fostering deeper engagement with the material and a sweet tooth. Graduate school can produce multiple roadblocks and opportunities for deviations from the end goal. When I veered off course or did not follow her advice, she did not gloat or discount me. On the contrary, Sandy reassured me and remained a steadfast champi on of my potential, believing in me even when I had doubts. I am forever indebted to Sandy for taking a chance on me, for her prescient comments on my research, and for always seeing the best in my work product. As many doctoral students can attest, the best scholars d o not always make the best advisers. My committee was the exception. My only hope is that I can match their generosity as I try to pay it forward fostering the next generation of civically engaged students and budding scholars. Many other MSU faculty and staff offered astute recommendations, thoughtful counsel, or continued reassurance during the dark parts of any research project. I am especially indebted to Michael Bratton, Melinda Gann Hall, and Sarah Reckhow for allowing me to execute some form of this research in their respective courses. All three endorsed the general premise of this project and provided sage advice for advancing it forward. Josh Sapotichne graciously talked through various aspects of my dissertation, directing me to different scholars ™ work and exhibiting as much viii enthusiasm for my ultimate findings as I did. Josh has also been instrumental to my career development, introducing me to the foremost policy process scholars and treating my professional experience advantage ously. Matt Grossmann spent hours reviewing early drafts and discussing different research frames, prompting me to pay more attention to the role of interest groups in the diffusion process. I cannot thank Matt enough for all his support o n this and other projects . His delegatory approach to research is empowering and made me a more confident scholar. Cory Smidt is one of the brightest political methodologists in the discipline , and he always had his door open to me and generously fielded methods and coding questions. Even though we attended rival undergraduate institutions, he never held that against me and always had my best interest at heart. I am also grateful to Bill Jacoby ™s sharp questions during the prospectus phase of the project, which helped me define the key contributions of my research relative to parallel studies. His comprehensive, illustrative, and energetic approach to teaching multiple regression, measurement, and scaling ignited my interest in learning and teaching research methods. My ca reer was further enriched when Bill invited me to serve as the managing editor of the American Journal of Political Science (AJPS) . I was afforded a front -row seat to the editorial process, interacted with numerous established and emerging scholars in our field, and apprised of the latest scientific findings. I also became an ardent disciple for reproducible research. And when the AJPS moved to American University, good fortune provided me the chance to work with Jan Leighley, another exceptional political scientist and mentor. Jan has always provided sage outside counsel. She was incredibly supportive of my career development and recognized my contributions to the Journal . In Jan, I gained another trusted adviser, ally, and friend. Chuck Ostrom, in his capacity as department chair, and Melinda Gann Hall, Tom Hammond, and Ani Sarkissian, in their capacit ies as graduate program directors, deserve special recognition for their abiding support of my studies, research, and career. All four were remarkably ix en couraging, willingly sought out funds to buoy my education, and checked -in with enough frequency to remind me of looming deadlines without stifling the creative process. I would also like to acknowledge Ben Appel, Valentina Bali, Cristina Bodea, Eric Chang , Mike Colaresi, Rachel Croson, Brian Egan, Richard Hula, the late Steve Kautz, Ben Kleinerman, Nazita Lajevardi, Sandy Marquart -Pyatt, Mariana Medina, Ian Ostrander, Dustin Sebell, Jakana Thomas, Michael Wahman, Bryan Wilcox -Archuleta, the late Ken Willia ms, and Krista Zeig for their roles in shaping my understanding of public policy and political science, helping to sharpen my use of methodological tools, or for simply offering words of encouragement or comic relief at some stage in this process. Still, I would not have been able to complete this project or my graduate studies without the constant and able succor from the department ™s assistants: Karren Battin, Rhonda Burns, and Sarah Krause. Scads of MSU graduate students read and commented on drafts, a ttentively listened to elevator pitches or presentations, and prodded with perceptive questions, guaranteeing that this scientific research was as much a public enterprise as any other. I am particularly grateful to the following individuals as their contr ibutions made this project better and made me a better scholar: Miles Armaly, Marcie Babcock McCalmont, Thomas Bentley, Melanie Bowers, Matt Breuer, Erica Briggs, Fang -Yu Chen, Lora DiBlasi, Kesicia Dickinson, Adam Enders, Daniel Fram, Michael Giles, Sung Min Han, Dan Hansen, Petra Hendrickson, Tim Hibbard, William Isaac, Tara Iseneker, Jason Kalmbach, Jonathan King, Alon Kraitzman, Elizabeth Lane, Caleb Lucas, Bob Lupton, Zuhaib Mahmood, Kate Morris, Shayla Olson, Chun Ho Park, Peter Penar, Jonah Ralston, Erika Rosebrook, Chrissy Scheller, Jessica Schoenherr, Jamil Scott, Emma Slonina, Nate Smith, Jacob Snyder, Dan Thaler, Doug Walker, and Matt Zalewski. I am fortunate to have studied with some of the best minds in the discipline, but even more fortunate to call so many of these colleagues friends. x With the financial backing of MSU ™s Department of Political Science, College of Social Sciences, and Graduate School, I was able to present different versions of this research at several academic conferences. I appreciate the feedback from panel chairs, discussants, and participants at the 2015 Michigan Academy of Science, Arts & Letters Annual Conference; the 2015 Midwest Political Science Association Annual Conference; the 2016 Southern Political Science Assoc iation Annual Conference; the 2018 Midwest Political Science Association Annual Conference; and the 2019 Measurement Meets Politics and Policy Conference in Chicago, IL. The following individuals provided noteworthy insights at these conferences or in othe r capacities: Brady Baybeck, Ann Bowman, Becky Bromley -Trujillo, Jerrell Coggburn, Paul Cornish, Daniel Hawes, Renée Johnson, Daniel Lewis, Kristin O ™Donovan, Srinivas fiChinnu fl Parinandi, Mallory SoRelle, and Kimberly Wiley. This project would have been much more challenging had various scholars and organizations not made their data publicly available. I cite the sources of data used throughout the dissertation in the appendice s and reference sections, but I extend a personal thanks to Kimberly Conger, Paul Djupe, Carl Klarner, and Daniel Lewis. MSU ™s Government Information & Political Science Librarian, Julia Frankosky Ezzo , also did some data digging for me as well. And t wo un dergraduate students at Hope College (my alma mater and where I was a faculty member for a year), Irene Gerrish and Joseph McClusky, provided tenacious research assistance collecting and coding nearly 7,800 ballot measures pursued at the U.S. state -level f rom 1902 -2016. I would be remiss, however, if I failed to acknowledge the undergraduate faculty mentors at Hope College that helped mold my early understanding of the discipline and approach to teaching: Annie Dandavati, David Ryden, Joel Toppen, and B oyd fiChacha fl Wilson. Just as influential to my formation and zeal for understanding public policy and politics were my eight years of professional experience before homing in on my vocation. The people I worked with in El Salvador to xi implement community an d economic development initiatives in impoverished communities will always hold a special place in my heart. My tenure there further opened my eyes to the complexity of military intervention, trade, immigration, foreign aid, disaster relief, poverty, and d ebt policies. Likewise, the courageous human rights defenders in Guatemala that I later lobbied on behalf of in Washington, DC as director of the Guatemala Human Rights Commission, and the tireless grassroots activists that stood in solidarity with them, m ade me better understand and appreciate the power of democracy. I not only gained direct knowledge of the legislative process but also learned how to use my voice of privilege for those fisin voz .fl My time managing sales for Neogen Corporation, a multinatio nal food and animal safety company, provided a holistic view of trade, economic development, tax, and food and drug policies. These professional experiences epitomized how policy is as personal as it is public. The people I interacted with during these adv entures influenced my passion for studying politics and thus had some bearing on my graduate studies and the output here. Nonetheless, while the aforementioned individuals immeasurably influenced my scholarship and teaching, the following people sustained and motivated me throughout this project and the arduous odyssey of graduate school. God gave me the strength to burn the midnight oil and opened windows of opportunity for me when doors seemed to close. Numerous neighbors, friends, and family members asked with genuine interest about my research or tenderly inquired about the progress of my dissertation. My extended family afforded me multiple passes to bow out of family parties, reunions, or impromptu ga therings. One pair of in -laws, Richard Douglass and Marian Horowitz, made dinner for my family nearly every week, lightening the household duties. Richard also gave career advice and helped tackle home repairs, while Marian offered writing and grammar tip s and fueled me with her delectable lemon bars. Another pair of in -laws, Tricia and Steve McEuen, visited us and invited us to Colorado xii for respites from the doldrums of academic work. Tricia ™s comforting spirit and scrumptious meals always reassured me, w hile Steve spent hours conversing with me about my research, asking perceptive questions that helped me better distill my contributions to a more general audience . Mary Beth Moore and Doug Ruby, my wife ™s aunt and uncle, provided opportunities for recess a t their cozy cottage in Pentwater, Michigan and at their home in Washington, D.C. Doug jokingly asked about my dissertation ™s progress in my first week of graduate school and Mary Beth ™s acclaimed chocolate chip cookies served as rewards after a hard day™s work. Those named here and so many others unnamed Šbrothers, sister -in laws, brother -in laws, nieces, nephews, grandparents, aunts, uncles, cousins, friends, neighbors Šwere incredibly encouraging and supportive. I would not have arrived at this destination without them. Deep appreciation and gratitude go to my parents, Susan and Clarence Jordan. I am especially proud to be their son. My persevering father worked for thirty -four years as a custodian for our local elementary school. My talented mother made an d sold artisan goods at craft shows around the state. Despite my parents™ best efforts, our family of six was a portrait of the working poor. We relied on government food assistance from time to time, and on subsidized housing loans. At the dinner table, w e discussed grievances in my father™s union, the influx of cheap crafts from abroa d, and the fate of the Earned Income Tax Credit. These childhood experiences fostered in me a passion for trying to understand the political world. Despite our financial circ umstances, my parents instilled in my brothers and me a deep passion for learning and encouraged us to use our skills and talents to serve the common good . More than anyone, they demonstrated the value of hard work and sacrificed everything for us. I would not be who I am today without them. I am forever grateful for their unwavering support at every step of this journey. During graduate school, my parents would frequently travel two and a half hours each way to visit and help care for my children. On several occasions, they dropped their work at a moment™s xiii notice to help us out in a pinch. They furnished my family with freshly picked blueberries, pistachio pie, and too many savory meals to count . My dad suppl ied me with origina l trivia questions to keep my general knowledge fresh . My mom offered words of fortitude, always believing in my abilities. And when this journey became incre asingly more difficult, they volunteered to do even more to lighten my load. I could not have aske d for better parents or an environment to begin to learn about politics and public policy. Still, the mo st significant recognition of all goes to my spouse, Elly Jordan, and our two precocious daughters, Gracie (6 years old) and Dot (3 years old). These th ree bore the burdens of this project and process more than anyone. My daughters, Gracie and Dot, made this journey immensely more gratifying and joyful. They did not treat me any differently when my statistical models would not converge, when my dissertati on prose was shoddy, or when I had to skip out on play time. They showered me with their love, wowed me with their development, and obliged me to tickle them until their laughter filled my soul. Gracie ™s and Dot ™s existence reminded me of what was truly im portant and motivated me forward. Words are insufficient to express my appreciation and awe for my partner, Elly. Elly encouraged me to take a leap of faith and pursue my vocational dream of becoming a political scientist and professor. She was the first to listen patiently to my ill -formed ideas, improved my own understanding of the political world with her insightful questions, and offered feedback that greatly enhanced my research. Elly provided first -rate advice on how to overhaul complex sentences and revamp syntax. She graciously laughed at my awful jokes while all together keeping me in good spirits. Elly even had the patience to help me format the dissertation ™s Table of Contents, the final straw that breaks most scholars ™ backs. What is more, she frequently put in overtime to care for our children and tackle household tasks so that I could have the space to read, collect and analyze data, and write. Further underscoring xiv how special she is, Elly was willing to ride the roller coaster of the academic job search and uproot our family to make my dreams come true. Elly did this all the while providing exceptional legal counsel to immigrants, sexual abuse victims, and other clients marginalized by society and the law. Her extraordinar y intellect is only matched by her level of compassion and generosity toward others. At various points along the journey, it was her steadfast support and belief in me that kept me going. Undoubtedly, my doctoral degree is as much hers as it is mine. I am incredibly blessed to share this life with Elly, Gracie, and Dot Šthree amazing and beautiful humans. I could not have achieved any of this without them. xv TABLE OF CONTENTS LIST OF TABLES ............................................................................................................................................. xviii LIST OF FIGURES .............................................................................................................................................. xx CHAPTER 1: INTRODUCTION ...................................................................................................................... 1 The Purpose of this Research ........................................................................................................................... 5 The Ca se for Venue Diffusion ......................................................................................................................... 9 Structure of the Dissertation ........................................................................................................................... 12 Advice Before Reading ..................................................................................................................................... 14 CHAPTER 2: POLICY DIFFUSION ............................................................................................................. 18 Mechanisms of Policy Diffusion .................................................................................................................... 22 Additional Factors Important to Policy Diffusion .................................................................................... 24 Conclusion .......................................................................................................................................................... 28 CHAPTER 3: DIFFUSION DYNAMICS OF BALLOT MEASURES ................................................. 29 Policy Diffusion ................................................................................................................................................. 30 The Myopic Focus on Legislative Arena ...................................................................................................... 32 Policy Diffusion Research™s Selection Bias .................................................................................................. 34 Ballot Measures Pursued in the U.S. States ................................................................................................. 36 The Diffusion of Ballot Measures: Expectations ....................................................................................... 46 Data and Methods ............................................................................................................................................. 51 Data .................................................................................................................................................................. 51 Variable Operationalization ........................................................................................................................ 54 Methods .......................................................................................................................................................... 58 Results for Diffusion of Ballot Measures ..................................................................................................... 59 Evaluating Selection Bias ................................................................................................................................. 68 Conclusion .......................................................................................................................................................... 69 CHAPTER 4: A THEORY OF VENUE DIFFUSION ............................................................................. 72 Venue Shopping ................................................................................................................................................ 74 Frequency of Venue Shopping ....................................................................................................................... 75 What M otivates Venue Shopping? ................................................................................................................ 78 Venue Diffusion and Political Learning ....................................................................................................... 81 xvi Alternative External Factors Driving Venue Choice ................................................................................ 92 Internal Factors Driving Venue Choice ....................................................................................................... 97 Implications of Venue Diffusion ................................................................................................................. 102 Conclusion ........................................................................................................................................................ 105 CHAPTER 5: POLITICA L LEARNING AND THE D IFFUSION OF GAY MARR IAGE POLICIES ............................................................................................................................................................. 106 Why the Policy Case of Gay Marriage? ...................................................................................................... 109 Mobilization for and Counter -Mobilization Against Gay Marriage ..................................................... 111 Past Diffusion Research of Gay Marriage Policies .................................................................................. 120 Four Missing Piece s to the Puzzle ............................................................................................................... 121 The Diffusion of Gay Marriage Policies: Expectations .......................................................................... 126 Data and Methods ........................................................................................................................................... 130 Data ................................................................................................................................................................ 130 Variable Operationalization ...................................................................................................................... 133 Methods ........................................................................................................................................................ 140 Results for Anti -Gay Marriage Policies ...................................................................................................... 144 Diffusion of Pro -Gay Marriage Policies ..................................................................................................... 156 Results for Pro -Gay Marriage Policies ........................................................................................................ 158 Robustness Checks .......................................................................................................................................... 163 Is Political Learning Simply Policy Learning? ........................................................................................... 164 Conclusion ........................................................................................................................................................ 165 CHAPTER 6: THE DIFFU SION OF VENUE CHOICE ...................................................................... 167 Venue Diffusion and Political Learning ..................................................................................................... 169 Qualitative Evidence of Venue Diffusion in Fight for Gay Marriage ................................................. 173 Role of State and National Interest Groups .............................................................................................. 178 Venue Diffusion in Fight over Gay Marriage: Expectations ................................................................. 181 Data and Methods ........................................................................................................................................... 186 Data ................................................................................................................................................................ 186 Variable Operationalization ...................................................................................................................... 188 Methods ........................................................................................................................................................ 190 Results for Anti -Gay Marriage Policies ...................................................................................................... 191 Results for Pro -Gay Marriage Policies ........................................................................................................ 200 Robustness Checks .......................................................................................................................................... 208 Conclusion ........................................................................................................................................................ 210 xvii CHAPTER 7: CONCLUSIO N ........................................................................................................................ 212 The Takeaway ................................................................................................................................................... 212 Unanswered Questions .................................................................................................................................. 215 Moving Forward .............................................................................................................................................. 218 APPENDICES ..................................................................................................................................................... 222 APPENDIX A ................................................................................................................................................. 223 APPENDIX B ................................................................................................................................................. 230 APPENDIX C ................................................................................................................................................. 234 APPENDIX D ................................................................................................................................................ 257 REFERENCES .................................................................................................................................................... 271 xviii LIST OF TABLES Table 1.1: Key Concepts ....................................................................................................................................... 15 Table 3.1: Institu tional Arrangements for Direct Democracy by State ..................................................... 38 Table 3.2: Description of 50 Randomly Selected Ballot Measures ............................................................. 53 Table 3.3: Ballot Measure Diffusion Models ................................................................................................... 63 Table 4.1: Assessing Venue Choice for Sample of 95 Policies .................................................................... 77 Table 4.2: Assessing Venue Cho ice for Sample of 95 Policies by Policy Category ................................. 78 Table 5.1: Policy Diffusion of Anti -Gay Marriage Policies using Binary Logistic Regression ........... 146 Table 5.2: Policy Diffusion of Anti -Gay Marriage Policies using Mult. Logistic Regression ............. 151 Table 5.3: Policy Diffusion of Pro -Gay Marriage Policies using Mult. Logistic Regression ............... 159 Table 6.1: Venue Choice to Pursue Anti -Gay -Marriage Policies, 1993 Œ 2015 ...................................... 172 Table 6.2: Venue Choice to Pursue Pro -Gay -Marriage Policies, 1993 - 2015 ........................................ 173 Table 6.3: Venue Diffusion of Anti -Gay Marriage Policies using Mult. Logistic Regression ............ 192 Table 6.4: Venue Diffusion of Pro -Gay Marriage Policies using Mult. Logistic Regression .............. 201 Table A.1: Ballot Measures Model™s Var. Descriptions, Descriptive Statistics, and Sources .............. 228 Table B.1: Choice of Venue and Diffusion Statist ics for Sample of 95 Policies ................................... 230 Table C.1: State by State Chronology of Anti - and Pro -Gay Marriage Policies, 1993 Œ2015 .............. 234 Table C.2: Pursuit of Anti -Gay Marriage Policies by Venue, Year, and State, 1993 -2015 .................. 243 Table C.3: Pursuit of Pro -Gay Marriage Policies by Venue, Year, and State, 1993 -2015.................... 244 Table C.4: Anti -Gay Marriage Models™ Variable Descriptions, Descriptive Statistics, and Sources . 248 Table C.5: Pro -Gay Marriage Models™ Var. Descriptions, Descriptive Statistics, and Sources .......... 250 Table C.6: Robustness Check: Policy Diffusion of Anti -Gay Marriage Policies Using CLogLog .... 252 Table C.7: R obustness Check: Policy Diffusion of Pro -Gay Marriage Policies Using CLogLog ...... 253 Table C.8: Robustness Check: Anti -GM Policies using Cox -Proportional -Hazards Model ............... 254 xix Table C.9: Robustness Check: Pro -GM Policies using Cox -Proportional -Hazards Model ............... 254 Table C.10: Policy Diffusion of Anti -GM Policies using Mult. Log. Reg. Clustered by State ........... 255 Table C.11: Policy Diffusion of Pro -GM Policies using Mult. Log. Reg. Clustered by State ............. 256 Table D.1: Anti -Gay Marriage Models™ Var. Descriptions, Descriptive Statistics, and Sources ........ 257 Table D.2: Pro -Gay Marriage Models™ Var. Descriptions, Descriptive Statistics, and Sources ......... 259 Table D. 3: Venue Diffusion of Anti -GM Policies using Logit, Comp. Log -Log, and Ord. Logit .... 263 Table D.4: Venue Diffusion of Anti -GM Policies using Cox -Proportional -Hazards Model ............. 264 Table D.5: Venue Diffusion of Pro -GM Policies using Logit, Comp. Log -Log, and Ord. Logit ..... 267 Table D.6: Venue Diffusion of Pro -GM Policies using Cox -Proportional -Hazards Model .............. 268 Table D.7: Venue Diffusion of Anti -GM Policies using Mult. Logistic Reg. Clustered by State ...... 269 Table D.8: Venue Diffusion of Pro -GM Policies using Mult. Logistic Reg. Clustered by State ....... 270 xx LIST OF FIGURES Figure 1.1: U.S. States Legalizing Recreational Marijuana via Citizen Initiatives, 2010 Œ 2018 .............. 4 Figure 3.1: Total Ballot Measures by Type by Decade, 1900 Œ2010 ............................................................ 40 Figure 3.2: Ballot Measure Passage Rate by Ballot Measure Type and Decade ....................................... 41 Figure 3.3: Ballot Measure Passage Rate by State ........................................................................................... 43 Figure 3.4: Total Ballot Measures Attempted by U.S. State from 1902 Œ 2016 ....................................... 44 Figure 3.5: Ballot Measures by Frequency of Policy Area ............................................................................ 45 Figure 3.6: Number of Ballot Measures from Sample that Have Diffused or Have Yet to Diffuse .. 61 Figure 3.7: Predicted Probability of Adopting Ballot Measure as Policy Learning Increases ............... 64 Figure 3.8: Average Marginal Effects of Key Variables from Diffusion of Ballot Measures Model .. 65 Figure 4.1: Policy Actors™ Venue Shopping Considerations ......................................................................... 81 Figure 4.2: Policy Actors™ Venue Shopping Considerations Also Includes fiPolitical Learningfl ........ 85 Figure 4.3: External and Internal State -Level Forces Influencing Venue Choice ................................. 102 Figure 5.1: Adoption of Gay Marriage Bans by U.S. State by Venues, 1995 Œ 2010 ........................... 115 Figure 5.2: Adoption of Pro -Gay Marriage Policies by U.S. State by Venues, 1995 Œ 2010 ............... 119 Figure 5.3: Average Marginal Effects for Key Anti -Gay Marriage Policy Predictors ........................... 153 Figure 5.4: Average Marginal Effects for Key Pro -Gay Marriage Policy Predictors ............................ 160 Figure 6.1: Average Marginal Effects of Key Variables on Venue Diffusion for Anti -GM Model .. 194 Figure 6.2: Political Learning™s Effect on Venue Choice Over Time for Anti -GM Policies .............. 195 Figure 6.3: Average Marginal Effects of Key Variables on Venue Diffusion for Pro -GM Model ... 203 Figure 6.4: Political Learning™s Effect on Venue Choice Over Time for Pro -GM Policies ................ 204 Figure A.1: Ballot Measures by Frequency of Policy Area by Type of Measure ................................... 223 Figure A.2: Random Sample of 50 Ballot Measures by Topic Area ......................................................... 225 Figure A.3: Random Sample of 50 Ballot Measures Pursued by Decade ................................................ 226 xxi Figure A.4: Random Sample of 50 Ballot Measures Pursued by Topic Area ......................................... 227 Figure C.1: Probability of Adopting Anti -Gay Marriage Policy by Key Explanatory Variables ........ 245 Figure C.2: Prob. of Adopting Anti -GM Policy by Venue as Political Learning Increases ................ 246 Figure C.3: Prob. of Adopting Pro -GM Policy by Venue as Political Learning Increases .................. 247 Figure D.1: Pred. Prob. of Picking Venue for Anti -GM Policies as Political Learn ing Increases ..... 261 Figure D.2: Pol. Learning and Time™s Interactive Effect on Venue Choice for Anti -GM Policies .. 262 Figure D.4: Pol. Learning and Time™s Interactive Effect on Venue Choice for Pro -GM Policies .... 266 1 CHAPTER 1: INTRODUCTION In 2009, Richard Lee Ša marijuana user, entrepreneur, and founder of Oaksterdam University, the first cannabis trade school in the United States Šlaunched a citizen initiative campaign in California to legalize the recreational use of marijuana (Hecht 2014; Martin and Rashidian 2014). California was the first state in the union to permit the use of marijuan a for medicinal purposes in 1996. Lee wanted the Golden State to be the first to legalize the smoking of pot for any Californian twenty -one years of age or older, explicitly flouting U.S. law (Kamin 2015; Pickerill and Chen 2008). And Lee believed that a b allot initiative, allowing voters to have a direct say on this policy, was the best route to get this done. Other activists, growers, and even representatives from statewide and national marijuana movements to legalize hemp use, including the California Le aders for the National Marijuana Policy Project and the National Organization for the Reform of Marijuana Laws (NORML), encouraged Lee to wait. They cautioned that it was too soon, that younger voter turnout would be too low in an off -year election, and th at adult recreational use was politically untenable (Hecht 2014). Lee persisted. He bankrolled more than $1.6 million to pay for signature gatherers, secure a spot for Proposition 19 (Prop. 19) on California™s November 2010 ballot, and rally support around his campaign (Hecht 2014; Martin and Rashidian 2014). 1 Lee™s campaign for Prop. 19 largely framed the legalization of cannabis as (1) a job creator (Starrs and Goin 2010), (2) a way to increase tax revenue, and (3) a way to save taxpayer money, by elimi nating costly criminal justice policies employed to enforce prohibition (Ballotpedia 2010; Hecht 2014; Martin and Rashidian 2014). A diverse coalition of individuals and interest groups endorsed 1 California™s November 2010 ballot measure to legalize recreational marijuana was not the first in the country, although it was the first in nearly a quarter century. In 1986, pro -reform activists and i nterest groups in Oregon successfully got Ballot 5 Measure on the voting ticket. The measure, intending to legalize the use of cannabis, was defeated having only garner ed 26% of the vote. 2 Prop. 19, including George Soros, Clint Eastwood, Snoop Dog, the United Food and Commercial Workers Union, the NAACP, the ACLU, and various elected officials (Ballotpedia 2010; Hecht 2014). Of course, the measure also had plenty of detractors. Many in law enforcement, dozens of elected officials, the state Attorney General, the National Black Churches Initiative, Mothers Against Drunk Driving, and even some marijuana activists opposed the initiative (Ballotpedia 2010). Indeed, several supporters and legal growers of medical marijuana worried that Prop. 19 carried too severe of penalties for minors, would drive down prices, or would put established growers out of business (Hecht 2014). Amid mounting opposition within the pro -marijuana movement, California legislative action decriminalizing recreational use, and threats of stiff federal enforcement, the ballot initiative failed (Martinez 2010). 2 Californians rejected Richard Lee™s measure 53.5% to 46.5% (Ballotpedia 2010). Regardless of the loss, Lee™s entrepreneurial spirit to change public policy and opinion in favor of cannabis spurred activists in other states also to pursue legal recreational use of marijuana and to do so via direct democracy. Following Lee™s lead, policy actors in three states ŠColorado, Oregon, and Washington Šput forward citizen initiatives in 2012 to legalize recreational use of marijuana in those states. At least one of the organizers of the Colorado ballot initiative, Dan Rogers, attended Oaksterdam University and directly strategized with Richard Lee about how best to achieve the legalization of marijuana in Colorado (Hecht 2014). Rogers drafted an initiative that 2 Witnessing the increasing support for some type of action on ca nnabis, the California legislature passed SB 1449 in October 2010 to lessen the criminal penalty for the possession of less than one ounce of marijuana from a criminal misdemeanor to a civil infraction (Ballotpedia 2010). Governor S chwarzenegger , an oppone nt of Prop. 19 despite at least one public incident of smoking pot during his body -building career (Grace 2002), signed the bill into law. With legislative action, many thought Prop. 19 was moot. Moreover, the federal government had sent contradictory sign als. In 2009, the Obama administration issued the Ogden Memo indicating that federal resources would not be used to pursue individuals using medical marijuana in compliance with existing state law (Kamin 2015). However, weeks before the November 2010 elect ion, Attorney General Eric Holder said the Justice Department would not allow California to blatantly flout federal law. He asserted he would rely on the Controlled Substances Act to fivigorously enforcefl federal law and go after individuals and organizatio ns using, growing, or distributing marijuana for recreational use, even if voters passed the ballot initiative (Hoeffel 2010). 3 included language guaranteeing greater state regulation of the cultivation, distribution, and sale of cannabis. Rogers™ effort also amassed greater unity of support among medical mariju ana growers and activists within the state (Hecht 2014). In the end, t he ballot proposals won in Colorado and Washington but narrowly lost in Oregon (Barcott 2015) . Following these wins in Colorado and Washington , pro -reform activists and interest groups collected enough signatures in Alaska and again in Oregon for November 2014 ballot proposals. Both direct democracy measures in Alaska and Oregon passed. 3 Ohio put forward a similar ballot measure in 2015. Although voters defeated the Ohio proposal, activi sts and interest groups put forward citizen initiatives for November 2016 in five other states: Arizona, again in California, Maine, Massachusetts, and Nevada (MPP 2016; NCSL 2018). These plebiscitary questions all passed except in Arizona, where it lost by a slim 2.5 percentage points. In 2018, Michigan voters passed recreational pot use also at the ballot box, 4 while activists in numerous other states are pla nning citizen initiative campaigns to approve adult -use of marijuana in future elections (NCSL 2018). 5 Figure 1.1 illustrates the passage of policies to legalize recreational marijuana use across U.S. states by citizen initiative. 3 Also in November 2014, voters in the District of Columbia overwhelmingly approved Initiative 71, which legalized the possession and cultivation of limited amounts of marijuana by adults twenty -one years of age and older. However, since the U.S. Congress has jurisdiction over the capital city (afforded by the U.S. constitution), lawmakers passed a series of measures curtailing the implementation of the voter -backed initiative. 4 At a recent roundtable, Sam Pernick, the organizing director for MI Legalize Šthe state group spearheading the legalization of recreational cannabis, commented that the decision to pursue a policy change via citizen initiative was due to political calculations within the state, the prior success of the medical marijuana ballot measure in Michigan in 2008, and witnessing the routes that other states had previously taken to sanction marijuana use (Pernick 2 019). 5 As of 2017, state -level direct democracy had been the only vehicle to legalize recreational marijuana. Nonetheless, in 2018, Vermont™s state legislature legally approved recreational marijuana, becoming the first and only state legislature to do so , although several other state legislatures are debating statutes to either legalize or decriminalize possession of pot. Importantly, Vermont does not allow citizen initiatives. Legislators in Maine, Massachusetts, and Washington are also considering bills to repeal the voter initiatives in those states that legalized the production, sale, and use of recreational marijuana (NCSL 2018). Due to the horizontal diffusion across U.S. states, there are now conversations at the federal level of legalizing recreati onal cannabis (Higdon 2019). 4 Figure 1.1: U.S. St ates Legalizing Recreational Marijuana via Citizen Initiatives, 2010 Œ 2018 Policies aimed at legalizing the recreational use of marijuana are spreading across U.S. states. Individual activists, like Dan Rogers, along with state and national interest g roups, like NORML and Marijuana Policy Project, are following Richard Lee™s innovative push for more lax marijuana laws. But they are emulating more than just the policy . These policy actors have also learned about the successful and failed tactics employe d by the Prop. 19 campaign in California. Beyond the policy itself, these actors are also copying Richard Lee™s choice of institutional venue Ša citizen initiative Što pursue policy change. Lee could have attempted statutory legalization of marijuana by way of the state legislature, or pressing legislators to call a referendum, or lobbying the governor to issue an executive order, or bringing forward a legal argument in the state or federal courts. For a host of political and institutional reasons, Lee pursue d change via a ballot initiative. Policy actors in many other states have followed suit, emulating Richard Lee™s choice of venue in their states™ effort to 5 legalize pot. These actors did not just copy the innovative policy; they also copied the choice of institutional site to pursue policy change. The central question is why? Why did advocates of recreational marijuana pursue policy change at the ballot box rather than via a different route? The Purpose of this Research The circumstances describe d above around the spread of legal -marijuana -use laws are not unique. More than five decades of research exploring policymaking in the U.S. states offer strong evidence that policy activity in one state depends on, at least in part, the policy activities i n other states ( Berry and Berry 1990; Boushey 2010; Gilardi 201 6; Graham, Shipan, and Volden 2013; Gray 1973; Karch 2007 a, 2007b ; Rogers 1962; Savage 1985; Shipan and Volden 2006 , 2008 ; Volden 2006; Walker 1969 ). Actors in one state facing a common societa l problem frequently turn to and seek out policy innovations Ša program or policy which is new to the governmental unit adopting it, no matter how old the program or how many other governments have already enacted it Šadopted in other states to address the s ame issue. Scholars refer to this phenomenon as policy diffusion . Given the increasing societal acceptance of marijuana use, medical research indicating cannabis™ s palliative health properties, and the increasing costs of ineffective enforcement and criminal justice policies targeting pot use, it is not surprising that policy actors in other states followed the lead of California, copying its policy innovation to le galize the recreational use of marijuana in their states. Nor is it that surprising that these policy actors, like Richard Lee in California, also pressed for more lenient marijuana laws via citizen initiatives, as citizen -driven ballot measures have becom e a popular venue to press for policy change. What is surprising, however, is that the spread of new ideas via different policy venues is se verely understudied. Save for a small number of publications researchers have overwhelmingly limited their focus on the spread of new policy innovations from one legislative body to another legislative body. 6 Diffusion scholars have mostly ignored the interdependence of decision -making outside of the traditional legislative context. This is curious considering that in t he American federated system, policy actors have multiple institutional venues Šgovernmental arena s with formal and informal rules that structure how actors make collective decision s and decide on public policies Šavailable at the state level to pursue new p rograms and policies. These institutional venues include state legislatures, citizen initiatives , popular referenda, legislative referenda, state court decisions, gubernatorial executive orders, state administrative agencies, Congress, federal courts, and the federal bureaucracy, among others. Indeed, since the 1970s, individuals and interest groups have increasingly pressed for policy change outside the fipeople™s branchfl and via a multitude of institutional venues (Damore, Bowler, and Nicholson 2012; Magl eby 1988; Miller 2009; Reilly 2010 ). For instance, citizen initiatives have been used to pass fithree -strikesfl laws to punish repeat criminal offenders, reinstate the death penalty, legalize marijuana for medical or personal use, and ban same -sex marriage. Governors have signed executive orders mandating renewable energy standards, while state and federal courts have struck down statutes and amendments outlawing same -sex unions. Nonetheless, we know little about the diffusion dynamics of these innovations wh ile accounting for the different venues in which they are pursued. This research contributes to the policy diffusion scholarship by documenting and examining the transmission of policy innovations across multiple institutional venues . More specifically , I leverage the spread of policies across multiple venues to address three omissions in the policy diffusion literature. First, I move us beyond the myopic legislative context by mapping the patterns of diffusion of ( a) ballot measures across U.S. states fro m 1902 Œ 2016 , and ( b) anti - and pro -gay marriage policies via state legislatures, legislative referenda, citizen initiatives, state judiciaries, and federal courts. Doing so helps reinforce and refine our understanding of the dynamics of policy transfer. Second, I employ an established, but underutilized, modeling strategy 7 that better accounts for the spread of innovative ideas via multiple competing arenas. Multinomial logistic regression allows us to simultaneously recognize each factor™s contribution to policy adoption in the respective venues and uncover inter -venue dynamics. Thir d, I address persistent selection bias in diffusion studies, whereby researchers examine policies that knowingly diffuse to a plurality of jurisdictions without considering pol icies that have yet to spread or only spread narrowly. I offer some evidence that policy scholars tend to overstate the occurrence of policy diffusion and overestimate the effect of its key mechanisms. But this dissertation does more. What is equally as su rprising as understudying policy diffusion via multiple venues is that we have not asked whether the choice of institutional venue to press for policy change in one state Ša vital aspect of the agenda -setting process Šinfluences the venue shopping process in other states. Rather than making insular, independent decisions about the fibestfl institutional venue to press for a favorable change, policy actors may learn about the paths taken by policy entrepreneurs and other actors to bring an innovation to the gove rnmental market. For example, Richard Lee™s decision to press for legal marijuana use via citizen initiative in California may have influenced Dan Rogers and policy actors™ decision in other states to pursue the same policy via ballot initiatives. Fusing t he policy -diffusion and venue -shopping literatures, this research also attempts to answer to what degree policy advocates™ choice of venue to press for a new idea in one state influences other states™ venue selection to pursue the same innovation. In particular, I theorize that a government™s choice of venue to pursue a policy is influenced by the prior venue choices of other governments pursuing the innovation, a phenomenon I term venue diffusion . I charge that policy actors look to an d learn from others, purposively seeking a solution to their common problem (i.e., policy learning) and how best to achieve that solution (i.e. , political learning). For example, Dan Rogers and policy actors in subsequent states all learned from Richard Le e™s choice of venue and campaign strategies to pursue recreational marijuana in their 8 states. Political learning has mostly been omitted from our models and understanding of policy diffusion ( Gilardi 2010; Heclo 1974; May 1992; Rose 1991). This presents an opportunity. By expounding on political learning and incorporating it into our models of the policy process, we are better able to explain the dynamics of policy diffusion and offer concrete evidence of venue diffusion. The upshot of this dissertation is fourfold. First, I offer evidence of policy diffusion via institutional venues beyond the legislative context. Although the results reinforce much of our understanding of policy diffusion, accounting for states™ varying institutional arrangements refines the precise nature of this process. Second, despite the occurrence of policy diffusion in other institutional venues, I show that policy scholars may be overstating the rate of policy diffusion and overestimating key mechanisms effect on the process. I fin d that nearly half of all ballot measures pursued across the U.S. states from 1902 Œ 2016 did not diffuse to other states , and nearly three -quarters of the ballot measures have yet to be enacted by other states or were enacted by less than a handful of sta tes. Moreover, when I rerun the models excluding the policies that do not diffuse or only spread narrowly, policy learning™s effect on enacting a policy is twice as large compared to the full set , potentially inflating the key mechanism™s role in the proce ss. Third, while past policy research may have overstated the occurrence of policy diffusion and policy learning, past research has also understated how frequently policy actors draw political lessons from policy entrepreneurs and early movers about other aspects of the policy process (i.e., political learning), including venue choice. The inclusion of political learning in policy diffusion models significantly improves our understanding of why states adopt new ideas. Policy actors not only learn about ava ilable policy solutions to shared problems but also gain insights on how to politically achieve the policy solution. 9 Lastly, I provide qualitative and quantitative evidence that venue diffusion occurs, at least in the context of a morality policy . States are more likely to pick a venue to pursue an innovation as other states successfully enact the policy via the same route. Evidence of venue diffusion also suggests that other elements of the policy process Šframing, routing policy opponents, coalit ion building, campaign tactics Što bring an innovation to market in the governmental arena may also transfer across states. The interdependence between states appears to go beyond the copying of a policy idea to emulating fundamental components of the polic ymaking process. Ultimately, what happens in your state doesn™t stay in your state. The Case for Venue Diffusion Policy entrepreneurs, those i nnovative individuals or groups that are t he first to pursue a new policy within their governmental jurisdiction, tactically pick an institutional venue . Entrepreneurs prioritize the venue in which they believe they have a comparative political and resource advantage, is most accessible and amenab le to the policy image, and has the best chance to bring about policy change and ensure policy longevity. Entrepreneurs investigate and consider the full set of institutional venues available to them to press for policy change. However, policy actors, t hose individuals or groups within and outside the public sector that follow the lead of policy entrepreneurs to advocate for the same policy innovation in other governmental jurisdictions , suffer from bounded rationality in their decisionmaking. At times t hey have limited information. Other times they face overabundant information. Still, these policy actors have limited resources (e.g., cognition, capital, energy, time), thus relying on heuristics to make decisions to optimize outcome s (Simon 19 72, 1985 ; Tversky and Kahneman 1974 ). As a result, policy actors engage in satisficing, looking to policy entrepreneurs in other states for the ‚best™ policy solution and ‚best™ political process to achieve policy change. Policy actors not only learn about the 10 conten t of a policy, including the problem, the goals, instruments, and implementation design of the solution (i.e., policy learning), a common tenet of policy diffusion. But policy actors also learn about how to navigate and manipulate the policy process to adv ance the policy (i.e., political learning) (May 1992). In turn, I charge that it is not only the innovative policy that spreads from state to state (i.e., policy diffusion), but also the choice of venue that diffuses (i.e., venue diffusion). Moreover, I credit political learning as the driving mechanism of venue diffusion. To be sure, policy actors weigh other internal and external factors in picking an institutional route to press for policy change. These may include considering the political, economic, institutional, demographic, or interest -group contexts within a state. And they may include external forces such as the venue choice of geographic neighbors, jurisdictions with similar institutional arrangements, federal intervention, the national politic al context, or policy coalition influence. Even after accounting for plausible alternative external and internal pressures, I contend that policy actors contemplating a path to enact a policy also consider the successful paths previously taken in other sta tes to adopt the same policy. The results presented in later chapters suggest a probabilistic relationship, not a deterministic one. Some policy actors within a state may have relied on an insular, independent assessment of the full set of venue options, w eighing only internal factors to pick a route. Most policy actors, however, likely picked a venue by considering both internal and external information. This proposition of venue diffusion is something marathon runners trying to qualify for the Boston Mara thon are familiar with. As the world™s oldest annual marathon and one of the most prestigious racing events, earning a spot in the Boston Marathon is challenging. To fibe in the running,fl racers must complete a certified marathon with a qualifying time for their age group within a specific period before the Boston Marathon. On average, o nly ten percent of marathon finishers qualify. 11 Knowing this, marathon runners trying to qualify for the Boston Marathon also engage in satisficing, learn from others, and st rategically pick the marathon that will help secure a spot at Boston. They do not research every possible qualifying marathon among the universal set of potential races and choose the venue that best optimizes their chances, independently deciding in a vac uum. Instead , they gain tips from racing articles and magazines , get advice from previous qualifiers, and follow the lead of others that have tactically selected the marathon course that increases their chance of qualifying. There are entire articles, blog s, and websites dedicated to promoting the fibest -Boston qualifiersfl Šthe marathon courses that will give the best chance at running the necessary qualifying time for the Boston Marathon. Runners are more likely to pick flat and fast courses, instead of elev ated and sluggish courses, to attempt their qualification. Understandably, Detroit is a popular choice for runners trying to qualify; Denver , not so much. Much like picking the right marathon course can increase a runner™s chances of qualifyi ng for the Boston Marathon, picking the right institutional venue can augment the odds a policy is adopted and entrenched in the political system. Just as current runners learn from and rely on the advice of former Boston qualifiers to pick the race that o ptimizes their chances of success, so too do policy actors learn from and rely on the venue shopping experience of policy entrepreneurs and other policy actors to choose the optimal arena to achieve successful policy change. Venue choice matters. It matter s to an innovation™s success and its entrenchment in the status quo. I charge that policy actors do not decide in which arena to press for new ideas in a vacuum. As they learn about a policy previously pursued by others, they also learn about the political tactics and paths others followed to achieve a policy win. 12 Structure of the Dissertation The next chapter offers a synthesis of what we know about policy diffusion. I recount what multiple generations of diffusion research have taught us about horizontal and vertical interdependence between governmental units. I describe the incremental learning proce ss that generally characterizes the spread of policy innovations, and I nod to the instances when diffusion is rapid and inconsistent with a model of learning. I summarize the pri mary external mechanisms driving policy diffusion (i.e., policy learning, geo graphic, competition, coercion, imitation), as well as highlight additional internal factors (e.g., policy, political, institutional, economic) that influence the adoption of ideas across states. In Chapter 3, I identify two gaps in the policy diffusion literature and articulate why they are problematic. I highlight past research™s myopic emphasis on the spread of policy innovations from legislative context to legislative context to the exclusion of other venues. Moreover, I point out past research™s omis sion from our models those policies that have yet to be enacted by others or have only been adopted by a few states. Relying on the full set of all legislative referenda, citizen initiatives, and popular referenda pursued across the U.S. states from 1902 Œ 2016, I describe how policy actors have increasingly pressed for policy change via ballot measures with varying degrees of success across issue area, ballot measure type, time, and space. I then use a random sample of ballot measures from the full set, in cluding measures not yet adopted by others to those enacted widely, to uncover the diffusion dynamics of ballot measures. While the forces driving the spread of ballot measures largely mirrors the forces responsible in the legislative context, I offer evid ence that past selection bias may overstate the occurrence of diffusion and overestimate the impact of the main mechanisms. I lay the theoretical foundation for venue diffusion in Chapter 4. I start by providing an overview of venue shopping and demonstr ate that policies are frequently pursued outside the 13 fipeople™s branch.fl I review the venue shopping literature™s complementary and sometimes competing rationales for what motivates venue choice. I then integrate these venue shopping theories with a politic al learning explanation between policy entrepreneurs and actors for the possible diffusion of venue selection across U.S. states. I also identify alternative external and internal forces that could account for policy advocates™ choice of venue. All the whi le, I lay out the hypotheses to be tested in the subsequent chapters. I end the chapter by identifying the implications of venue diffusion and outline why evidence for such a phenomenon matters. Chapter 5 recounts the policy fight over same -sex marriage by the religious right and gay rights movements and explains why this policy case is useful to better understanding both policy diffusion and venue diffusion. I then identify four missing pieces to the fuller puzzle of policy diffusion from past research on this policy case including failing to account for the spread of these policies across multiple venues, ignoring the diffusion of pro -gay marriage policies, controlling for the opposition™s policy successes, and integrating political learning into the di ffusion process. Next, I lay out my expectations for political learning and other known external and internal determinants in predicting the adoption of anti - and pro -gay marriage policies. I follow this by describing the data, detailing my measurement cho ices, and justifying a multinomial logistic regression modeling strategy. The empirical results reveal that a multinomial logistic approach allows us to capture the inter -venue dynamics of policy diffusion better and establish political learning as a centr al mechanism of policy transfer. Political learning™s marginal effect was more substantial than nearly every other external and internal factors™ impact on explaining a state™s decision to prohibit or permit same -sex unions. I conclude the chapter by carry ing out robustness checks of the models and variable operationalizations, and by making the case that political learning is different conceptually and empirically from policy learning. 14 In Chapter 6, again relying on the policy case of gay marr iage, I provide both qualitative and quantitative evidence for venue diffusion. The qualitative narrative describes how early policy entrepreneurs engaged in strategic venue shopping in the state courts to press for marriage equality and how subsequent pol icy actors learned from these tactics to also advocate for same -sex unions in other state courts and venues. I highlight how both national and state -level interest groups played a role in pursuing and spreading these policy innovations, and that treating t he fight over gay marriage as only a top -down process would be a mistake. After detailing expectations, describing the data, and variable operationalizations, the empirical results from both anti - and pro -gay marriage models support the existence of venue diffusion and identify political learning as the principal mechanism driving venue choice across states. Policy actors learn from the successful venue shopping choices by early mover states and especially prioritize cues from states analogous along institu tional and political dimensions. Although interest groups™ role in the process is less clear, what is clear is that political learning™s effect on venue choice varied over time as policy actors processed in real time other states™ venue successes and failu res. Robustness checks of the models at the end of the chapter further bolsters the chapter™s claims. The conclusion trails in Chapter 7. I summarize the key takeaways and also suggest avenues for future research to build on and expand this new knowl edge base. I finally offer supplemental materials, including tables, figures, and detailed lists of states™ adoptions for ballot measures and anti - and pro -gay marriage policies in the Appendices. Advice Before Reading Before turning to the next chapter which review s the policy diffusion literature, I offer some advice to facilitate the reader™s understanding of this research. First, I provide a description, some examples, and references for the key concepts that underlie this research in Table 1.1. My h ope is 15 Table 1.1: Key Concepts Concept Definition Sources bounded rationality limited information, limited cognition to deal with overabundant information, or limited resources, thus relying on heuristics to make decisions to optimize the outcome Simon 1972, 1985; Tversky and Kahneman 1974; Weyland 200 6 institutional venue a governmental arena with a set of informal and formal rules that structure and guide how actors make collective decisions and decide on public policies; also known as a fipolicy venuefl Examples: state legislature, state legislative referendum, citizen initiative, state high court, gubernatorial executive order, state bureaucratic agency, federal government, etc. Baumgartner and Jones 1993; Kingdon 1984; Lubell 2013 policy actor individuals or groups within and outside the public sector that follow the lead of policy entrepreneurs to advocate for the same policy in other governmental jurisdictions ƒ Author™s definition policy diffusion the process by which new policy ideas are transmitted across governmental units over time; a government™s policy choices are influenced by the policy choices made in other governmental units; an external force Berry and Berry 1990; Boushey 2010; Gilardi 201 6; Graham, Shipan, and Volden 2 013; Gray 1973; Karch 2007 a, 2007b ; Rogers 1962; Savage 1985; Shipan and Volden 2006; Volden 2006; Walker 1969; policy entrepreneur innovative individuals or groups within and outside the public sector that are the first to pursue a new policy ƒ Author™s definition; also see Kingdon 1984; Mintrom 1997 policy innovation a program or policy which is new to the governmental unit adopting it, no matter how old the program may be or how many other governments have adopted it Examples: fithree -strikesfl habitual offender laws, tax and expenditure limitations, medical marijuana laws, smoking restrictions, seat -belt requirements, auto lemon laws, Berry and Berry 2014; Gray 1973; Rogers 1962; Walker 1969; policy learning learning about the content of a policy, including the problem, goals, instruments, and implementation design of the solution Heclo 1974; May 1992; Rose 1991; Sabatier 1988 political learning learning about how to maneuver within and manipulate the policy process to advance an idea or policy Freeman 2008; Gilardi 2010; May 1992; Nicholson -Crotty and Carley 201 5; Rose 1991 venue diffusion a government™s choice of venue to pursue a policy is influenced by prior venue choices of other governments pursuing the policy Author™s definition venue shopping the process of strategically choosing among the variety of institutional settings where policy change can occur to lobby for an issue, press for a new policy, or maintain the status quo Baumgartner and Jones 1993; Holyoke 2003; Ley and Weber 2015; Lubell 2013; Pralle 2003; Sabatier and Jenkins -Smith 1993; Note: The table reports characteristics for ten important concepts in this research. ƒ This definition deviates slightly from the policy literature for the purpose of explaining the process of venue diffusion. 16 that the accompanying definitions elucidate the myriad, sometimes overlapping, terms used to articulate key actors and components o f the dynamic policy process. Second, to clarify, policy innovations are not policy inventions. This distinction is subtle but important. Policy inventions are the products of fithe process through which original policy ideas are conceivedfl (Berry and Berry 2014). Policy innovations, however, are the reincarnations of the invention in the governmental jurisdictions that have yet to enact the fresh idea. Each new policy starts as an invention but becomes an innovation for those who have yet to adopt it. Policy inventions turn into innovations. As Karch (2007 a: 1) put it, innovations fineed not be new in an objective sense,fl rather fithey need only be perceived as new by an individual or another unit of adoption. If the idea, practice or object seems new to a potential adopter, it is an innovation.fl The proliferation of diffusion research has nearly uniformly focused on innovations, ignoring the policy invention process. Save for Polsby™s (1985) work to document policy invention at the federal level and Bou shey and Knight -Finley™s (2016) more recent effort to analyze the policy winnowing of ideas as the design stage, we have little understanding of how ideas are molded into policies that then transfer across governments. Like most policy diffusion scholars, my research here focuses on the spread of ideas and programs that are new to the adopter, but not necessarily new. Thus, policy invention is outside of the scope of my research here, but it is another essential element of the diffusion process that merits more considerable scholarly attention. Fourth, policy diffusion scholarship has the liability of conflating terminology around who is diffusing what. Since the unit of analysis is generally the state -year, country -year, city -year, or an equivalent space -ti me monadic unit, researchers often talk about the government taking action, adopting a policy, diffusing an innovation, or the like: e.g., states copying other states, cities influencing other cities, nations learning from other nations. The reality, howev er, is that governments do not act, do not adopt, do not learn, nor do they diffuse. Rather actors within 17 governments carry out these tasks. While I also rely on this convention Šreferring to fistatesfl taking direct action, for consistency with the literatur e, readers should know that policy advocates are the real protagonists. Finally, the concept of venue diffusion w as formulated with the American state context in mind. As such, unless otherwise specified, when I refer to governments, the implication is U.S. state governments. That said, the theory of venue diffusion should be generalizable to other levels of government where there is variation in the opportunity to venue shop. Other countries with federated systems (e.g., Canada, India, Switzerland) may witness and be prone to similar dynamics. Cities may emulate other cities™ decisions to pursue a policy via the city council or a ballot question. Policy actors in nation states may also learn from and follow the choice of venue in peer nations. This resea rch stems from the American context, but many of its findings and implications should be transferable to additional settings . 18 CHAPTER 2: POLICY DIFFUSION Policy choices within a governmental jurisdiction can be a result of both internal and external forces. Different economic, social, political, institutional contexts, as well as diverse policy actors within a state (i.e., internal forces), can stimulate ne w policy ideas. In fact, states experiencing similar conditions and a common problem may arrive at the same policy solution to address the issue , independently of one another. States may happen upon, or ficonverge,fl on equivalent policy solutions (Bennett 1 991; Boehmke 2009 b; Freeman 2008). Moreover, some problems are unique and isolated to that jurisdiction; in turn, a solution likely emerges from within (Volden, Ting, and Carpenter 2008). External forces, however, can also supply policy choices. States f acing a common problem or issue may look to fipeerfl jurisdictions Šthose units that are geographically proximate (Berry and Berry 1990; Berry and Baybeck 2005; Walker 1969); similar along economic, social, political, cultural, or institutional dimensions (Bu tler et al. 2015; Desmarais, Harden, and Boehmke 2015; F ay and Wenger 2015; Lewis 2011; Lupia et al. 2010; Volden 2006, 2015); or with similar preferences Šfor an innovative solution. Or activity at lower or higher levels of government may also influence or compel states to consider a set of policy options (Karch 2009, 2012; Shipan and Volden 2006, 2008; Welch and Thompson 1980). These external forces explain the phenomenon of policy diffusion : the process by which new policy ideas are transmitted across governmental units over time (Rogers 1962; Walker 1969; Gray 1973). The bedrock of policy diffusion is that a government™s policy decisions are fisystematically conditioned by prior policy choi ces madefl by other governments (Simmons, Dobbin, and Garrett 2006: 787). Or put more plainly by Virginia Gray, diffusion rests on the idea that policy fi adopters influence those in the social system who have not yet adoptedfl (1973: 1176). Since policy actor s do 19 not live in a vacuum, previous actions in one state likely affect subsequent actions in other states. Governments are linked together through their policy decisions. Scholars have documented the transmission of various policy innovations across sp ace and time, ranging from anti -money laundering protections (Sharman 2008); education reform s (Mintrom and Vergari 1998); capital punishment laws (Mooney and Lee 1999); curtailment of welfare benefits (Volden 2002); child seat belt and lemon -aid laws (Sav age 1985); environmental policies (Daley and Garand 2005); health insurance programs (Volden 2006) ; the liberalization of global economic policies (Simmons and Elkins 2004) ; lotteries and gaming ( Berry and Berry 1990; Baybeck, Berry, and Siegel 2011) ; pens ion reform (Weyland 2005, 200 6); social security programs (Collier and Messick 1975); structural changes in cities (Frederickson, Johnson, and Wood 2004); to water fluoridation (Crain 1966), among other policies. Furthermore, researchers have shown that di ffusion can both be horizontal and vertical (Mintrom 1997; Shipan and Volden 2006), across a multitude of dyadic relationships, including diffusion from cities to cities (Crain 1966; Frederickson, Johnson, and Wood 2004), cities to states (Shipan and Volde n 2006, 2008), states to federal governments (Boeckelman 1992; Karch and Rosenthal 2016), and countries to countries (Elkink 2011; Pitlik 2007; Simmons and Elkins 2004 ; Weyland 2005, 200 6). There is no shortage of evidence for the diffusion of myriad publi c policies across myriad jurisdictions and myriad points in time ( Graham, Shipan, and Volden 2013 ). Policy diffusion is generally viewed as an incremental -learning process. Succinctly stated by Volden (2015:3): fipolicy diffusion is not just the adoption of similar policies by similar states but rather–a learning -process leading to more effective policies over time.fl Policy actors within a state facing existing economic or social problems engage in a limited search of potential policy solutions attempted i n peer jurisdictions. Elected officials, in particular, face time constraints and are motivated by concerns for reelection. In turn, policymakers rely on heuristics and information from 20 fitrustedfl sources, learning about these innovations, evaluating their outcomes, and picking the ‚best™ available option they believe will meet constituents™ demands (Berry and Berry 2014; Boushey 2010; Freeman 2008). As Lubell (2013: 545) explains: fi[H]umans heavily engage in social learning from others, sometimes conforming to the behavior of the majority, and other times adopting the behavior of the most successful or prestigious individuals. Such social learning influences how individuals make decisions across different [policy] games in which they participate, and learn over time about different ways of solving collective action problems.fl In a nutshell , policy actors learn about and adopt new developments employed in other governmental units. Much like new technologies gain a small following of early -adopters, then an i ncreasingly larger majority of backers, bookended by laggards joining the bandwagon, new policy solutions also attract early to late adherents. As more states adopt an innovation, more information is available to reduce uncertainty for the holdouts. The cu mulative frequency distribution for the adoption rate of policy innovations typically follows a sinusoidal -shaped curve (Gray 1973; Rogers 1962). The initial pace of policy adoption starts slowly, then gains speed as more states adopt the policy until a widespread majority of states follow the trend, and finally plateaus as the straggler states enact the innovation over a more extended period (Gray 1973; Rogers 1962). Of course, the incremental -learning narrative of policy diffusion does not always compor t with reality. While some policies are adopted and implemented gradually across states, other policies are enacted suddenly across a swath of states, implying imitation rather than cumulative learning (Boushey 2010, Nicholson -Crotty 2009). Some ideas seem to experience an fioutbreak,fl whereby fia positive feedback cycle [leads] to the extremely rapid adoption of policy innovation across statesfl (Boushey 2010: 5). Sudden changes in public opinion, national crises or focusing events, or the advent of policy fa ds can all spur multiple states to adopt nearly identical policies simultaneously . Lawmakers in these states, trying to capitalize on potential electoral benefits of quick action, may 21 fiforgo the gathering of information in favor of immediate adoption, crea ting a rapid diffusion processfl (Nicholson -Crotty 2009: 194). These environments may produce non -incremental patterns of policy transfer (Boushey 2010). Acute innovation may be more common for some types of policies than others, distinct policy actors, or even different institutional venues where policies may be pursued (Boushey 2010; Makse and Volden 2011; Nicholson -Crotty 2009). Importantly, whether a result of incremental learning or imitation, policy diffusion is a fimulti -stage processfl (Elkins and Si mmons 2005; Givan et al. 2010; Karch 2007 b; Karch and Cravens 2014). Unfortunately, much of the early literature solely focused on the dichotomous enactment of an innovation: did ‚State A™ adopt ‚Policy X™ or not? Yet policy innovations do not simply succe ed or fail. New ideas make it onto the agenda, are pursued via one or multiple venues, are enacted or rejected. If the innovations are adopted, they are then implemented and evaluated and can be modified, reinvented, or repealed. Moreover, this feedback fr om the initial policy can influence the process for future innovations. Treating policy diffusion as anything less than a multi -dimensional process fimay underestimate the impact of certain forces while overestimating the impact of othersfl (Karch 2007 b: 26) . Fortunately, more recent scholarly attention, although still limited, has been paid to the various stages of policy diffusion. Many of these works have primarily focused on the later stages of the policy process, including the modification (Karch and Cr avens 2014), reinvention (Clark 1985; Glick 1992; Glick and Hays 1991; Hays 1996; Mooney and Lee 1999), or repeal (Lowry 2005) of the innovation. Only a few scholars have emphasized the earlier stages of the policymaking process. Wilkerson, Smith, and Stra mp (2015), for example, highlight an essential part of the agenda -setting phase by analyzing the reuse of legislative text across multiple state legislatures. Similarly, Boushey and Knight -Finley (2016) explore the possible diffusion of the policy winnowin g that occurs at the design stage. And Gilardi, Shipan, and Wueest (201 9) examine how the framing and perception of 22 the innovation might affect its diffusion. They find that frames which emphasize the concrete aspects of the policy (i.e., learning about the policy) are more predictive of states adopting smoking restrictions than normative frames. Giving attention to all the stages of the policy process will offer a more thorough view of h ow policies move from one governmental unit to another. Mechanisms of Policy Diffusion The task of policy diffusion scholars has been to distinguish between the internal and external forces influencing policy choice within a governmental unit. 6 Careless t heory and empirical analysis can reinforce fiGalton™s problem.fl It occurs when a researcher infers incorrectly that just because two or more characteristics are highly correlated that they are also causally related. But, as the maxim goes, ficorrelation does not imply causation.fl Just because two states that shar e similar traits and fac e similar circumstances end up adopting the same policy does not mean external forces are at play. As such, policy transfer scholars, relying on the fundamentals of the process described above, have identified several key mechanisms to theoretically and empirically account for possible external forces driving diffusion. Rather than merely being interested in if policies diffuse, scholars have increasingly become interested in why they might diffuse. The four main mechanisms identified are learning, imitation, competition, and coercion (Gilardi 201 6; Graham, Shipan, and Volden 2013; Shipan and Volden 2008). 7 6 Teodoro (2009), examining the role of policy entrepreneurs within local bureaucratic agencies, offers a unique perspective on the diffusion of innovations. Instead of treating diffusion as a product of internal and external forces, Teodoro suggests diffus ion is an artifact of supply and demand dynamics. Teodoro argues that forces outside of government Še.g., interest group lobbying, public ideology, fiscal conditions, economic competition, etc. Šdemand new solutions and some individuals within government, ac ting as policy entrepreneurs, are motivated to innovate and supply those new solutions. Seeing policy diffusion as the resultant outcome of supply and demand dynamics puts the individuals making policy decisions, rather than the governmental units where de cisions are made, front -and -center as the unit of analysis. Future diffusion scholarship should incorporate and build on this perspective. 7 Of course, other mechanisms (and the terminology for those mechanisms) for policy diffusion have been identified (Gilardi 2016; Graham, Shipan, and Volden 2013). The international relations literature, for example, has also put forward finorm diffusion,fl whereby normative judgments of a policy may trump rational considerations of its usefulness and effectiveness. It i s the idea that states adopt a policy (e.g., woman™s suffrage) because others are doing it; rather than learning from the experience or other adopters, states enact the policy because of a bandwagon effect (Berry and Berry 23 As articulated above, po licy actors facing a particular problem within on e state may look to policy entrepreneurs in other states for innovative solutions. These actors may learn about the policy (e.g., policy implementation, policy effect) and the political (e.g., venue selection, framing, policy opponents, etc.) dimensions of the innovation (Freeman 2008; May 1992; Seljan and Weller 2011; Shipan and Volden 2008; Volden 2006). 8 Actors search for solutions elsewhere because of a genuine need for information in the face of considerable uncertainty. The policy and political infor mation about the innovation become more abundant and accessible as more states adopt the policy ( Makse and Volden 2011). Rather than filearningfl about a policy, however, some states may imitate the policies adopted by states with similar political, demogra phic, budgetary, or cultural characteristics (Shipan and Volden 2008). For example, some states may look to copy policies adopted by their contiguous -geographic neighbors (Berry and Berry 1990; Cohen -Vogel and Ingle. 2007; Foster 1978; Walker 1969). Althou gh learning and imitation appear similar and can be challenging to parse empirically, learning is a purposive search for information while imitation is conformity (Meseguer 2005). The competition mechanism implies that a state adopts a policy to gain or k eep an economic, resource, or image advantage over other states or other governments. For example, states may permit a lottery or casino to draw in revenue from other states or keep gambling dollars within the state (Baybeck, Berry, and Siegel 2011; Berry and Berry 1990), or firace -to -the -bottomfl in offering the least amount of redistributive benefits (Bailey and Rom 2004; Volden 2002). Finally, diffusion may also occur because another jurisdiction, such as the federal government, influences or ficoercesfl a particular policy innovation through incentives, penalties, or court rulings (e.g., increase the legal drinking age, adopt stem -cell related legislation) (Karch 2009, 2014; Braun and Gilardi 2006; Fin nemore and Sikkingk 1998). Other scholars see policy diffusion as a continuum from lesson -drawing to coercive transfer (Dolowitz and Marsh 2000). 8 For an excellent overview of the concept of learning in the public policy process, see Freeman (200 8). 24 2012; Shipan and Volden 2006, 2008; Welch and Thompson 1980). Bottom -up or top -down press ures can add a vertical component to traditional horizontal diffusion patterns. Indeed, federal intervention, action, or even greater national attention may speed the diffusion of a policy (Allen, Pettus, and Haider -Markel 2004; Boushey 2010, 2012; Mallins on 201 6; McCann, Shipan, and Volden 2015; Nicholson -Crotty 2009; Welch and Thompson 1980). Of course, various policy actors, including networks of state or national interest and advocacy groups can also play a role in the dynamics of policy diffusion (Ball a 2001; Garrettt and Jansa 2015; Mintrom 1997; Mintrom and Vergari 1998; Shipan and Volden 2006). Additional Factors Important to Policy Diffusion Beyond laying the theoretical foundation for the diffusion of new ideas across governmental jurisdictions, providing empirical evidence that other governments™ prior decisions influence a government ™s policy choices , and identifying the key reasons why som e states copy other states, diffusion scholars have also highlighted additional factors that augment, hamper, or inhibit the contagion of innovative policy solutions. These factors include policy type, policy attributes, the capacity of policy actors, and the state™s political environment, institutional considerations, and resource conditions, among others. The diffusion process is not uniform across all types of policies. Shifting from the traditional fistate -centricfl focus of much of the research to a fipol icy -centricfl approach helps to explain variation in patterns of policy transfer (Mallinson 201 6; Makse and Volden 2011; Nicholson -Crotty 2009). Depending on the policy category Šmorality, regulatory, development, redistributive, etc. Šthe pattern, speed, and determinants of diffusion may differ in important ways. For example, Mooney and Lee (1999) demonstrate that the spread of the death penalty, a morality policy, across U.S. states was rapid, largely driven by value judgments and public opinion about capita l punishment 25 rather than any lesson drawing by policymakers. According to the authors, fit he decisionmaking process [was] not one of increment al learning but rather it [was] on e of competit ion to validate majority valuesfl (Mooney and Lee 1999). Nicholson -Crotty (2009) point out that energy, environmental, healthcare, tax, trade, and regulatory policies generally diffuse at a slower pace than other policy types. The dynamics of diffusion are also different for policy reversals , the undoing of past policy (Eyestone 1977; Lowry 2005). Examining the spread of water management efforts, Lowry™s (2005) research indicates that the politics of policy reversals are di fferent from the politics of policy adoptions in at least three ways. First, Lowry finds that the p attern of diffusion for policy annulment is not necessarily geographic, especially since policy repeals gain national attention. Second, the speed of reversals appears to be more gradual than for adoptions; this could be attributable to the fact that polic y disinnovations must compete with established institutions, constituents, and a subsystem of support that must be overcome. Third, at least in the context of the removal of dams, Lowry finds that a state™s fiscal health and interest group pressures are th e leading determinants for the diffusion of reversals. In fact, different types of policies diffuse in dissimilar ways primarily because of their particular attributes. Eshbaugh (2006) and Nicholson -Crotty (2009) identify two features of a policy that affect its diffusion: its salience and complexity. Salient policies are those proposals that affect constituents in important ways and gain the attention of a l arge share of the American public. More salient policies may spread more quickly since they boast greater public awareness and can earn a more prominent spot on the agenda. Complex policies are those programs that require substantial technical expertise to design and address a policy problem, expertise often beyond the capacity of state legislators (Gormely 1986; Nicholson -Crotty 2009). Complexity may hamper the spread of a policy since greater attention and knowledge is required to formulate the policy. Ni cholson -Crotty™s 26 analysis of the effect of these two policy attributes on the speed of diffusion of 57 policies suggests that salience can hasten policy transfer, especially for more straightforward policies. Makse and Volden (2011) also examine the effect of policy attributes on patterns of diffusion. Relying on the attribute typology introduced by Rogers (1983), Makse and Volden analyze how a policy™s relative advantage (i.e., the perceived return of going with the new policy vs. remain ing with the status quo), compatibility (whether policy comports with current values), observability (if policy results recognized by others), trialability (amount policy can be experimented with), and complexity (difficulty in understanding and using the policy) condition its diffusion. They find that all of these policy attributes (scored by outside policy experts) affect the degree and rate of states emulating other states. The rate of adoption was higher for policies with a relative advantage over the s tatus quo, compatible with current values, observable by many, and trialable. Similar to Nicholson -Crotty™s (2009) findings, Makse and Volden conclude that the more complex a policy the slower its rate of adoption. Also of note, policy learning was most ev ident for highly observable policies, while least evident for complex solutions. In addition to the policy type and characteristics, there are also policy -actor, political, institutional, and resource dynamics, among others, that may affect the ado ption of policy innovations. The capacity of the policy actors is critical for the diffusion of any new idea. Contemporary research has documented that more professionalized legislatures are better equipped to innovate and seek out innovations (Shipan and Volden 2006, 2014; Volden 2015). Professionalized policymakers, compared to amateur or part -time legislators, have the time, resources, and motivation to address constituents™ demands for new policies (Shipan and Volden 2006). Those policymakers with highe r policy and political expertise can learn from their own experiences as well as look to other states for successful policies and processes of adoption. 27 Highlighting the significance of capacity, Shipan and Volden (2014) examine states™ adoption of polici es to limit youth access to cigarettes and show that states look to peers that demonstrate success, measured in this case as the largest reduction in teen smoking. But, this learning is conditional on policymakers in early -adopter to laggard states having both the policy (e.g., understanding state conditions, past policy successes and failures) and political (e.g., navigating political obstacles and institutions) capacity. As Shipan and Volden (2014) put it, learning firequires the time and ability both to g ather relevant information and to process it in a way that is appropriate and meaningfulfl (2). Diffusion is more likely when potential adopters are capable of both policy and political learning. The political context also impacts the pace and pattern of p olicy diffusion. For example, adopters™ ideological predispositions can play a central role in learning among governmen ts (Butler et al. 2015; Desmarais, Harden, and Boehmke 2015; Volden 2015). Butler et al. (2015) provide experimental evidence that decisi on makers within a state engage in ideologically motivated reasoning. Legislators predisposed against a particular policy are less likely to learn from policy actors in other states. This ideological hurdle can be overcome if the policy is especially succe ssful or co -partisan peers in other states adopt the policy. The degree of electoral competition within a state (Barrilleaux, Holbrook, and Langer 2002; Holbrook and Van Dunk 1993), state electoral cycles (Berry and Berry 1990, 1992; Mintrom and Vergari 19 98), citizen ideology and public opinion (Erikson, Wright and McIver 1993; Pacheco 2012; Wright, Erikson, and McIver 1987), national attention (Boushey 2016; McCann, Shipan, and Volden 2015), as well as interest group capacity and pressure (Desmarais, Hard en, and Boehmke 2015; Haider -Markel 1998; Mintrom 1997; Sabatier and Jenkins -Smith 1999; Savage 1985) can all condition the spread of new solutions to common societal problems. Some research even suggests that Democratically -controlled state governments ar e more 28 likely to innovate than Republican -controlled states given the party™s penchant for greater governmental intervention ( Calvert , McCubbins, and Weingast 1989 ). Institutional structure and resource conditions also influence whether and how a state c opies another state. Procedural variations in how states adopt policies or amend their constitutions, for example, can slow or speed the spread of policies ( Dinan 2018; Fay and Wenger 2015; Lewis 2011; Lupia et al. 2010). Or a state™s regulatory environment can hurry or hamper diffusion (Stream 1999). States with more resources, including higher wealth or better fiscal health, can also be more innovative (Berry and Berry 1992; Boehmke and Skinner 2012; Desmarais, Har den, and Boehmke 2015; Walker 1969). In sum, depending on the policy type or policy attributes, a state™s policy actor capacity, and political, institutional, or resource environment may contribute to its propensity to innovate (Boehmke and Skinner 2012; W alker 1969). Many of these factors serve as the fiprerequisitesfl for policy diffusion (Savage 1985). And occasionally the fitime comesfl for a policy to be enacted (Savage 1985). Conclusion The policy diffusion literature is replete with rich research on h ow and why innovative ideas spread across different governmental jurisdictions. Despite this voluminous body of research, omissions and opportunities remain to further flesh out (1) the forces driving diffusion in institutional arenas outside of the legisl ative context; (2) the reality of policy diffusion by including innovations that have yet to be adopted by others or have only spread narrowly; (3) the inter -venue dynamics at play when policies are enacted via multiple institutional paths; and (4) whether political learning facilitates the spread of policy solutions and venue choice across U.S. states. I now turn to tackl e the first two items in the next chapter. 29 CHAPTER 3: DIFFUSION DYNAMICS OF BALLOT MEASURES More than f ive decades of research e xploring policy making in the U.S. states offer substantial evidence that a government™s policy choices depend, at least in part, on the policy decisions previously made by other governments Šthat policy innovations diffuse (Berry and Berry 1990; Boushey 20 10; Gilardi 201 6; Graham, Shipan, and Volden 2013; Gray 1973; Karch 2007 a, 2007b ; Shipan and Volden 2006; Volden 2006; Walker 1969). Scholars have assiduously documented if and how a multitude of different policies representing contrasting topic areas have spread across distinct governmental jurisdictions (e.g., cities, states, countries). But existing policy diffusion research has primarily traced the spread of policies from one legislative unit to another legislative unit. Save for a few dozen articles, the literature has mostly overlooked the diffusion of policies in other venues : e.g., ballot measures, gubernatorial executive orders, court rulings, agency decisions. This is unfortunate because state actors have increasingly pursued policy change via alt ernat e venues outside of the legislative context. And the patterns and explanations for policy diffusion from legislature to legislature may not hold for diffusion in other sites . Furthermore, scholars have almost exclusively studied policy diffusion by ob serving policies that have been already widely adopted by a large number of governmental jurisdictions Ša fipro -innovation biasfl (Karch et al. 2016). This selection bias has thus omitted critical data on policies that have failed to diffuse or only been enacted by a small number of governments. By ignoring other sites where policy change can occur and by only modeling policies that have knowingly spread , our understanding of the policy interdependence between governmenta l units is potentially limited. In this chapter, I attempt to address both gaps in the literature by relying on a random sample of ballot measures from the full set of nearly 7,800 ballot measures Šlegislative referendums, citizen initiatives, popular refer endums, and other sŠpursued at the U.S. state level from 1902 Œ2016 (Jordan and Grossmann 201 8; NCSL 2016). Importantly, this supply of ballot measures includes 30 initiatives and referendums that have and have not diffused . Leveraging this unique data, I exami ne three empirical questions. First, do state ballot measures diffuse across U.S. states? Relatedly, if not, why not? Second, if ballot measures d o spread across governmental jurisdictions, what are the primary external mechanisms driving this diffusion? Third, how has our past selection bias (i.e., modeling only policies that spread) affected our understanding of the diffusion process? Although there are some limitations in my ability to answer those questions, this chapter does provide a clearer picture of the states™ use of ballot measures, their relative success rates, and the main topic areas that voters are asked to decide . Moreover, I offer evidence that we, as policy scholars, have been overstating the occurrence of policy diffusion. Nearly half of all ballot measures are only adopted by one state and do not appear to spread to other jurisdictions. And nearly three -quarters of ballot measures are pursued by fewer than a handful of U.S. states. Besides , by excluding the policies that have only spread to a limited number of governments or have yet to diffuse entirely, our models run the risk of inflating key mechanisms™ impact on the diffusion dynamics. Still, for the ballot measures that do diffuse, the axial forces found to drive diffusion in the leg islative context Špolicy learning and imitation Šmake a cent er stage appearance in the ballot measure context as well. Policy Diffusion Policy choices within a state can be a result of both internal and external forces. A state™s given political, economic, social, and institutional context (i.e., internal forces) may stimulate new policy ideas. Sometimes, states face unique problems isolated to their jurisdiction; in turn, a policy solution likely emerges from within (Volden, Ting, and Carpenter 2008). Othe r times, states experiencing similar conditions and a common problem may ficonvergefl upon equivalent policy solutions independently of one another (Bennett 1991; Boehmke 2009 b; Freeman 2008). 31 External forces, however, can also supply policy choices. States facing a common problem or issue may turn to fipeer jurisdictions flŠthose states that are geographically proximate (Berry and Berry 1990; Berry and Baybeck 2005), or similar along economic, social, political, cultural, or institutional dimensions (Butler et al. 2015; Desmarais, Harden, and Boehmke 2015; Lupia et al. 2010; Volden 2006, 2015) Šfor an innovative solution. States™ policy options can also be influenced by activity at the federal or local levels of government (Karch 2009, 2012; Shipan and Volden 2006, 2008; Welch and Thompson 1980). These external forces explain the phenomenon of policy diffusion: the process by whic h new policy ideas are transmitted across space and time (Rogers 1962; Walker 1969; Gray 1973). Scholars have documented the spread of various policy innovations across different units and over time, ranging from education reforms (Mintrom and Vergari 1998 ), capital punishment laws (Mooney and Lee 1999), curtailment of welfare benefits (Volden 2002), to child seat belt and lemon -aid laws (Savage 1985), health insurance programs (Volden 2006), and lotteries and gaming (Berry and Berry 1990; Baybeck, Berry, a nd Siegel 2011). Furthermore, researchers have shown that diffusion can both be horizontal and vertical (Mintrom 1997; Shipan and Volden 2006), across a multitude of governmental jurisdictions, including diffusion from cities to cities (Frederickson, Johns on, and Wood 2004), cities to states (Shipan and Volden 2006, 2008), and states to federal governments (Karch and Rosenthal 2016). The literature has also theoretically developed and empirically identified several key mechanisms that explain why policy inn ovations spread. Although policy diffusion is generally thought of as an incremental learning process, that is not always the case. While some policies are enacted in a gradual fashion across states, other policies are adopted suddenly by a large swath of states, implying imitation rather than cumulative learning (Boushey 2010, Nicholson -Crotty 2009). In addition to learning from or imitating their peers, states may try to compete with or gain an 32 advantage over other states or feel coerced with inc entives or penalties to adopt a policy (Gilardi 2016; Graham, Shipan, and Volden 2013; Shipan and Volden 2008). The Myopic Focus on Legislative Arena Despite hundreds of articles detailing if and how policies diffuse (Graham, Shipan, and Volden 2013), the overwhelming bulk of the literature has emphasized the transfer of policies from one legislative unit to another legislative unit (e.g., state legislature to state legislature, city council to city council, city council to state legislature). As detailed in Chapter 2, the myopic focus on the legislative arena is surprising for at least three reasons. First, early scholars suggested that policy innovation occurs in various venues outside of the fipeople™s branchfl (Polsby 1985; Walker 1969). Althou gh Walker™s (1969) canonical piece focused on the diffusion of ideas from one legislative body to another, he explicitly acknowledged that innovations are pursued fiby regulatory commissions or courtsfl (881). And Polsby (1985) asserted that fithere is no dou bt that political innovations take place within...diverse [institutional] arenas.fl The second reason why it is unexpected that few studies investigate diffusion outside of the legislative context is that policy change can and does occur in multiple venues (e.g., citizen initiatives, legislative referendums, gubernatorial executive orders, court rulings, agency decisions). This is one of the defining advantages of American federalism. Policies can be enacted via citizen ballot initiatives or referenda, legis lative referenda, state supreme court decisions, gubernatorial executive orders, bureaucratic agency decisions, and even by way of the federal government. Although numerous horizontal and vertical venues suggest multiple veto points to impede change, they also represent various opportunities to pursue change. Third , this singular focus is remarkable especially since policy actors are increasingly turning to alternative venues to press for new ideas ( Ferraiolo 2008 ; Magleby 1988; Miller 2009; Reilly 2010). 33 There is strong evidence that other forums are increasingly being used to set the policy agenda (Damore, Bowler, and Nicholson 2012). Indeed, due to heightened polarization and gridlock at the federal level, interes t groups and citizens have sought policy windows at the subnational level (Dinan and Krane 2006). To be sure, there are a couple dozen studies that have directly or indirectly explored policy diffusion in venues beyond American legislative bodies. For exa mple, Fay and Wenger (2015) and Lupia et al. (2010) highlight how states ™ institutional hurdles slow the adoption of constitutional amendment s for lottery policies and anti -gay marriage bans. Both conclude that higher institutional barriers to amending sta te constitutions slow policy diffusion, even if the public favors the policy. Lewis (2011 , 2013 ) also explores the diffusion of same -sex marriage bans and finds that those states equipped with the capacity for citizen -driven initiatives were more likely to outlaw gay marriage than states without direct democracy. Analyzing the emulation of capital punishment and Indian gaming policies, Boehmke (2005) offers some support for the notion that interest groups within a direct -democracy state look to interest gro ups in other direct -democracy states for policy ideas. Similarly, Seljan and Weller (2011) recount the diffusion of tax and expenditure limits (TELs) via direct - and non-direct democracy states. They find that policy failure in geographically proximate sta tes caused both plebiscite and non -plebiscite states to be less likely to adopt TELs. 9 However, the few articles that examine the spread of policy ideas outside of the legislative context are the exception rather than the rule. This is regrettable because the diffusion of policy innovations at the ballot box, by governors, in courtrooms, or via state agencies may not parallel the patterns of diffusion in legislative arenas. Just as the diffusion process is not uniform across all types of policies, neither should we expect it to be uniform across all venues. Moreover, the previously 9 A few other scholars have also documented how the courts (Caldeira 1985; Canon and Baum 1981; Dear and Jessen 2007; Glick 1992 ; Hinkle 2015; Hinkle and Nelson 2016 ), gubernatorial offices (Bowman, Woods, and Stark 2010), and bureaucratic agencies (Pa rinandi 2013; Teodoro 2009; Volden 2006) serve as venues for policy innovation. 34 identified mechanisms driving diffusion Šlearning, imitation, competition, coercion Šmay be more appropriate in the legislative context than in other institution al settings with varied institutional arrangements. In fact, new, unidentified mechanisms may be driving force s as well . Besides , the policy actors typically important in fostering diffusion in the legislative context (e.g., legislators, interest groups, c itizens activists) may play a more or less prominent role in the spread of innovations in other forums . Finally, the policy attributes (e.g., salience, complexity, observability, trialability) deemed critical to diffusion in the people™s branch ( Makse and Volden 2011; Nicholson - Crotty 2009) might also wax and wane in importance to this process in other venues. Put simply , the picture of policy diffusion in the legislative context may not reflect the dynamics in other institutional settings. Policy Diffusi on Research™s Selection Bias A myopic focus on one venue, however, is not the only gap in current policy innovation research. Nearly every policy diffusion study tries to examine transfer patterns of policies (usually from one policy domain) that have alre ady been adopted by numerous governments, rather than consider the full set of policies that are (or are not) pursued by governments. Consequently, researchers have selected on the dependent variable ( King, Keohane, and Verba 1994 ), and left a significant amount of essential data, policies not yet enacted by other jurisdictions, out of their models of policy diffusion. Experts have tried to explain the fihitsfl without also accounting for the fimisses.fl This omission is understandable since policy scholars frequently face data and modeling limitations, and because studying both policies that have diffused and have yet to diffuse requires scholars to have a complete dataset Šan arguably onerous if not impossible demand in some circumstances. Ye t, to have a richer and more holistic understanding of the dynamics of policy diffusion, we must model the full set of policies that are fiat riskfl of being adopted by governmental units. 35 Fortunately, recent scholarship has identified and started to address this issue. Karch and colleagues (2016), for example, call out diffusion research for its fipro -innovation bias,fl whereby scholars select policies that have already diffused broadly. In analyzing the adoption of interstate compacts by a handful of states t o a plurality of states, they find that modeling innovations that only gain large traction causes us to underestimate the role of learning and professional associations and overestimate any geographic or regional forces (Karch et al. 2016). In addition, Vo lden (2015) explores whether a state™s abandonment of a specific Temporary Assistance to Needy Families (TANF) requirement makes another state more likely also to abandon or fail to adopt that specific requirement as well. He offers firm support that it do es, suggesting that policy reversals and abandonments may also diffuse. Volden also finds that commensurate levels of professionalism and ideological similarity between the states facilitate this learning and desertion of the policy. However, because this is a salient, complex, and politically contentious policy area, we are still left wondering how generalizable these findings are to other policy topics, venues, and periods. But my argument here goes beyond the need to account for fipro -innovation biasfl and the modeling of policy reversals and abandonment. I charge that to comprehend fully if, why, how, and when policies diffuse, we cannot simply model policies that disseminate narrowly or widely. Instead, we must also consider for policies that have yet to gain traction outside of one state. Much like congressional scholars track and model bills that remain in committee and do not make it to the floor, or international relation scholars account for countries that do and do not go to war, diffusion scholars s hould also account for policies that remain in one domain and do not spread across subnational governments. Leveraging the inclusion of policies that have yet to spread can help us better grasp how policies diffuse in a dynamic, interdependent environment. No article to my knowledge has explored the diffusion of all possible bills, measures, orders, rulings, or decisions. 36 Luckily , the National Conference of State Legislatures Ballot Measure Database (Jordan and Grossmann 201 8; NCSL 2016) provides an opportu nity (1) to explore policy diffusion in key venues outside of the purely legislative context and (2) to include in our models innovative measures that have yet to launch. The database contains the full set of successful and failed ballot measures Šlegislati ve referendum, citizen initiatives, popular referendums, among others Špursued a cross all 50 U.S. states from 1902 Œ2016. Moreover, the ballot measures cover an array of policy areas (from abortion, bonds, and morality policies to government reform, veteran benefits, and tax policies ) to examine the effect of this potentially pernicious selection bias . Ballot Measures Pursued in the U.S. States Citizens acting as lawmakers is one of the unique aspects of American federalism. Buoyed by concerns about machine politics, corruption, and a powerful few supplanting the will of the many, populist and Progressive -Era reformers in the late 1800s and early 1900s were able to push for the adoption of new political institutions across the states: direct election of U.S. senators, Australian secret ballots, civil service standards, and direct democracy (e.g., Bowler and Donovan 2006; Lawrence, Donovan, and Bowler 2009; Smith and Fridkin 2008). Within two decades, twenty states had adopted direct democracy, whereby voters could directly or indirectly initiate policies, repeal legislation, or recall elected officials. 10 There are multiple motivations for why elected officials, interest groups, or citizen activists might pursue a ballot measure. Policy entrepreneurs wi thin a state may turn to alternative venues because they have greater knowledge and experience with one venue over another (Pralle 2003). Or perhaps state actors are facing increasing polarization and gridlock in state legislatures (Hinchliffe 10 Five states adopted direct democracy provisions much later than the turn of the 20 th century. Alaska became a direct democracy state in 1959, followed by Wyoming in 1968, Ill inois in 1970, Florida in 1972, and Mississippi in 1992. 37 and Lee 201 6; Shor and McCarty 2011) and need another roadmap to policy change. Or legislators may capitalize on their partisan legislative majorities to enshrine their policy preferences in the status quo ( Damore, Bowler, and Nicholson 2012 ). Interest groups and advo cates also frequently desire to codify or annul policies in state constitutions (Miller 2009; Fay and Wenger 2015), attempting to preempt other institutions (Boehmke, Osborn, and Schilling 2015; Dumas 2017; Gerber 1996) or for greater popular sovereignty ( Bowler and Glazer 2008; Lewis 2013), among other reason s. Still , actors may turn to the states to try to challenge federal policy or address inaction at the national level (Ferraiolo 2008). Regardless of the motivations to pursue an initiative or referendu m, not all states™ access to ballot measures is created equal. Today, 24 states and the District of Columbia allow their citizens through direct or indirect means to press for new statutory or constitutional language at the ballot box. 11 I classify these me asures as ficitizen initiatives.fl Forty -nine of the states, with the exception of Delaware, allow the legislature to refer constitutional questions to voters, although only 24 states and the District of Columbia permit the electorate to have a say in statut ory questions. I term these referrals as filegislative referenda.fl S lightly more than half of the states, 26 in total, allow citizens to check the legislature by repealing public policy via plebiscite or referendum. I mark these measures as fipopular referen da.fl Finally, some states allow fiother measuresfl for constitutional conventions, nonbinding questions, or advisory votes. See Table 3.1 for each state™s access to these different types of ballot measures , where citizens have an expanded or limited direct say on policymaking. However, even this simple categorization is not exhaustive because there exist varying degrees of hurdles for policy actors trying to pursue an initiative or referendum . To secure a pl ace for a measure on the ballot, most states require a specific number or proportion of voters™ signatures 11 Although, Illinois is one of these twenty -four states, it only permits citizen initiatives to amend Article IV of its constitution dealing with legislative procedures. 38 Table 3.1: Institutional Arrangements for Direct Democracy by State State Legislative Referendum (Statute) Legislative Referendum (Amendment) Citizen Initiative (Statute) Citizen Initiative (Amendment) Popular Referendum Recall Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming TOTALS 24 49 21 18 23 19 Note: The table provides the different institutional arrangements for ballot measures for each state, whereby state electorates hav e a direct vote on policymaking, including via legislative referenda for statutory and constitutional policies, citizen initiatives for statutory or constitutional policies, popular referendum to annul a policy, or the recall of an elected official. Source: Ballotpedia (2016); Lupia et al. (2010) ; NCSL 2016; and Waters (2003). 39 (e.g., some percentage of the votes cast for governor in the prior election) (Donovan, Bowler, and McCuan 2001). Several states also require approval by the legislature or a committee . Still , other states require a supermajority of support from the legislature or approval in two legislative sessions. For example, Tennessee requires a supermajority of legislators supporting the measure from two separate sessions before the proposal makes the ballot (Lupia et al. 2010). Despite thes e institutional hurdles , elected officials, interest groups, and citizen activists have increasingly turned to referendums, initiatives, and other ballot measures to pursue policy change in the U.S. states over the last century ( Ferraiolo 2008; Magleby 198 8; Miller 2009; Reilly 2010). Of the 7,772 ballot measures pursued at the state level from 1902 Œ 2016, nearly two -thirds (64%) have been put before the voters since the 1970s. Of the cumulative total, some 62% (4,814) were legislative referenda, with the overwhelming majority of those occurring in the last five decades. Another 32.1% (2,494) were citizen initiatives, with more than half of those being pursued since the 1970s. And a smaller portion of the total ballot measures were popular referenda (4.2%, or 325) and other measures (1.8%, 139). 12 Figure 3.1 displays the total number of ballot measures pursued by type by decade from 1900 Œ2010 (although the 2010 decade only includes ballot measures through 2016). As we can see, there has been a marked increase in the use of legislative referendum and citizen initiatives to bring about policy change since the 1970s. This activity appears to have reached an apex in the 1990s, gran ted that the number of legislative referendums and citizen initiatives have held rel atively steady since. Opposite of this trend, the use of popular referendums to recall legislation has drastically declined since the 1910s, with 70 such measures pursued that decade compared to only 29 in the 12 For an additional and exce ptional breakdown of the number of ballot measures adopted by the U.S. states since the early 1900s by decade, policy type, and the quantity challenged in the courts see Miller (2009). 40 Figure 3.1: Total Ballot Measures by Type by Decade, 1900 Œ2010 Note: Bar chart displays the total number of ballot measures pursued across the U.S. states by type (citizen initiatives, leg islative referendums, popular referendums, and other ballot measures) pursued in the U.S. states by decade fr om 1900 Œ2010. The 2010 decade only includes ballot measures through 2016. Source: National Conference of State Legislatures (NCSL). 2016. Ballot Measures D atabase. 2010 s. There has been an uptick in fiotherfl measures as well. This category includes measur es for constitutional conventions, nonbinding questions, or advisory votes. 41 State policymakers have been asking for voter input more frequently , while citizens and interest groups have been more willing to give legislators a pass (by not repealing legislat ion via popular referenda) and press for new policies via plebiscite . Aside from an increase in the absolute numbers, however, the passage rate of ballot measures also appears to be higher compared to the earlier decades. See Figure 3.2 for the success rate of ballot measures by type by decade. When legislators refer policies to the electorate via legislative referendum, the success rate of those referenda ha s increased over time. In 1910, less than half of legislative referendum were approved; i n 2010, nearly 80% of legislative referend a passed . This difference in proportions is statistically reliable Figure 3.2: Ballot Measure Passage Rate by Ballot Measure Type and Decade Note: Line chart displays the passage rate for Citizen Initiatives, Legislative Referenda, Popular Referenda, and other ballot measures pursued in the U.S. states by decade. Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. 42 There has also been a slight increase in the success rate of citizen initiatives (35% in 1910, 50% in 2010) but not as steep as has been for referred measures. This passage rate for legislative referendum is somewhat surprising considering that prior research ( Dumas 2017; Gerber 1996) suggests that legislators desire to directly tackle policy issues to achieve outcomes closer to their preferences, rather than allow public input via legislative referendum , citi zen initiatives , or court cases .13 The s uccess rate also varies widely by state. Figure 3.3 illustrate s the average pass rate for ballot measures by state. South Dakota has the lowest average success rate at under 40%, closely followed by Michigan, Colorado , and New Hampshire. Eleven states have a success rate below 50%. Meanwhile, Indiana, North Carolina, Pennsylvania, Tennessee, and Washington DC had the highest success rate at 100%. However, these latter states only put a limited number of legislative ref erenda before voters in the last century: Indiana (11), North Carolina (19), Pennsylvania (12), Tennessee (11), Washington DC (1). Fourteen states in total have a success rate for measures higher than 80%. Of course, not all states turn to initiatives and referendums as frequently as others. Figure 3.4 shows the aggregate number of measures that each state has allowed on the ballot. Two states, in particular, stand out. California has attempted more than 1,238 ballot measures, while Oregon has pursued 859 i nitiatives and referendums since 1902. The next closest state, Oklahoma, has only attempted 440 in the same period, while the average for the states without California and Oregon included is 229 ballot measures. For these states , excluding California and O regon, this works out to four ballot measures per two -year election cycle. Clearly , California and Oregon are leaders (and likely influential observations in any models of diffusion) in pressing for policy change at the ballot 13 This sizeable increase in the success rate for legislative referendum over time should be explored further in future research. Do legislatures with larger partisan majorities simply capitalize on a more supportive electoral climate to secure their policy prefe rences ( Damore, Bowler, and Nicholson 2012 ). Is this due to legislators profiting from geographic sorting and polarization (Lang and Pearson -Merkowitz 2014 ). Or have legislatures become better at discerning the policy preferences of the public and carefull y craft policy language directed at the median voter? Or are voters simply more likely to approve of a referendum because it came from the legislature rather than an interest group or citizen activist? 43 Figure 3.3: Ballot Measure Passage Rate by State Note: Dot plot displays the average passage rate of ballot measures (e.g., citizen initiatives, legislative referendum, popular r eferendum, others) attempted by each U.S. state from 1902 -2016. Source : NCSL (2016 ) Ballot Mea sures Database. box, but there is also sizable variation across the other states. Certainly, the aforementioned hodgepodge of institutional arrangements by state (Bowler and Donovan 2004; Fay and Wenger 2015; Lupia et al. 2010) affects whether or not meas ures even make it on the ballot let alone adopted. Oregon and California have fewer requirements than most states to get a measure on the ballot 44 Figure 3.4: Total Ballot Measures Attempted by U.S. State from 1902 Œ 2016 Note: Dot plot displays the absolute number of ballot measures (e.g., citizen initiatives, legislative referendum, popular refere ndum, others) attempted by each U.S. state from 1902{2016. Source : NCSL (2016 ) Ballot Measures Database. 45 Figure 3.5: Ballot Measures by Frequency of Policy Area Note: Bar chart displays the percent frequency of ballot measures (e.g., citizen initiatives, legislative referendum, popular ref erendum, others) by policy area attempted across the U.S. states from 1902 -2016. Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. 46 (Banducci 1998) . But that is not the only factor. Organized interest groups also play a crucial role, especially in pressing for citizen initiatives (e.g., Boehmke 2005; Damore, Bowler, and Nicholson 2012). Still, polarization (Hicks 2013), larger partisan legislative majorities (Damore, Bowler, and Nicholson 2012), and less competitive elections (McGrath 2011) also likely explain the increased frequency in b allot measures in some states over others. States appear to turn to ballot measures, however, for some policy areas more than others. Figure 3.5 provides the percent frequency of measures by policy area. Despite the media attention around more salient meas ures pertaining to morality policies such as gambling (Berry and Berry 1990), gay marriage (Lewis 2011), the death penalty (Mooney and Lee 1999), or medical marijuana (Hannah and Mallinson 201 8), the top three policy areas for ballot measures actually pert ain to government reforms (17.57%), taxes and exemptions (11.07%), and bonds and budgets (10.18%). 14 Anecdotally, some of the most frequently appearing topics are bonds to raise revenue (e.g., schools, transportation, construction, hospitals, research facil ities), and tax exemptions for different groups or individuals viewed favorably by the public (e.g., non - profits, churches, veterans, farmers). The Diffusion of Ballot Measures: Expectations The above description and figures illustrate the wide variation in access to, frequency of use, and type of ballot measures pursued across the American laboratories of democracy. Considering these differences in states™ institutional arrangements and the use of ballot measures, do states copy the ballot measures adopted in other states? Because past diffusion research has demonstrated the 14 Also see Figure A.1 in the Appendix which shows the frequency of policy area by type of ballot measure. Citizen initiatives were more likely to emphasize government reform (21.4%), taxes and revenue (20.9%), elections and campaigns (12.5%), and business and economic development (12.2%) issues. Legislat ive referendums, however, appear to be pursued to reform government (32.5%), solicit bond and budget funding (24.4%), or make changes to tax policy (18.7%). 47 interdependence between states™ policy choi ces in the legislative context, I believe it is sensible to expect some conditionality in the ballot context as well. However, the degree of interdependence may be overstated in the literature (Karch et al. 2016). A fundamental assumption of policy diffus ion is that jurisdictions face common problems that merit universal policy solutions. E lected officials, interest groups, or concerned citizens are likely to look to other jurisdictions for these universal solutions. Nonetheless, states frequently face pro blems that are unique and that require particularized policy solutions (Volden, Ting, and Carpenter 2008). Not all policy ideas are germane to all governments. This fact has been obscured in the policy diffusion literature as scholars have selected cases t hat have spread to offer support of policy diffusion (Karch et al. 2016). Furthermore, even some universal policy solutions for mutual issues may transfer slowly (over decades or centuries) to other jurisdictions or not diffuse at all (Rogers 1962). 15 Inde ed, achieving policy change is challenging and infrequent (Baumgartner et al. 2009; Baumgartner and Jones 1993; Kingdon 1984). There are abundant institutional and political roadblocks to altering public policy. Even when policy change is achieved, failure in implementation, reform, and policy reversal can all result (May 1992; McConnell 2010; Patashnik 2014; Volden 2015), thus undermining the policy™s potential for diffusion in other jurisdictions (Seljan and Weller 2011; Volden 2015) . As a result, our gen eral expectation should be one of stagnation rather than dispersion . Most innovations should not spread to other jurisdictions. Therefore, I offer the following hypothesis: H1: Minimal Diffusion : Most ballot measure policy ideas pursued in one state will not be pursued by other states. 15 One cannot help but think of the s talled spread of the solution (i.e., vitamin C) to fight scurvy among sailors during the Age of Sail (mid -16 th century to mid -19 th century) It took several hundred years before fresh fruit and vegetables were commonplace on ships (Rogers 1962). 48 Nevertheless, because of past demonstrated interdependence among states, it is reasonable to assume that some policy innovations will transfer to other states. 16 Turning to the potential mechanisms driving th e diffusion of ballot measures across subunits , I consider two external factors from the existing literature: policy learning and imitation. Because policy actors are boundedly -rational, facing limited cognition and resources, they engage in satisficing an d learn from other entrepreneurs and actors about potential solutions to universal problems (Gilardi 2010, 201 6; May 1992; Shipan and Volden 2008). As first movers are successful in adopting a new policy to address a common issue, remaining states will gra dually become familiar with these solutions and are likely follow suit. Therefore, I propose the following policy learning hypothesis, whereby an increase in the number of states enacting the policy innovations augments subsequent states™ likelihood of ado pting the ballot measure. H2: Policy Learning: A state™s likelihood of adopting a given ballot measure increases as the number of other states pursuing that ballot measure increases . Even though I anticipate states deliberately seeking out and learning about available solutions from other states, it is also possible that states may emulate the ballot measures pursued by peer states ( Butler et al. 2015; Shipan and Volden 2006, 2008, 2014; Volden 2006, 2015; Zel izer 2019 ). A desire for homophily may lead policy actors to look to comparable states with a similar partisan composition, citizen ideology, economic circumstances, institutional settings, or demographics. Rather than a comprehensive search for informatio n, state actors may simply take cues from similar states that have adopted a ballot measure. Acknowledging a possible imitation mechanism for diffusion, I hypothesize the following : 16 Although it may be unreasonable to assume the same mechanisms from the legislative arena apply for initiatives and referendums. Pursuing referendums and initiatives usually requires overcoming multiple veto points and players (more than in the legislative context), as we ll as mobilization and campaign efforts. 49 H3: Imitation : A state™s likelihood of adopting a given ballot measure in creases as its similarity with other states adopting that ballot measure increases . U.S. states also operate within a federal system. As a result, the national political environment may make some policies or venues more attractive than others (Baumgartner and Jones 1993; Berry and Berry 1990, 1992; Ley and Weber 2015; Mintrom and Vergari 1998; Smith et al. 2006) . Presidential election s, for example, offer an opportunity for legislators and organized interests to mobilize support for or against a measure . Because the national political environment and timing of elections might influence the likelihood of a ballot measure™s adoption, I p ropose: H4: National Environment : A state™s likelihood of adopting a given ballot measure increases during presidential election years . Aside from the se external factors 17 associated with a state adopting a ballot measure , numerous i nternal factors may h asten or hinder the adoption of a ballot measure. For example, a state™s political environment Šparty control of governing bodies, ideological predisposition of officials, public opinion regarding the policy issue Šare known determinants of policy adoption ( or inaction) (Butler et al. 2015; Calvert et al. 1989; Desmarais, Harden, and Boehmke 2015; Enns and Koch 2013; Erikson, Wright and McIver 1993; Holyoke, Brown, and Henig 2012; Pacheco 2012; Volden 2015; Wright, Erikson, and McIver 1987). Even the degree o f electoral competition in the state could lead to more ballot activity (Barrilleaux, Holbrook, and Langer 2002; Holbrook and Van 17 In addition to the policy learning and imitation mechanisms, there are three other mechanisms that are typically accounted for in policy diffusion studies: geographic, competition, and coercion factors . I decide to not control for a geographic effect because of the sheer number of policies pooled in the dataset and the separate neighbor to neighbor proportions that would be required for each state -year. While future analyses could include a fineighborfl variable, the re are several studies that suggest regional policy diffusion in the U.S. is increasingly rare (Haider -Markel 2001 a; Karch et al. 2016). 17 As such, I do not believe the omission of a geographic variable will alter the inferences we can draw from the current analyses. Regarding the competition and coercion mechanisms, these factors tend to be topic area dependent. For instance, states are more likely to adopt a lottery as they lose out on tax revenue to neighboring states with an existing gambling system (Ber ry and Baybeck 2005; Berry and Berry 1990), or states are more likely to adopt welfare or embryonic stem -cell policies as the federal government signals or incentivizes its preferences (Karch 2006; Karch and Rosenthal 2015). Identifying variables that woul d encompass potential competition or coercion across such a range of policies (from taxes to the environment to veteran affairs) and time is quite daunting, to say the least. I decide to leave the inclusion of such variables for future research with the da ta. 50 Dunk 1993) . Furthermore, measures of state wealth and resources correlate highly with more innovative states (Walker 1969). Va riation in a state™s demographics has also been shown to matter in the spread of new ideas or in pressing for policy change outside of the legislative arena. Policy actors with larger state populations may face higher hurdles to enacting a ballot measure g iven the steeper costs of informing or mobilizing voters (Boehmke 2005; Donovan and Bowler 1998 ; Lewis 2013) . And states located in the southern part of the U.S. are known to behave differently than their peers in the North, Midwest, and West for a host of historical, political, and cultural reasons (Foster 1978; Key 1949). Institutional availability (i.e., whether a state has access to direct or indirect citizen initiative, statutory legislative referendum, popular referendum) ( Füglister 2012; Gilardi and Wasserfallen 2019) as well as the degree of difficulty in pursuing a ballot measure (e.g., amendment to the states constitution) may also influence the diffusion of ballot measures ( Dinan 2018; Fay and Wenger 2015; Lupia et al. 2010; Lutz 199 4). More hurdles to pursue a measure (e.g., ranging from the number of signatures required for the petition to legislative approval in subsequent sessions and a supermajority of support from the voters) may impede the transfer of policies. Policy diffusion research identifies interest groups as central characters in the spread of new ideas across jurisdictions ( Balla 2001; Garret and Jansa 2015; Haider -Markel 200 0, 2001a, 2001b ; Karch 2007 a; Mintrom 1997; Mi ntrom and Vergari 1998; Shipan and Volden 2006; St one 2012 ). Interest group presence may play an even bigger role in direct -democracy states (Boehmke 2005; Gray et al. 2004). Boehmke™s argument , in particular, is that ballot measures provide an additional route for epistemic networks to influence public p olicy. And since the opportunities for accomplishing one™s goals are more plentiful in direct -democracy states, Boehmke suggests this will produce a greater number of interest groups in these states, especially citizen organizations championing and advocat ing on behalf of underrepresented groups. Of course, interest groups™ 51 impact on the process is easier to imagine than to always see in our models (Banfield 1961; Lowery 2013). Lastly, p olicy attributes are also known to influence the adoption as well as th e pace of diffusion of new policies ( Makse and Volden 2011; Nicholson -Crotty 2009). For example, depending on the policy category Šmorality, regulatory, development, re -distributive, etc. Šthe pattern, speed, and determinants of diffusion may differ in impor tant ways. For example, Mooney and Lee (1999) demonstrate that the spread of the death penalty, a morality policy, across U.S. states was rapid, largely driven by value judgments and public opinion about capital punishment rather than any lesson drawing by policymakers. According to the authors, fithe decisionmaking process [was] not one of incremental learning but rather it [was] one of competition to validate majority valuesfl (Mooney and Lee 1999). Nicholson -Crotty (2009) point out that energy, environment al, healthcare, tax, trade, and regulatory policies generally diffuse at a slower pace than other policy types. Therefore, controlling for policy area may further elucidate the underlying processes at play. Data and Methods Data I attempt to explore and test the diffusion processes of ballot measures by relying on the NCSL Ballot Measure Database (Jordan and Grossmann 201 8; NCSL 2016). To check the potential for diffusion for all ballot measures from 1902 Œ2016, however, similar policies must b e identified within the dataset. Intending to match analogous ballot measures across time and space, I drew a random sample of fifty ballot measures (out of the full set of 7,772 initiatives and referendums). For each randomly selected measure, I then sear ched for keywords to find comparable ballot measures pursued by the same or other states from 1902 Œ2016. This matching exercise produced a sample of 579 ballot measures (or 7.4% of the full set of initiatives and referendums) , co mprised of measures 52 paired across states (suggesting po ssible diffusion) and measures that were not matched with other states (indicating no diffusion) . Notably, unlike past policy diffusion studies, this dataset contains both policies that appear to have diffused and have not diffu sed to other states. Modeling both diffusion successes and failures should help us better flesh out the diffusion dynamics of ballot measures. Table 3.2 lists the fifty randomly selected ballot measures, along with the first state to pursue the measure, the first year it was attempted, a description of the measure, the type of measure, and how many states also attempted to enact the policy. For more information on my strategy for matching and coding ballot measures, please see Appendix A . Also, to demons trate that the fifty randomly selected measures mirror the underlying population, please see Figure A.2 and Figure A.3 in the corresponding Appendix. After gathering this sample of 579 ballot measures (by taking the random 50 measures and matching analogo us propositions from the full set) , I constructed the relevant data universe for these measures. Since each state can pursue multiple ballot measures in any given year, I examine each ballot measure choice simultaneously. States are at firiskfl of pursuing o r adopting a particular ballot proposal from the time it was first adopted to the end of the dataset (2016) (as states are still fiat riskfl of adopting a policy pursued decades ago by other states). As a result, the unit of observation is state -measure -year , rather than state -year, as is most common in policy diffusion studies. This approach is a pooled events history analysis (EHA) ( Box -Steffensmeier and Jones 2004 ) and is an established modeling strategy in the literature (Shipan and Volden 2006 , 2008 ; Vol den 2015). EHA is useful because it examines each ballot measure for each state in each year to determine if the state adopted a specific measure in a given year, distinguishing between external (e.g., learning, imitation) and internal (e.g., state resou rces, state politics, state institutions, etc.) 53 Table 3.2: Description of 50 Randomly Selected Ballot Measures Policy ID 1st State to Pursue Yr. Pursued Ballot Measure Description Ballot Measure Type # of States Pursuing 10001 Colorado 2004 Remove obsolete constitutional amendments Leg. Referendum 1 10002 Arizona 1976 Motor vehicle emissions inspections Leg. Referendum 1 10003 California 1996 Allow medical marijuana Initiative 15 10004 Arizona 1972 Regulation around the employment of children Leg. Referendum 1 10005 Oregon 1996 No discrimination against health care providers Initiative 1 10006 California 1980 Bonds for Lake Tahoe conservation and restoration Leg. Referendum 2 10007 Oregon 1990 Use pollution control bonds in OR for related activities Leg. Referendum 1 10008 Vermont 1903 Permit sale of alcohol or liquor at county or state level Leg. Referendum 16 10009 Arkansas 1912 Provide free textbooks for schools and students Initiative 5 10010 New Jersey 1984 Bond for job, science, and technology in NJ Leg. Referendum 1 10011 Oregon 1910 Repeal of poll taxes Initiative 9 10012 Arkansas 1914 Establishing children's home and welfare for minors Initiative 2 10013 Minnesota 1998 Providing constitutional right to hunting, fishing, trapping Leg. Referendum 5 10014 Alaska 2014 Oil and gas production, taxes, in AK Pop. Referendum 1 10015 Arizona 1972 Preemption of taxes for municipalities in AZ Initiative 1 10016 New Jersey 1985 Bonds for solid waste management facilities Leg. Referendum 3 10017 Washington 1991 Allowing assisted suicide Initiative 7 10018 Oregon 1974 Allows state employees to be state legislators in OR Leg. Referendum 1 10019 California 1922 Increasing loans / bonds for veterans' support Leg. Referendum 4 10020 Michigan 1984 Water and natural resource protection trust fund Leg. Referendum 4 10021 Massach . 1998 Referendum on deregulation of electric industry in MA Pop. Referendum 1 10022 Alaska 1982 Claiming state ownership of federal land Initiative 1 10023 Washington 1972 Transportation bonds and funding Leg. Referendum 13 10024 Utah 1966 Abolish board of examiners in UT Leg. Referendum 1 10025 California 1944 Public officers called to active military service Leg. Referendum 2 10026 California 2003 Preventing classification by race, ethnicity, nat origin Initiative 1 10027 Massach . 1986 Outlawing abortion Leg. Referendum 8 10028 Oklahoma 1935 Public assistance and welfare for needy and elderly Initiative 8 10029 Oregon 1908 Requiring railroads to give public officials free passes Pop. Referendum 1 10030 California 1911 Legislative sessions Leg. Referendum 20 10031 Oregon 1920 Voter registration Leg. Referendum 6 10032 Maine 1964 Guarantee and insure state payment of loans Leg. Referendum 1 10033 California 1949 State school building, construction, facilities bond Leg. Referendum 10 10034 Nebraska 1914 Construction of armory Leg. Referendum 2 10035 Georgia 2000 Property tax exemptions for non -profits Leg. Referendum 1 10036 North Dak . 1920 Legalize sale of cigarettes Initiative 1 10037 California 1948 Railroad brakemen Initiative 1 10038 Pennsyl . 2006 Bonds for vet s of Persian Gulf / Afghanistan conflicts Leg. Referendum 2 10039 California 1928 Allowing mutual water companies Leg. Referendum 1 10040 California 1986 Elected district attorney Leg. Referendum 1 10041 California 1952 Oaths for public officials Leg. Referendum 3 10042 Missouri 1910 Tax levy for higher education Initiative 2 10043 Montana 1908 Bonds for higher education / universities Leg. Referendum 17 10044 California 1922 Bonds for energy, utilities, and power Initiative 9 10045 North Dak . 1934 Regulate where alcoholic beverages are sold Initiative 1 10046 Missouri 1984 Sales / use tax for soil and water conservation Leg. Referendum 1 10047 Montana 1914 Establishing an athletic commission Pop. Referendum 3 10048 California 1984 Disqualification for libelous / slanderous campaigns Leg. Referendum 1 10049 California 1990 Changes to criminal code and law Initiative 1 10050 California 1911 Tax exemption for veterans Leg. Referendum 14 Note: Table above displays 50 ballot measures that were randomly selected from the full Ballot Measure database to be matched acros s analogous measures pursued across the U.S. states from 1902 - 2016 . The table shows the first state to attempt the specific ballo t measure, the first year it was attempted, a description of the measure, the type of measure, and how many states also attempted to enact the policy. Source: NCSL (2016 ) Ballot Measures Database. 54 explanations ( Berry and Berry 1990; Buckley and Westerland 2004; Volden 2006). Potentially, if all of the 5 0 policies were first adopted in 1902, then all fifty states would be at risk of passing these measures from 1902 through 2016, creating a maximum universe of 285,000 observations: 50 50 114 = 285 ,000 . However, because most of these measures came much later in time , the actual universe and risk pool for these initiatives and referendums is much smaller: 146,242 observations. And because of missi ng variable values , the empirical models rely on 60,000 to 80,000 observations. The dependent variable is whether a state adopted a ballot measure of interest in a given year. An adoption in one state makes it at risk for diffusion in other states. As is typical with EHA data , the dependent variable takes on a value of zero until the state enacts the policy specific measure in the given year , when it takes on a value of one (Blossfeld, Golsch, and Rohwer 2007) . Once the state adopts th e measure of interest , it is removed from the dataset for the remaining measure -specific years (as the state is no longer at risk of adopting that specific proposition ). But the state remains in the dataset for other ballot measures that it is at risk of e nacting. Variable Operationalization This longitudinal dataset spanning more than a century provides many rich opportunities to better understand the dynamics of plebiscitary action in the states over time. But the data™s breadth also presents challenges in finding relevant explanatory variables that also span this range. Despite major gains in data collection and dissemination of state policy and politics variables, few measures track to the early 1900s. For example, despite Gray and Lowery™s ( 1988) ex traordinary efforts to capture state interest group density by sector, these measures only date to the mid -1980s. As such, I use facially valid surrogate measures that cover as much of the timespan as possible. Variables™ names, descriptions, descr iptive statistics , and sources are referenced in Table A.1 in the Appendix. 55 In trying to identify the mechanisms driving the propagation of ballot measures across U.S. states, recall that my central hypothesis is Policy Learning (H 2). I anticipate that as more states pursue the analogous ballot measure, laggard states are more likely also to adopt the proposition . I operationalize Policy Learning as the cumulative number of states pursuing the given ballot measure. 18 I anticipate a positive relationship betw een a state learning about available solutions and adopting a given ballot measure. I test the Imitation Hypothesis (H 3) by relying on four variables to capture the economic, political, and institutional similarities between states: Similarity in State Re venue per capita , Similarity in State Party Control , Similarity in Citizen Ideology , and Similarity in Difficulty in Amending State Constitution . The similarity in state revenue per capita relies on Klarner™s (2013 b) measure of a state™s total income divided by the state™s population. The similarity in state party control uses Klarner (2013 a) and Ranney™s (1976) measure, where a 0 indicates Republican control, 1 indicates Democratic control, and 0.5 indicates biparti san control of the state government. Berry et al. ™s (1998, 2010) measure of a state™s congressional ideology scores is used to calculate the similarity in citizen ideology variable. 19 And I employ Lupia et al. ™s (2010) index of a state™s institutional hurdl es to amending its constitution via ballot measure to compute the similarity in difficulty changing the state constitution. To construct all the similarity variables, I calculate a state™s Euclidean distance from the average of all states in a given year. I then reverse code the variables so that an increase point s to greater similarity . These variables give us a sense of how extreme or typical 18 Operationalizing policy learning as the success rate of other states pursuing the ballot measures is another way of capturing this learning process. However, such an operationalization mirrors political learning (May 1992) rather than policy lea rning since it involves drawing lessons about the political process to achieve the ballot measure instead of simply learning about the policy solution. Subsequent chapters in this dissertation attempt to parse the difference between policy and political le arning. Nonetheless, measuring policy learning here as the success rate of other states pursuing the ballot measure produces similar, if not stronger, findings compared to using the current operationalization. 19 I include both party control and citizen id eology variables because they capture distinct, albeit related, factors that may affect the passage of a ballot measure. State party control represents the institutional political dimension, while citizen ideology proxies for the public opinion dimension. to a moderate, and not empirically pernicious, relationship. 56 a given state is relative to the average of all the other states along these yardsticks pertinent to diffusion. Pe r the Imitation Hypothesis, I expect that as states mirror others on these economic, political, or institutional dimensions (i.e., peer states), they will be more likely to adopt a given ballot measure. 20 To evaluate the National Environment Hypothesis (H 4), I rely on a Presidential Election Year dummy variable which is coded one for all years when a presidential election occurred and zero otherwise. Because ballot measures are designed to allow voters a direct say in policy and because national elections offer an opportunity to engage and mobilize citizens around issues, I anticipate a positive coefficient for the presidential election year variable. In addition to these covariates capturing external pressure , I also include a host of variables to c ontrol for the internal determinants of state policy change. For example, I include four dummy variables to account for how states ™ varying institutional arrangements affect ballot measure adoptions . I control for states that allow direct or indirect citiz en initia tives (Direct Democracy State ), states that permit changes to statutory language via legislative referend a (Statutory Legislative Referendum State ), and states that grant citizens an opportunity to repeal legislation throug h popular referenda ( Popular Referendum State ) (Fay and Wenger 2015; Lutz 1994). I anticipate that states™ institutional settings largely dictate the pursuit and adoption of new policies, with states that permit these additional avenues to voters to be more likely t o adopt a given measure. 21 I also consider other internal political and demographic factors. For instance, I account for a state™s Electoral Competitiveness . I suspect that a tighter electoral environment may lead 20 An advantage of these variables is that they simultaneously account for a state™s own internal context relative to other states™ contexts. Because of this, I do not include the root Šstate revenue per capita, party control, citizen ideology, and difficulty in amending the state constitution Šof these similarity variables in the models. Nonetheless, the inclusion of the root variables does not alter the overall findings. 21 Refer to Table 2.1 for details on each state™s access to different types of ballot meas ures. 57 legislators to shirk, delegating tough po licy decisions to voters via ballot measures so as not to risk electoral defeat, or compelling interest groups to circumvent any legislative impasse at the ballot box. I rely on Ranney™s four -year moving average of electoral competitiveness , which is booke nded between 0.5 and 1 where higher scores indicate greater competitiveness. Because passing a ballot measure is more challenging in more populous states (Boehmke 2005; Donovan and Bowler 1998 ; Lewis 2013), I include the natural log of State Population (Ln ). I expect a negative coefficient. And because politics in the South are markedly different from other regions (Foster 1978; Key 1949), I include a Southern State dummy variable based upon the U.S. Census Bureau™s regional classification. To weigh the ro le of state -level interest groups in the adoption of ballot measures, I include three proxy variables. First, to capture organized interests™ effect on ballot measures invoking moral values, I account for the percentage of a state™s Evangelical Population that identifies as Evangelical Christian or Mormon. I add a Union Membership Density variable capturing the percentage of the state™s workforce that is unionized to control for the labor movement™s influence. Still, with the hope of including a broader int erest group measure, I include a GINI Inequality Measure provided by Frank (2009). According to Morehouse™s (1981) state -level research on parties and organized interests, states with greater wealth disparity correlate with greater pressure group strength. The argument is that competing interest groups curtail the rights of others producing greater inequality. As such, wider inequality should be a stand -in for greater interest group activity. Therefore, I anticipate a positive relationship between the GINI coefficient and the adoption of a given ballot measure. Acknowledging the disproportionate number of ballot measures that both California and Oregon have pursued in the past century, I also created California Dummy and Oregon Dummy variables to re move their potential out -sized influence on any inferences we make. Finally, I also 58 include three policy topic dummy variables ŠGovernment Reform Measures , Bond and Budget Measures , and Tax and Revenue Measures Što account for the three most common policy domains for all types of ballot measures. The policy domain for each measure was coded by NCSL (2016) with most ballot measures coded for multiple policy areas. I opted to categorize each ballot measure by the first policy topic coded by NCSL. Accounting for these variables ensures th at no one topic area drives the empirical results. Methods I rely on a complementary log -log approach to estimate the parameters for my ballot measure diffusion models . Given the discrete nature of the dependent variable, uncertainties about the exact parametric relationship between the variables, and possible time duration dependence, complementary log -log models are appropriate (Box -Steffensmeier and Jones 2004; Buckl ey and Westerland 2004). The complementary log -log is better suited for the estimation of sporadic events (Buckley and Westerland 2004), which is the case for th is dataset . In fact, out of the 60,000 to 80,000 observations estimated in the models, there ar e only a couple hundred ballot measure adoptions. The probability equation for complementary log -log regression is as follows: Pr(=1|)= 1exp {exp ()}, where the probability of a state adopting a specific ballot measure in a given year, Pr (=1), is a function of the covariates, , and the coefficients, , are expressed as hazard ratios in discrete time fashion for each covariate ( Box -Steffensmeier and Jones 2004; Long 1997). Complementary log -log 59 regression parallels that of logistic regression, 22 relying on a complementary log -log link function (instead of a logit -link function) to specify parameters in terms of the hazard ratio of the event occurring to it not occurring (much like Cox model parameters). The coefficients are then exponentiated to be interpreted as hazard ra tios. But, because the interpretation of these exponentiated parameters is not always straightforward, I provide predicted probabilities and average marginal effects where appropriate. To account for temporal dependence Šthat the probability of pursuing a measure by a state in any year is related to its probability of adoption in previous years ŠI also include time and time -squared count variable s (Beck, Katz, Tucker 1998; Buckley and Westerland 2004). And to reduce potential heteroskedasticity in the error term, thus jeopardizing our inferences, I estimate all the models with robust standard errors clustered by state (Box -Steffensmeier and Jones 2004; Buckley and Westerland 2004). Results for Diffusion of Ballot Measures Before empirically testing the mecha nisms driving ballot measure diffusion, I start by evaluating the Minimal Diffusion Hypothesis (H 1). To do so, I categorize the ballot measures (from the random sample of 50) by the number of states that adopted them . Figure 3.6 is a bar chart displaying t he number of ballot measures from the random sample that were pursued by only one state, by 2 Œ 5 states, by 6 Œ 14 states, or by 15 or more states. Twenty -four of the measures (48%) appear to have been pursued by only one state. That is, nearly half of th e policies have yet to diffuse to other jurisdictions. Of course, it is possible that these innovations were pursued and adopted by states in other venues (e.g., legislature, courts, gubernatorial executive order, bureaucratic agency 22 Although unlike logit and probit regression, complementary log -log models produce an asymmetric function that approaches zeros more slowly and ones more quickly. 60 decision). In those ca ses, the idea may still diffuse but via a different venue not captured here. Therefore, we are unable to say for sure that these measures did not diffuse. Nonetheless, this descriptive statistic offers a reference point whereby almost half of all ballot me asures have yet to diffuse. 23 This non -diffusion rate could be much lower or higher in other institutional venues, 24 depending on the obstacles or openings to achieve policy change in those alternative arenas. Still looking at Figure 3.6, we see that a sizeable number of measures Š13 measures or 26% of total Šwere pursued by between two to five states. Added together with the number of measures that did not diffuse, this suggests that nearly three -quarter s of new policies either do not diffuse or diffuse to a very limited number of jurisdictions. 25 Only 20% of the 50 ballot measures (10 in total) were pursued by six to fourteen states, and a mere 6% of the measures (3 in total) were pursued by fifteen or more states. Yet, as Karch et al. (2016) point out, policy scholars have disproportionately selected these policies that d iffuse widely to study and understand policy transfer. In fact, looking at 23 different policies reviewed by Graham, Shipan, and Volden (2013), Karch et al. (2016) find that the average number of state adopters per policy is 29.3 states. This diffusion rat e is not reflective of the low or non -existent diffusion rate for the overwhelming majority of policy innovations, as evidenced by this sample of ballot measures . 23 Undoubtedly , some of the ballot measures that have yet to spread to other jurisdic tions are individual in nature. For example, Colorado legislators asked voters via referendum in 2004 to remove obsolete amendments from the state constitution (Policy ID: 10001). Oregon legislators wanted voter input in 1974 to allow state employees to ru n for elected office (Policy ID: 10018), while Oregon voters sought to repeal a 1908 statute that required railroads to give elected officials free transit (Policy ID: 10029). Still, many of the other measures that have yet to spread seem ripe for emulatio n. California ™s ballot initiative in 1984 disqualifying candidates from office if they are found to have defamed their opponents during the campaign (Policy ID: 10048) is the type of reform that most voters would likely support. 24 A recent analysis by USA Today in conjunction with The Arizona Republic and the Center for Public Integrity show that over an eight -year period (2010 Œ 2018), some 10,163 bills were introduced in state legislatures that were essentially copied from bills promoted by interest group s (O ™Dell and Penzenstadler 2019). If more than 109,000 bills are introduced in state legislatures every year, then roughly 1.2% of all pieces of legislation introduced at the state -level are copied from interest groups and have the potential to diffuse ac ross states. 25 Of course, just because the same policy was pursued by another state does not implicate diffusion. Not only is pursuit not equivalent to enactment, but states can simultaneously converge upon a policy solution independently of other states (Bennett 1991; Boehmke 2009 b; Freeman 2008). 61 In sum, Figure 3.6 offers some evidence (even if the sample size for each category is too small to achieve a statistical difference between categories) for the Minimal Diffusion Hypothesis (H 1). Nearly half of the ballot measures did not diffuse, and another quarter of the propositions were only pursued (and not necessarily adopted) by a handful or less of other states. At the very least, this offers additional evidence beyond Karch et al. (2016) that policy scholars are overstating the phenomenon of policy diffusion by selecting cases that have already widely diffused. 26 Figure 3.6: N umber of Ballot Measures from Sample that Have Diffused or Have Yet to Diffuse Note: Bar graph displays the number of policies from the 50 ballot measures that were randomly selected from the full Ballot Measure database (1902 Œ 2016) that were adopted by only one state (i.e., did not diffuse), by 2 Œ 5 states, by 6 Œ 14 states, or by 15 or more states. Source: National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. 26 One might wonder whether policy topic area matters in this diffusion process. Are some policies more likely to spread than others? Dividing the 50 ballot measures randomly selected into two groups Š(1) those ba llot measures only pursued by one state and (2) those ballot measures pursued by more than one state ŠI categorize the measures by topic area. Figure A.4 in the Appendix provides the topic areas most likely to diffuse and not diffuse, based on the random sample of measures. Because of the small sample size, any conclusions are cautious at best, but from this sample it appears that measures dealing with Bonds and Budgets as well as the Military and Veterans are more likely to be pursued (and potentially diffu se) to other states. 62 Although there is some eviden ce that most ballot measures do not diffuse, what factors drive the ballot measures that do diffuse? The next step is to analyze which external and internal factors drive the enactment of these measures across states. Table 3.3 displays the se empirical res ults. Recall that the dependent variable is whether a state adopted a given ballot measure in a given year. Model 1 fiAdopt Ballot Measurefl accounts for both external (i.e., policy learning, imitation, national environment ) and internal (i.e., political, institutional , demographic ) forces. Model 2 fiInterest Group Influencefl contains three variables to account for the role that state -level interest groups may play in the pursuit and adoption of ballot measures. Model 3 fiRemoving CA & ORfl include s dummy variab les for all California and Oregon observations so as not to allow two influential observations to cloud our understanding of these diffusion dynamics in the forty -eight other states. Finally, Model 4 fiWith Key Policy Typesfl includes dummy variables that co ntrol for the three main policy domains represented by initiatives and referendums, so our inferences are unencumbered by the variation due to any one major policy area . The resounding takeaway from all the models is that some ballot measures, even if not all or most measures, diffuse across the U.S. states. And the main external force for this diffusion is policy learning (H 2), whereby policy actors learn from and emulate the solutions pursued in other states to address common problems. States do not appea r to arrive upon these policy solutions independently of one another, but rather seek out, process, and act on this external information. Based upon estimates from Model 4, Figure 3.7 displays the predicted probability of a state adopting a given ballot me asure in a given year as the number of other states pursuing the same measure (i.e., policy learning) increases. From the figure we see that a state™s risk of adopting a particular measure in any given year remains low: around 0.007 percentage points with no states pursuing the measure. But as ten states pursue, a state™s propensity to adopt increases to 0.4 percentage points in any given year. 63 Table 3.3: Ballot Measure Diffusion Models Explanatory Variables Model 1: Adopt Ballot Measure Model 2: With Interest Groups Model 3: Removing CA & OR Model 4: With Key Policy Types Policy Learning [+] 0.174* (0.015) 0.185* (0.016) 0.185* (0.016) 0.177* (0.021) Similarity in State Revenue per Capita [+] -0.002 (0.086) 0.030 (0.071) 0.004 (0.065) 0.001 (0.066) Similarity in Party Control [+] 1.233 (0.905) 0.343 (0.976) 0.313 (0.825) 0.297 (0.834) Similarity in Citizen Ideology [+] -0.015 (0.010) 0.008 (0.014) -0.003 (0.009) -0.003 (0.009) Sim. in Difficulty Amending Constitution [+] 1.050* (0.382) 1.338* (0.386) 1.583* (0.396) 1.598* (0.396) Presidential Election Year [+] 0.892* (0.175) 0.705* (0.197) 0.724* (0.198) 0.731* (0.194) Direct Democracy State [+] 1.082* (0.452) 0.870* (0.373) 0.724ƒ (0.374) 0.729* (0.370) Statutory Leg. Referendum State [+] -0.031 (0.851) 0.299 (0.623) 0.741 (0.568) 0.762 (0.567) Popular Referendum State [+] 0.301 (0.929) 0.068 (0.753) -0.559 (0.680) -0.567 (0.679) Electoral Competitiveness [+] 2.009 (1.909) 0.787 (2.041) 1.711 (1.611) 1.711 (1.626) State Population (Ln) [ -] 0.205 (0.352) 0.164 (0.276) -0.254 (0.205) -0.245 (0.203) Southern State [ -] -0.397 (0.634) -0.215 (0.586) 0.125 (0.588) 0.122 (0.588) Evangelical Population [-] --- -0.036* (0.016) -0.029* (0.013) -0.030* (0.013) Union Membership Density [+] --- -0.012 (0.028) -0.014 (0.026) -0.016 (0.026) GINI Inequality Measure [+] --- 7.244* (3.127) 5.855* (2.853) 6.228* (2.982) California Dummy [+] --- --- 2.728* (0.604) 2.709* (0.599) Oregon Dummy [+] --- --- 1.382* (0.350) 1.389* (0.351) Governmental Reform Measures [+] --- --- --- -1.011* (0.271) Bond and Budget Measures [+] --- --- --- 1.487* (0.323) Tax and Revenue Measures [+] --- --- --- 1.143* (0.340) Constant -8.515* (2.978) -11.801* (3.110) -8.787* (2.331) -9.693* (2.300) N 81,513; 64,241; 64,241; 62,241 2 (14, 17, 18, 20) : 542.46 *; 417.66*; 2208.71*; 2099.59* AIC / aROC 3049.91; 2309.82; 2246.71; 2141.59 / 0.87; 0.87; 0.88; 0.90 Log Likelihood: -1509.96; -1136.91; -1104.35; -1049.8 ƒ 5, two tailed. Dependent variable is likelihood of adopting a given ballot measure. Statistically significant complementary log -log regression coefficients are in bold face. Robust standard errors, clustered by state, are in parentheses. Models also include a time and time squared count variables to account for temporal dependence; coefficient s are omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. 64 Figure 3.7: Predicted Probability of Adopting Ballot Measure as Policy Learning Increases As twenty states pursue the measure, a state™s likelihood of enacting rises to 2.3 percentage points in any given year. Another useful way to interpret policy learning™s overall influence on a state™ s likelihood of adopting a ballot measure is to calculate its Average Marginal Effect (AME). AMEs can be interpreted as the instantaneous rate of change in adopting a ballot measure (i.e., the dependent variable) following a one unit increase in the indepe ndent variable. AMEs are computed using the delta method by calculating the marginal effects for every observed value of the independent variable and then averag ing across the resulting estimates. 27 In essence, AMEs provide a summary measure of a predictor variable™s average influence on the outcome variable of interest by relying on 27 Unlike Marginal Effects at the Means (MEMs), which is another common approach, AMEs provide a single statistic using the full distribution of the explanatory variable rather than a few selective values and better capturing the variab ility of the independent variable. 65 variables™ actual values (rather than averages of those values). Figure 3.8 presents the Average Marginal Effects for the key variables in Model 4. Policy learning™s AME is 0.1 percentage points, where each additional state that pursues a given ballot measure increases subsequent states™ risk of adopting that measure by 0.1 percentage points. A lthough policy learning™s effect on a state™s probability of adopting a measure may appear substantively smal l, these numbers reflect the likelihood of enacting a specific measure in any given year over the full time a state is at risk. Indeed, the aggregate risk of a state adopting any one of the ballot measures during a given year in the time period is only 0.3 percentage points. Considering this , as well as the political and institutional obstacles that policy actors must overcome to achieve policy change, policy learning™s 0.1 marginal effect on the likelihood of adoption is not so trivial. Figure 3.8: Average Marginal Effects of Key Variables from Diffusion of Ballot Measures Model Note: Figure displays A verage Marginal Effects (AMEs) calculated for key variables from Model 4 in Table 2.3. AMEs can be interpreted as the instantaneous rate of change in adopting a ballot measure (i.e., the dependent variable) following a one unit increase in the independent v ariable (located on the Y -axis). AMEs are co mputed using the delta method by calculating the marginal effects for every observed value of the independent variable and th en average across the resulting estimates. 66 Beyond states looking to and learning about policy solutions pursued by other governm ents, states also appear to take cues from those with similar institutional hurdles to amending their state constitutions. Offering some support for the Imitation Hypothesis (H 3), states are more likely to adopt a ballot measure if their inst itut ional arra ngements mirror others ™ settings for changing their constitutions. Not surprisingly, states with high hurdles to amending their constitutions (e.g., approval by multiple legislative sessions or a supermajority of voters) are less likely to adopt a given ballot measure. Moving a state one unit closer to the average degree of difficulty in amending a state constitution (e.g., signatures and legislative approval to get measure s on the ballot) raises a state ™s likelihood of adopting the measure by 1.1 percentag e points. As suspected, presidential election years also make states more likely to adopt a ballot measure , supporting the National Environment Hypothesis (H 4). States are 0.3 percentage points more likely to enact a ballot measure during presidential election years than during off -year elections. Also as expected, interest groups appear to play a role in states pursuing and adopting ballot measures. Two of the thr ee variables are statistically significant across the models. States with higher proportions of Evangelical Christians are less likely to pass ballot measures. Perhaps knowing the Evangelical Christians ™ political clout, legislators and interest groups may be less likely to put some ballot measures (especially those invoking moral values) before voters. States with greater income inequality , thus suggesting stronger interest group presence , also appear more likely to adopt ballot measures. Nonetheless, labor unions™ presence in the state workforce does little to predict whether a state will adopt ballot measures, perhaps underscoring organized labor™s waning influence in the last half -century. Most of these results in Table 3.3 reinforce w hat we know about policy diffusion . And t he results hold even when the major ballot players ŠCalifornia and Oregon Šare removed from the analyses. But some findings refine our current understanding. For example, it is somewhat surprising 67 that only institutio nal variable to have any influence on the adoption of ballot measures was the direct democracy variable. While direct democracy states were 0.2 percentage points more likely to adopt a ballot measure in any given year, states™ that allowed statutory legisl ative referenda and popular referenda were no more likely to pass ballot measures. The null finding is paradoxical since we might expect policy actors with more avenues available to achieve policy change to utilize those avenues. Of course, this result may be an artifact of the random sample of ballot measures. The three main policy variables also offer insightful results. Ballot measures on reforming government (e.g., legislature, governor, courts, localities) are less likely to be adopted, while those dea ling with raising revenue via bonds or taxes are more likely to be enacted. Perhaps these latter measures allow state policymakers to get what they want with direct approval by the electorate or are required by law (as some states require citizen approval for taking on debt or increasing taxes) .28 Finally, it is worth acknowledging the robustness of these models. All four models in Table 3.3 produce superior aROC statistics, a measure indicating a model™s accuracy where a score of 0.5 suggests a random c lassification and a score of 1.0 suggests a perfect classification. All four models aROC statistics range between 0.87 for Model 1 to 0.90 for Model 4. The results are robust to alternative estimation techniques to boot . The findings are nearly identical t o the complementary log -log parameter estimates. In addition, different operationalizations for interest groups, including an education variable, Morehouse™s (1981) measure of pressure group strength, and a dummy variable for when a state™s Chamber of Comm erce was founded had no statistical effect on whether a state enacted a ballot measure. 28 Interactions with the policy topic variables and the policy learning variable had no effect. That is, policy learning™s role held regardless of the policy domain under consideration by voters. Moreover, including contro ls for the other ballot measure topic areas had no statistical effect on the dependent variable. 68 Evaluating Selection Bias Recall that a concern with past research is the tendency to choose and model policies that knowingly diffuse to explain policy diffusion. B ut this selection on the dependent variable is akin to only modeling countries that go to war to explain conflict (King, Keohane, and Verba 1993). I leverage the fact that my dataset includes both ballot measures that do and do not diffuse to evaluate the potential effect of this selection bias on our understanding of diffusion dynamics. The Model 4 parameters in Table 3.3 from the pr evious section were estimated relying on ballot measures that were pursued in only one state and ballot measures pursued by two or more states. I re -estimate Model 4 using data from this latter group: only those ballot measures pursued by two or more states. I then compare Model 4™s original estimates (relying on all the ballot measures) with Model 4™s re -estimates (relying onl y on ballot measures that were pursued by two or more states). I do not report the new coefficients here out of consideration of space. While the new estimates largely comport with the original estimates and lead to the substantively similar conclusions, t 2(22) p -estimated Model 4). Furthermore, any difference bet ween estimates of policy 2(1) p Even if ill -advised, modeling only those policies that have been pursued by two or more states does not produce substantively different results or interpretations, at least for this small random sample of ballot measures. But what if we model only those ba llot measures that have knowingly been pursued by multiple governments, falling prey to what Karch et al. describe as a fipro -innovation biasfl? I re -estimate Model 4 relying solely on ballot measures that were pursued in six or more states and compare them to Model 4™s original estimates. Again, both models are 69 2(22) 23597.19 , p -estimated Model 4). Alt hough the coefficient estimates for policy learning between the two models are not statistically different, their estimated marginal effects are different. Recall that the original model estimated that an additional state pursuing a given ballot measure in creased subsequent states™ risk of adopting the measure by 0.1 percentage points. But the revised Model 4, relying only on ballot measures that knowingly diffused, put policy learning™s marginal effect at 0.2 percentage points. This estimated effect is twi ce as large as reality, inflating policy learning™s actual impact on policy change. Although the results are not definitive, this suggests that by only modeling policies that have spread widely, we may be overstating policy diffusion™s existence in the bla ck box of the policy process. Conclusion I admit three main limitations to this research and the inferences we can draw. First, leveling criticism against policy scholars for predominantly siloing our focus to the legislative context and then narrow ing my attention to another sole context may seem disingenuous , if not hypocritical. I do turn my attention to the spread of policy ideas across multiple, competing venues in subsequent chapters. What is more, the increase in use and success of ballot meas ures, as well as the vast institutional variation within states, makes the ballot measure context ripe for exploration. Second, by only focusing on the ballot measure context, I concede that I may be missing the spread of these policy innovations in other venues. That is, State A may have enacted the new idea via legislative referendum (captured in this data), but State B adopts it in the legislature (not captured in the data). Given this, it is possible I am understating the existence of policy diffusion. But short of having the entire universe of all state policies pursued and enacted across all possible venues, this is a challenging limitation to resolve. 70 Lastly, like other state politics and policy research, this undertaking is hampered by limited time -series data for key policy, economic, political, interest group, and demographic covariates. Although I capitalize on a massive repository of state -year variables compiled by numerous scholars and aggregated via the fiCorrelates of State Policy Projectfl (Jo rdan and Grossmann 2018), few variables span back to the early 1900s. As a result, several units of observation are missing values and dropped from the models. Moreover, the task of matching analogous ballot measures across some 7,800 initiatives and refer endums is a laborious one. Given this, I only present preliminary findings based on a small random, although representative, sample of ballot measures. Because of these limitations, I am unable to fully flesh out answers to initial empirical questions. Sti ll, by relying on the full set of ballot measures pursued across the U.S. states over the last century, this chapter offers substantial descriptive information about the ebb and flow of the use of and success rate of ballot measures by states over time. Po licy actors have relied on legislative referenda, citizen initiatives, and other ballot measures much more frequently since the 1960s, while the ir use of popular referenda to repeal certain policies has drastically declined over time . Not only has the freq uency in use of these measures increased, their rate of enactment (especially for measures referred by state legislatures) has also increased. This variation extends to topic areas as well, with those measures pertaining to governmental institutions or ref orm, tax and revenue, bonds and budgets, and elections and campaigns compris ing nearly half of all measures put before vote rs. The empirical results offer further evidence that states do not operate in vacuums. What happens in one state affects the policy decisions of others. The diffusion dynamics of ballot measures largely parallels the diffusion dynamics documented in the myopic legislative context. Policy actors purposively learn about policy solutions elsewhere, as well as look to peer states wi th similar institutional contexts for potential policy solutions. Despite offering evidence that many ballot propositions do diffuse to other states, however, the evidence also suggests that policy 71 scholars run the risk of overstating the occurrence of pol icy transfer. I find that nearly half of the ballot measures in my sample do not diffuse, and almost three -quarters are only pursued by less than a handful of states. Just six percent of the ballot measures were pursued by fifteen or more states. Furthermo re, I show that by only including ballot measures pursued by six or more states in my model, the empirical results inflate key mechanisms™ marginal effect on the outcome. These findings should caution all of us from making generalizations about policy diff usion relying on such limited data. 72 CHAPTER 4: A THEORY OF VENUE DIFFUSION It is no longer a novel assertion that the public policy choices made in one governmental unit influence the policy choices made in another governmental unit. Five decades of research have provided abundant evidence that governments frequently adopt new po licy ideas enacted by prior governments Šthat innovative policies diffuse. Jack Walker ™s (1969) and Virginia Gray™s (1973) seminal articles exploring patterns of policy adoptions across U.S. states spurred a burgeoning interest in policy diffusion within an d beyond the American context. Hundreds of subsequent studies have documented and detailed the transfer of various policy innovations across space and time. Graham, Shipan, and Volden (2013) point to more than 800 publications documenting policy diffusion since 1958, with half of these written in the last decade. 29 Scholars do not dispute that policy diffusion occurs. We recognize that the public policymaking process is dynamic and interdependent across multiple layers of government. But, as Chapters 2 and 3 highlighted, much of our understanding of this interdependence emanates from research almost entirely focused on the diffusion of (1) the policy output itself (2) from one legislative body to another legislative body. The overwhelming majority of policy d iffusion studies have emphasized the transfer of ‚policy X™ in ‚legislative body A™ to ‚legislative body B.™ As a result, we have largely overlooked the potential diffusion of other key parts of the policymaking process beyond the policy itself, such as po licy winnowing, the framing of the problem and policy solution, choosing an institutional venue, routing opponents, implementing the policy, evaluating the policy, and spinning the policy evaluation, among other components. And we have not yet fully 29 See Graham, Shipan, and Volden (2013) for a comprehensive review of policy diffusion research from the American Politics, Comparative Politics, and International Relations™ perspectives. They perform a network analysis of nearly 800 diffusion articles written since 1958 with the aim of identifying the broad them es and conclusions within the respective subfields. They find that, often times, subfields are talking past one another and make a call for greater integration of diffusion research between subfields. Furthermore, they encourage diffusion scholars to go be yond whether or not policies diffuse, asking and answering the more challenging questions of how, whe re, and whe n policies diffuse. 73 analyz ed the diffusion of innovations via other institutional venues where policy change can also occur. Put another way, we have a good sense that the innovative policy pursued in one state influences another state™s decision to pursue the same policy. But we d o not know if, say, State A™s decision to pursue an innovation via one institutional venue influences State B™s decision to pursue the same innovation via the same institutional venue. Does the adoption of fiPolicy X fl via fiVenue Y™ in one governmental unit increase the likelihood that fiPolicy X fl is adopted, especially via fiVenue Yfl in subsequent governmental units? This chapter builds on past research mapping the patterns of policy adoptions across space and time and lays the theoretical underpinnings for why we might expect venue choice (and possibly other key elements of the policymaking process beyond the innovation itself) also to be copied. More concretely, this chapter relies on a political learning explanation, paralleling the polic y learning account for the diffusion of innovations, whereby policy actors not only learn about an adopted policy and its effect but also actively learn about the political processes and tactics employed to bring about change in the innovative jurisdiction s. Venue shopping is a crucial step in optimizing the chances of a policy™s enactment and entrenchment in the political system. Innovator states™ choice of institutional venue to attempt a new policy idea may affect early -adopter, early -majority, late -majo rity, and laggard states™ selection of venue to pursue the policy idea. Depending on a state™s institutional arrangements, policy actors may follow their predispositions to attempt policy change or learn from the paths being taken in other states, especial ly those with similar institutional arrangements. If policy diffusion implies that a government™s policy choices are conditional on the prior choices of other governments ( Gray 1973 ; Walker 1969), then it is also plausible that a government™s choice of ven ue to pursue a policy is influenced by the prior venue choices of other governments pursuing the innovation. Simply put, venue selection for a policy innovation may also diffuse, a phenomenon I term venue diffusion . 74 Venue Shopping The process by which i nnovative policies enter the governmental arena matters. One critical element of this process is venue shopping, the act of strategically choosing among the variety of institutional settings where policy change can occur. Elected officials, policy advocate s, interest groups, nonprofit organizations, bureaucrats, concerned citizens Šthose individuals or groups within or outside the public sector Šdevelop and advance policy solutions for societal problems, relying on their knowledge of and connections within th e political system to press for policy change to upend the status quo at the most opportune time (Baumgartner and Jones 1993; Kingdon 1984; Sabatier and Jenkins -Smith 1993). Since there are multiple avenues (both horizontal and vertical) for policy adoptio n in the U.S. federated and fragmented system with shared jurisdictional authorities, advocates can fivenue shopfl (Baumgartner and Jones 1993). These policy promoters, weighing the political, financial, cultural, and institutional constraints, can select the venue they believe will be most feasible and favorable to their problem definition and innovative policy solution, and where they can more equally compete with challengers (Constantelos 2010; Pralle 2003). If policy change does occur, these individuals or groups not only achieve the desired policy outcome but also gain new institutional rules, actors, and constituencies around the policy to help rebuff short - and long -term attempts at reform (Karch 2009; Lubell 2013; Maltzman and Shipan 2008; Pralle 200 3). Surely this is partly what E.E. Schattschneider implied with his maxim that finew policies create new politicsfl (1935: 288). Or as Karch (2009) put it: fisuccessful venue shopping may alter the terrain on which subsequent decisions are madefl (38). Policy entrepreneurs and policy actors pushing for new solutions at the state level are increasingly pursuing policy change in institutional venues other than state legislatures (Miller 2009; NCSL 2016; Piott 2003; Smith and Tolbert 2004, 2007). Conventionally , change actors looking to adopt new policies or reform previously enacted policies would start in the fipeople™s branch.fl They 75 would lobby state legislators to introduce a bill, push the policy through various committees, secure a floor vote, and if the bill passed both chambers, reconcile different versions of the legislation in conference and convince the governor to sign it. However, different routes on this classic roadmap are progressively being taken to attempt policy change. Policy promoters are often pursuing solutions to public problems at the ballot box, via gubernatorial executive orders, through state high court rulings, and even within bureaucratic agencies. Indeed, policy actors are capitalizing on the federated and fragmented U.S. structure, on e of the most essential yet often overlooked features of American democracy, to seek policy change. Policy entrepreneurs and actors are turning to alternative institutional venues for a variety of reasons, including greater knowledge and experience with on e venue over another (Pralle 2003); increasing polarization and gridlock in state legislatures ( Hinchliffe and Lee 201 6; Shor and McCarty 2011) and at the federal level (Hetherington and Rudolph 2015; Poole and Rosenthal 1997, 2007); a desire to codify or annul policies in state constitutions (Miller 2009; Fay and Wenger 2015); attempting to preempt other institutions (Boehmke, Osborn, and Schilling 2015; Dumas 2017; Gerber 1996); and a desire for greater popular sovereignty (Bowler and Glazer 2008; Lewis 2 013); among others. 30 Although multiple venues may contain various veto points to deny policy change, they also offer multiple opportunities for participation and to pursue change, especially if change agents initially encounter failure in one or more of th e venues (Lubell 2013; Pralle 2003). Frequen cy of Venue Shopping Despite this increase in policy activity in other political institutions, it is unclear from the literature how frequently and for which topics entrepreneurs and actors fishopfl for ven ues outside 30 These are additional claims that could benefit from further examination, especially exploring how polarization at the fed eral and state levels may contribute to augmented venue shopping. 76 the standard legislative process. That is, how often is more than one venue used to pursue policy change? Chapter 3 provided some clues that policy actors pursue ballot measures especially for policies related to government reform, bonds and bu dgets, and taxes and revenues. But I was unable to account for policy innovations that were pursued as ballot measures in some states and as legislation in other states. To get an initial sense of the frequency and variation in institutional venue shoppin g for policy innovations across states, I rely on a sample of 95 diverse innovative policies attempted from 1916 Œ 2009 compiled by Boehmke and Skinner (2012). 31 The authors acknowledge that while most of the policies were pursued in state legislatures, som e were also pursued via ballot measures. But I want to know which policies were only pursued in one state forum or in multiple state arenas. Therefore, f or each policy, I researched and coded whether the policy was pursued (1) only in state legislatures; (2) via state legislatures, legislative referenda, or citizen initiatives; or (3) only via legislative referenda or citizen initiative. 32 This straightforward exercise reveals the variation in venue where these innovations were pursued , pointing to the freq uency of venue shopping for a given set of policies . Table 4.1 provides a breakdown of where these 95 policy ideas were pursued. Six out of every ten of the policy innovations in the dataset were only channeled through the traditional state legislati ve processes. But nearly four out of every ten innovations were endeavored in multiple 31 See Table B.1 in the Appendix for the full list of policies, as well as the venues where the policies were pursued, the years of the policies™ first and last adoption, the number of sta tes that successfully enacted the innovations, and the rate of adoption. To be sure, this is a convenience sample of policies. However, the dataset includes a diverse group of policies covering a broad array of issue areas: e.g., abortion, criminal justice , economic development, health, gambling, tax, welfare, among others. Moreover, these policies were selected by other researchers (e.g., Boehmke and Skinner 2012; Walker 1969) and not chosen based upon the main interest of this project: the institutional v enues where the policies were adopted. 32 I used a variety of sources to identify the venues where the policies were pursued, including Ballotpedia, LegiScan, LexisNexis, National Conference of State Legislators, among other search databases. I relied on a similar matching strategy used in Chapter 3 for ballot measures (see Appendix A for an explanation of this strategy) to identify which of these 95 policy innovations compiled by Boehmke and Skinner (2012) were pursued as ballot measures, cross -referencing NCSL™s Ballot Measure Database (NCSL 2016). 77 venues, with actors in at least one state pursuing those policies via the state legislature, legislative referendum, or citizen initiative. Meanwhile, two percent of the policies in the dataset w ere only attempted via a ballot measure. These simple statistics suggest that while most new policies are still being pursued in the fipeople™s branch,fl the people are also being asked to vote directly on a fair share of innovation s. Moreover, these numbers imply that venue shopping occurs and occurs on a fairly regular basis . Table 4.1: Assessing Venue Choice for Sample of 95 Policies Policies Pursued via: Number of Policies Percentage of Policies Only State Legislature 57 60% State Legislature, Legislative Referendum, or Citizen Initiative 36 38% Only Legislative Referendum or Citizen Initiative 2 2% Total 95 100% Note: A sample of 95 diverse policies (1916 Œ 2009), compiled by Boehmke and Skinner (2012), were assessed for the choice of institutional venue Šstate legislature, legislative referendum, citizen initiative or popular referendum Šwhere the policies were pursued by at least one state via those venues. Perhaps the frequency of forum shopping depends on the type of policy pursued. To explore this, I broke down the rate of venue shopping by policy category for the sample of 95 innovations. Table 4.2 displays the findings. Gun legislation, health care polic ies, welfare laws, women™s rights bills, and miscellaneous regulations appeared to witness the least amount of activity outside of state legislatures. Morality policies, however, predominately encompassing abortion, gambling, and gay rights policies, as we ll as tax and economic policies, experienced higher activity outside the standard legislative context. Not surprisingly, nearly all innovations in the sample were pursued by at least one state legislature, the most popular policy venue (Chubb 1983). But a quarter of the policies were also attempted via legislative referendum by at least one state , and a third of the policies were decided by citizen initiative or popular referendum by at least one state in the union. Although the preceding categorizations ar e descriptive and narrowly focused on only a few venues (excluding state courts, gubernatorial executive orders, bureaucratic agency decisions, federal forums ), they illustrate that venue fishopping aroundfl happens, especially for some policy types. 78 Table 4.2: Assessing Venue Choice for Sample of 95 Policies by Policy Category Policy Category Number of Policies Percent of Policies with in category where at least one state pursu ed policy via Legislature Percent of Policies with in category where at least one state pursu ed policy via Legislative Referendum Percent of Policies with in category where at least one state pursu ed policy via Citizen Initiative / Popular Referendum Abortion 3 100 67 67 Crime 17 94 29 29 Drugs and Alcohol 7 100 0 29 Economic 4 100 50 50 Education 4 100 25 50 Environmental 4 100 25 25 Gambling 2 100 100 100 Gay Rights 1 0 100 100 Governmental Issues 9 100 44 67 Gun Laws 1 100 0 0 Health 19 100 5 11 Labor Rights 1 100 0 100 Miscellaneous Regulation 5 100 0 20 Racial Issues 1 100 0 100 Tax 4 100 75 50 Transportation 7 100 43 29 Welfare 4 100 0 0 Women's Rights 2 100 0 0 Total or Average 95 98% 26% 34% Note: A samp le of 95 diverse policies (1916 Œ 2009), compiled by Boehmke and Skinner (2012), were assessed for the choice of institutional venue Šstate legislature, legislative referendum, citizen initiative or popular referendum Šwhere the policies were pursued by at least one state. The data above ref lect the percentage of policies within the different policy categories attempted in the respective institutional venues. What Motivates Venue Shopping? Over the years, scholars have advanced various theories for what motivates policy actors to pick diff erent venues to press for policy change. Baumgartner and Jones (1991, 1993) portray venue shopping as a strategic exercise of matching the right policy image frame to the receptive venue. As one example, the authors recount how groups painted a negative, e nvironmentally dangerous image of nuclear power following the Three Mile Island nuclear accident and other incidents to break the decade -long policy monopoly between energy companies and the federal government. These groups used this new image to garner pu blic support and to press for change in multiple receptive venues, including Congress and the courts. Sabatier, Jenkins -Smith, and colleagues, however, suggest that advocacy coalitions frequently venue shop, picking the avenue or avenues where they will ha ve a competitive advantage (Sabatier and Jenkins -Smith 1993; Jenkins -Smi th et al. 2014). This narrative 79 suggests that these groups try to upend the status quo by targeting as many venues as possible, reducing risk through diversification (Boehmke, Gailmard , and Patty 2013; Constantelos 2010; Jourdain, Hug, and Varone 2017 ). Still, other scholars acknowledge the challenges in changing policy and the resource limitations of policy advocates. Holyoke, Brown, and Henig (2012) theorize that policy actors consid er their resources, opponents™ resources, and the venue location of ideologically congruent officials when picking a venue . These change agents prefer to pressure friends instead of foes. And they are especially drawn to venues already working on the issue of interest. According to Lubell and colleagues™ fiecology of gamesfl perspective, acknowledging a dynamic policy process where outputs are the fifunction of decisions made in multiple games over timefl (Lubell 2013 : 538 ), policy stakeholders have limited inf ormation, limited cognition, and limited resources, thus relying on heuristics to select the institutional venue they believe will optimize the outcome. Over time, more experienced policy advocates may cultivate a particular set of skills and resources for a specific venue, producing a penchant for one forum over others in a policy game (Lubell, Henry, and McCoy 2010; Lubell 2013). Pralle (2003) also supports the notion that individuals and groups pressing for policy change are boundedly rational, face int ernal and external constraints, and suffer from a positive feedback loop. Upon selecting a particular venue to advance an issue, the issue monger™s decision fi shapes the kind of issues and campaigns promoted by the advocacy group, such that it becomes a self -reinforcing process fl (Pralle 2003: 243) . Rather than pursuing all venues in an instrumental fashion, Pralle suggests that policy actors engage in informed venue shopping. As a result, these policy professionals produce a pseudo path -dependen ce for one venue over another, relying on the skills, resources, and connections they have developed to advance new causes and defend old ones. 80 Ley and Weber (2015) make a significant contribution to the literature by trying to combine these various, sometimes competing, narratives into a new Adaptive Venue Shopping (AVS) framework. They charge that emergent groups tactically choose a venue based on assessments of their own political, legal, and technical strengths; assessments of their opponents™ reso urces and capacity; and the degree to which their opponents control a venue as well as the receptivity of an image within the site (Ley and Weber 2015: 706). These actors rank the venues available to them based on these dimensions, pursuing policy change i n the fibestfl venue where the group maintains a relative advantage in resources over opponents, can gain control of the venue that is favorable to the policy image. Importantly, Ley and Weber add that when policy advocates fail in one forum , they can learn, adapt, and transfer their resources to another institutional venue that may yield a more favorable outcome. While Ley and Weber™s (2015) AVS Framework is perhaps the most ardent attempt yet to synthesize the complementary and rival arguments for venue sho pping, it does not account for external political learning that may also influence venue choice. The AVS implicitly acknowledges internal learning by individuals and groups (e.g., adapting strategies post -failures) but ignores the interdependence between a dvocates across peer states. Figure 4.1 summarizes policy actors™ primary considerations in picking a venue identified in the literature. 81 Figure 4.1: Policy Actors™ Venue Shopping Considerations Venue Diffusion and Political Learning Relying on and integrating the policy diffusion and venue shopping literatures , I conceptualize the venue shopping process in pursuing new policy innovations in the following three ways. First, I differentiate between policy entrepreneurs and policy actors ™ processes to pick an institutional venue. Much of the literature conflates terminology for the cast of characters that press for policy change, including terms like policy entrepreneur, policy advocate, policy professionals, policy actors, advocacy coali tions, interest groups, mass membership organizations, policy stakeholders, among many other analogs. 33 I define policy entrepreneurs as those innovative individuals or groups within and outside the public sector that are the first to pursue a new policy. Policy entrepreneurs, through their fiskillful mobilization of substantive justifications and the accurate identification and thoughtful cultivation of allies, can and do bring new policy into being fl 33 I define fipolicy entrepreneursfl somewhat differently than other scholars. Mintrom (1997), for example, following others (e.g., Kindgon 1984; Baumgartner and Jones 1993), terms anyone fiwho seek[s] dynamic pol icy changefl as a policy entrepreneur. I see this label as too general (and encompassing of policy actors) and distinguish between first -movers and followers. Perhaps a more appropriate term for my purposes here might be fipolicy inventors,fl although I hesit ate to use such a label as there is still a distinction between those who invent solutions (e.g., think tanks, academics) and those who are the first to advance them in the political arena. 82 (Polsby 1985: 172) . I designate policy actors a s those ind ividuals or groups within and outside government that might follow the lead of entrepreneurs to advocate for the same policy in other governmental jurisdictions. The critical distinction is that policy entrepreneurs innovate and lead; policy actors follow. Policy entrepreneurs present new solutions ; policy actors seek out solutions that have been tried elsewhere but are unique to their governmental unit. Why is this distinction between policy entrepreneurs and policy actors important? Because policy entrepreneurs play a crucial role in strategizing, taking risks, building support, rebuffing opposition, and spurring policy innovation (Kingdon 1984; Polsby 1 985). Policy entrepreneurs also interact with and influence policy actors in other states to bring about change in subsequent jurisdictions (Cobb and Elder 1983). The distinction also matters because much like the decisionmaking behavior of early -adopter, early -middle adopter, late -middle adopter, and laggard adopter states is different from the innovator states (Walker 1969; Gray 1973), so too entrepreneurs™ venue shopping process should differ from followers™ processes to pick a venue. Second, like other scholars (Lubell, Henry, and McCoy 2010; Lubell 2013; Pralle 2003), I assume those individuals and organizations engaging in venue shopping have limited cognition, limited time, and limited resources. They are strategic and tactical actors, but still boun dedly rational advocates. Information about policy actors™ capacities, opponents™ assets , and venue accessibility is not always available, incomplete, or too abundant to process. In turn, the pursuit of policy change via the venue that offers the greatest return on investment is based on these bounded beliefs and limitations. Described by Lubell (2013: 546): fi[D]ue to cognitive constraints, it is costly for actors to expand their behavioral repertoire to adjust to a new institutional setting. Thus, actors w ill not optimize their decision making across the ensemble of institutions in which they participate. Instead, actors will develop a series of simplified heuristics that they use to choose the [institutional venue] in which they participate, and how to mak e decisions within policy institutions of different types.fl 83 Indeed, the decision to pursue one venue over another is not always so straightforward, as evidenced by the fact that some actors fail in the process. Much like scholars™ conceptualization of venue shopping, Weyland (2005, 200 6) emphasizes the bounded rationality of decision makers in pursuing innovations. Analyzing the diffusion of the Chilean pension model throughout Latin America, Weyland argues that policymakers depend on cognitive heurist ics and shortcuts when weighing different policy solutions. These include the representativeness heuristic (i.e., relying on perceived initial success), the availability heuristic (i.e., looking to nearby examples), and the anchoring heuristic (i.e., relyi ng on the fact that adopted elsewhere). Bounded rationality is central to venue shopping (Ley and Weber 2015; Lubell 2013; Pralle 2003) and policy diffusion (Gilardi 2010; Mooney 2001; Weyland 2005, 200 6). Third, I propose that external political learning influences the venue shopping process when policy actors consider adopting new ideas previously pursued by policy entrepreneurs and policy actors in other jurisdictions. Pralle (2003) and Ley and Weber (2015) point to an internal learning process about fu ture decisions on picking a forum . Pralle acknowledges the existence of a positive feedback loop that keeps individuals and actors invested in the same venue, while Ley and Weber (2015) suggest that policy promoters learn from their failures in a venue. Bu t it is also possible that policy actors learn about the venue shopping done by policy entrepreneurs and actors in other governmental jurisdictions, fulfilling an external learning process. Here, I argue that policy actors learn from the venue shopping pre viously done by policy entrepreneurs and other policy actors. In addition to learning from their own experiences with site selection, policy actors also learn from others™ venue shopping. Collectively, relying on these three frames of venue choice, as wel l as the prior venue -shopping and policy -diffusion literatures, I theorize that a government™s choice of venue to pursue an innovation is influenced by the pr eceding venue shopping of other governments previously 84 pursuing the new policy. If policy diffusio n implies that the fiprior adoption of a trait or practice in a population alters the probability of adoption for the remaining non -adoptersfl (Strang 1991, 325), then I posit that the prior selection of a forum to pursue a new idea alters the likelihood of selecting the same venue for the remaining non -adopters, a process I call venue diffusion . I charge that the mechanism driving venue diffusion is political learning, whereby policy actors not only learn about a new solution to a current problem from other jurisdictions (i.e., policy learning) but also gain information about the political processes employed (i.e., political learning) to bring about change in the innovative jurisdiction. Articulated by May (1992: 340): fiPolitical learning is concerned with l essons about maneuvering within and manipulation of policy processes in order to advance an idea or problem.fl One of the key political processes that policy actors learn about from the policy entrepreneur in the innovative jurisdiction is venue choice. Dep ending on a state™s institutional arrangements, policy actors may follow their predispositions to attempt policy change or learn from the successful paths being taken in other states, especially those with similar institutional settings. Like Figure 4.1 a bove, Figure 4.2 again illustrates the extensive (although not exhaustive) list of considerations that policy actors might contemplate when shopping for an institutional venue. I add to this list at least one more factor that has been overlooked Špolitical learn ing: the drawing of lessons from prior policy actors™ success rate in a given venue. 85 Figure 4.2: Policy Actors™ Venue Shopping Considerations Also Includes fiPolitical Learningfl How might political learning drive venue diffusion? As described in previous sections, policy actors within governmental jurisdictions facing societal problems can either look inward or outward for possible solutions. Policy actors can undoubtedly learn fr om their own experiences or vicariously through the experiences of others within their governmental unit (Volden, Ting, and Carpenter 2008). However, policy actors can also learn from policy entrepreneurs in innovative jurisdictions or other policy actors outside their jurisdictions that have previously taken action on possible solutions. Policy entrepreneurs, those innovative agents within or outside government, propose and advance new policy solutions for current societal problems (Baumgartner and Jones 1 993; Elder and Cobb 1984; Kingdom 1984; Mintrom 1997; Polsby 1985). fiThey may be motivated by personal convictions, ideological zeal, the imperatives of office, or simply self -promotion. In any case, the[y] [sic] often play a critical role in mustering sup port and sheparding [sic] new issues and ideas to the governmental agendafl (Elder and Cobb 1984: 122). Counting on their expertise and relying on their connections within the political system, policy entrepreneurs expend energy, resources, and time to res earch the problem and design a solution (Elder and Cobb 1984; Kingdon 1984). They also spend appreciable effort determining the 86 most practical political path to enact and implement their new idea. They ruminate over how to define and frame the problem and solution, how to get a spot on the agenda, how to mobilize a coalition of support, how to shape the terms of debate, and how to counter and defeat opponents. And policy entrepreneurs tactically pick the institutional venue in which they believe they have a comparative political and resource advantage, is most accessible and amenable to the policy image, and has the best chance to deliver the win and ensure policy longevity. Importantly, entrepreneurs consider the full set of institutional venues available t o them to press for policy change. While policy entrepreneurs also suffer from bounded rationality, they are the first to put forward a new policy and press to upend the status quo . In doing so, they expend considerably more effort to gain information, s tudy the problem and develop a solution, make contacts and build coalitions, craft arguments in support of the idea, strategize on how to rebuff challengers, and decide the appropriate venue to achieve the policy innovation. Policy entrepreneurs™ fiboundsfl are not as tight as they are for subsequent policy actors. They engage less in satisficing and more in strategizing. In contrast, p olicy actors, given their limited cognition and resources (e.g., time, financial and political capital, political access), en gage in satisficing to learn about new policy ideas, the policy idea™s success, and the political viability of the innovation ( Holyoke, Brown, and Henig 2012; May 1992; Mooney 2001; Seljan and Weller 2011; Workman et al. 2009). T hese organizations and indi viduals rely on filesson -drawing,fl asking themselves fiunder what circumstances and to what extent would a programme now in effect elsewhere also work here?fl (Rose 1991: 4). Beyond the filesson -drawingfl regarding the policy and its effect, these followers als o learn about the successful and failed political processes and tactics employed to bring about policy change in the innovative jurisdictions. And it is part of the pre -contemplation and knowledge -gathering stage articulated by Rogers (1962). 87 Seljan and Weller (2011) provide a contemporary example. In addition to emphasizing diffusion of state tax and expenditure limits via direct democracy, the researchers also explicitly model political viability. Rather than lumping together policy and political learni ng, they disentangle the political viability of a given policy from the diffusion process by assessing whether the policy failure of some states affected neighboring states™ decision to pursue the policy. They find that, indeed, states with neighbors who h ad pursued TELs and failed were far less likely to try to adopt TELs. In short, political learning occurred, impacting the spread of policies across states. This further demonstrates that drawing lessons about the feasibility and political processes to ado pt and implement a policy are as crucial as learning about the policy itself. Recent work by Nicholson -Crotty and Carley (2015) suggests that policy learning is more than Jurisdiction B simply asking fiwas that policy effective in Jurisdiction A,fl but also, fican we make that policy work for us?fl Policy actors within states are not only concerned with policy outcomes but also with if and how they are best able to implement the policy. This is political learning. The political learning around venue shopping is perhaps best articulated by May (1992: 339): fiThe prima facie evidence for political learning consists of policy advocates™ change in political strategy. They may shift arenas for their advocacy from one committee in Congress to another, among branches of government, or among levels of government. They may make strategic use of litigation to call attention to a problem or force decisions. They may emphasise policy arguments that have proven to be more successful in mobilising attention. Or, they may try ou t new tactics in using the media, mass protests, letter -writing and so on to call attention to a problem or policy proposal.fl Policy actors may learn about the success of the policy, successful frames of the problem and solution, successful venues for ado ption, successful strategies for implementation, and electoral and political consequences of these events (Fetner 2008; Heclo 1974; May 1992; Sabatier 19 88). Rather than gather information about and strategize regarding the return on investment for all ava ilable venues within a state™s set of venues, policy actors take venue choice cues from policy 88 entrepreneurs and other actors that have previously attempted an innovation. They do not need to rehash the complete venue shopping process (just like they do no t need to invent their own solution) because it was previously done for them by prior adopters. Much like policy learning reduces uncertainty about the innovation, political learning helps pare down overabundant information or fill the information gap abou t which venue may be fibestfl to bring the innovation to the governmental marketplace. The pursuit of policy change is not a one -off source of learning, but rather an eternal spring of policy and political information. Political learning may be especially m anifest among state - and national -level interest and advocacy groups. Individuals and groups working within a policy network established within one or across many states may be better situated to communicate and share both policy and political successes. T o be sure, past diffusion scholarship has demonstrated persuasively that organized groups play a crucial role in spreading policy ideas. For example, Mintrom and colleagues relied on surveys of interest groups to show how policy actors work within inter - and intra -state networks to press for the adoption of education policies across states and localities (Mintrom 1997; Mintrom and Vergari 1998). 34 Hai der -Markel (2001) finds that interest group campaigns aided the diff usion of gay -marriage bans. Balla (2001) suggests that states whose insurance commissioner participated in a national -level committee were more likely to adopt a policy innovation. Moreover, Garrett and Jansa (2015) theorize and offer strong evidence that interest and advocacy groups not only affect policy change within a state but also contribute a complex network of information , including model legislation, that facilitates the diffusion of policy ideas across states. Given the role of state - and nationa l-level organized interest and advocacy groups in the diffusion of new ideas (e.g., Balla 2001; Garret and Jansa 2015; Haider -Markel 2001 a; Karch 2007 a; 34 Nongovernmental organizations and non -state actors can also drive the diffusion of new ideas across transnational networks as well; see True and Mintrom 2001 for one example. 89 Mintrom 1997; Mi ntrom and Vergari 1998 ; Shipan and Volden 2006; Stone 2012 ), it also seems plausible th at such groups could reduce the information or resource costs for affiliated policy actors in choosing an institutional venue. fiEpistemic communities organized around a particular policy area, sharing principled and causal beliefs, can profoundly influence policy diffusion, in part by facilitating learningfl (Graham, Shipan, and Volden 2013). And fi[a]t some level of aggregation, organizations face the same limits to attention as individual decision makers dofl (Workman et al. 2009). Hence, networks of interes t groups may be especially prone to facilitating political learning for the diffusion of venues, playing a more significant role in the diffusion process than understood initially . Of course, political learning does not imply indifferent, blind copying of venue choice by policy actors within a state. As Givan et al. (2010: 2) put it: fiDiffusion–does not simply mean that tactics or frames are transplanted in whole cloth from one site to another; creative borrowing, adaptation, and political learning are ofte n vital to its success.fl Just as fipolicymakers are not agnostic with respect to where they search for [policy] information,fl they are also not indifferent to where they obtain political information (Parinandi 2013: 245). Policy actors turn to their peers f or policy and political information: geographic neighbors, states with similar institutional arrangements and constraints, governmental units with similar ideological or political environments, etc. Communication between and among entrepreneurs and actors is central to this story (Rose 1991). Those motivated to pursue change may communicate with and receive information from multiple sources, including coworkers within agencies, between decision makers at professional conferences, from various correspondence s and publications, the media, interest groups, academics, concerned citizens, among others. 35 35 There is abundant anecdotal support for this theory. At a recent roundtable, Nancy Wang and Amelia Quilon, organizers for fiVoters not Poli ticians,fl told the story of why they decided to press for an end to gerrymandering in Michigan via direct democracy. Early organizers were familiar with the ballot campaign to end legislative redistricting in 90 Nor does this proposition imply that policy actors will never deviate from the venue paths previously pursued by other actors. It is not that policy actors are n ever aware of and will never contemplate resources, opponents, venue accessibility, and venue amiability to the policy image, among other considerations. Different venues are more or less receptive to the type of resources possessed by a group (Sabatier an d Jenkins -Smith 1999: 143). Indeed, not all forums are available in all the states, and institutional hurdles make some venues more feasible than others (e.g., Boehmke and Patty 2007). Undeniably, policy actors partake in some degree of deliberation and st rategy and possibly follow the lead of other states with similar institutional settings. But policy actors will strategize to a lesser degree (and satisfice more) than policy entrepreneurs and other advocates who have previously pursued the innovation. Whi le considering their resources and capacities, their opponents™ resources and capacities, political and institutional hurdles, venue accessibility, among other factors, change agents will also weigh the paths previously taken by others. Also, somewhat cha llenging the proposition that policy actors attempt policy change in as many available venues as possible (Boehmke, Gailmard, and Patty 2013; Holyoke 2003; Jourdain, Hug, and Varone 2017 ; Sabatier and Jenkins -Smith 1993), my proposition implies that actors engage in satisficing and select the most appropriate venue(s) for enactment. 36 Policy entrepreneurs and actors are not akin to consumers on a shopping spree with unlimited resources. Instead, these change agents are careful consumers on a resource budget. Nor are policy actors like novice archers Ohio and executed a fifty -state survey to learn about successful policies and campaigns around the country. Following the survey, they decided to emulate the citizen redistricting commissions established via ballot initiatives in California and Arizona. (Quilon and Wang 2019). At the same roundtable, S am Pernick, the organizing director for MI Legalize, the state group promoting the legalization of recreational marijuana, attributed the decision to press for policy change via plebiscite was due to state political factors, prior success with medical mari juana at the ballot box in Michigan in 2008, and seeing the successful recreational cannabis ballot measures in other states (Pernick 2019). 36 Notably, the number of venues an individual, group, or policy network involves themselves may vary depending on whether these actors are trying to dominate an issue area or to pursue an innovation. Lobbying activity can involve more than just pressing for policy change. In turn, other scholars™ proposition that individuals and groups target as many venues as possible , and my proposition that they narrow their focus to push for change in one or a few venues (rather than the complete set) may both hold depending on whether we are discussing general lobbying activity or the pursuit of a new idea. This is yet another area ripe for additional investigation. 91 shooting as many arrows as possible with the hope that one will land in the bullseye. They do not pursue a policy in all of the venues with the promise of success in one out of many attempts. Seeking policy change is costly (Buffardi, Pekkanen, and Rathgeb Smith 2014); it requires knowledge of the political environment and institutions, mobilized support and relationship building, financial assets, and time, among other resources. In turn, change agents turn to and learn from the political processes of those who have previously pursued or enacted the policy. Indeed , selecting a particular venue for policy change may be more experimental than exact (John 1999; Pralle 2003). This is likely to be the case, especially f or policy entrepreneurs as they are the first to push for the adoption of a new policy. But, the uncertainty of which venue that early -, mid -, and late -adopters ought to pursue should decline precipitously as the successful adoption rate of a particular po licy in one or more venues increases. Much like the effect of policy learning is greatest for the first few states adopting the innovation with fia smaller added value for each additional adoptionfl ( Makse and Volden 2011: 117), the impact of political learning on venue choice should also be most influential for the early adopters, decreasing as more states successfully achieve policy change via a given path(s). This is learning in practice (Freeman 2008). Given this, I advance the following hypothesis for a political learning mechanism that drives venue diffusion: Political Learning Hypothesis: The likelihood of a state p icking a venue to pursue a policy increases as the proportion of other states successf ully pursuing the same policy via the corresponding venue increases . Undoubtedly, the inverse of this hypothesized relationship is also possible. Political learning can include both positive and negative signals. As a result, as the proportion of other states successfully pursuing a policy via a given venue decreases over t ime, implying failure via that venue, the likelihood that a state also pursues the policy via that same venue should also decrease. Further, political learning™s effect may vary over time. We should not expect a linear relationship between political learni ng and choice of venue over the lifetime of the policy, especially if success ebbs and 92 flows by arena over time. Since policy actors are known to be fiadaptive venue shoppersfl (Ley and Weber 2015), we should anticipate that they process these external signa ls of success and failure and update their choice of venue in real time. Also of particular note, this hypothesis and corresponding measure emphasize fisuccess.fl As a metric, accounting for success is vital for two reasons. First, the successful pursu it of a policy via a particular venue is the clearest signal that the path chosen worked. Moreover, it is a consistent signal across institutional arenas regardless of variation in venue type. Second, a critique leveled against several of the measures oper ationalized for learning is that they fail to directly factor in success (Gilardi 201 6; Volden, Ting, and Carpenter 2008). 37 Thus, my operationalization of political learning, discussed in more detail in the following empirical chapters, heeds this criticis m and directly accounts for the cumulative success and failure of states pursuing a policy via a given venue. Alternative External Factors Driving Venue Choice There are, however, plausible alternative explanations for the diffusion of venue selection. Many of the mechanisms that drive policy diffusion may also contribute to venue diffusion. Policy actors within a state are likely to turn to states they perceive as fileadersfl or as fipeersfl for policy solutions. Similarly for venue diffusion, the notion of fi peerfl can take on a variety of forms, including actors emulating the venue choice of states that are alike along economic, social, political, cultural, or institutional dimensions , geographically proximate , or exhibit shared preferences. Given this, per haps policy actors mimic the choice of venue of pr evious adopting states with similar institutional arrangements or political contexts. Since the set of venue options varies across states, it seems appropriate that if policy actors look externally for poli cy ideas, gaining both policy 37 Of course, fisuccessfl can take on different meanings and can be operationalized in different ways. Volden (2006), for example, measures policy success as the degree to which the health insurance amendment lowered the un insured rate among poor children. Success can also be operationalized as electoral retention (Gilardi 2010) or as a quantification of the policy adoption™s impact Gilardi (2015). 93 and political information about the processes to enact those ideas, they would especially look to states with analogous institutional settings or political environments. For example, in considering pursuing a constitutional am endment, a fidirect -democracyfl state will likely look to other fidirect -democracyfl states to emulate their path of enactment, rather than look to states that require multiple sessions, constitutional conventions, or voter supermajority for ways to enshrine t he policy. Policy actors might also look to their fipeerfl states with similar ideological predispositio ns ( Butler et al. 2015; Butler and Pereira 2018; Desmarais, Harden, and Boehmke 2015; Volden 2015; Zelizer 2019); degrees of legislative professionalism ( Shipan and Volden 2006, 2014; Volden 2015 ), judicial professionalism (Squire 2008), difficulty in amending the state co nstitution ( Dinan 2018; Fay and Wenger 2015; Lupia et al. 2010), or another institutional or political attribute. I offer the foll owing hypothesis: Institutional / Political Similarities Hypothesis: The likelihood of a state picking a venue to pursue a policy increases as more institutionally and politically similar states opting for the same venue increases. Another possibility is that policy actors™ choice of venue is the result of a geographic phenomenon (Berry and Berry 1990 ; Berry and Baybeck 2 005; Cohen -Vogel and Ingle. 2007; Walker 1969 ). Policy promoters may copy the venue choice of contiguous neighboring states. Policy actors in states may look to their neighbors for new ideas and follow their neighbors™ choice of venue to pursue the idea. In fact, fi[s] tate policymakers and citizens look to other states in a satisficing search for solutions to problems, and the states to which they look first are their neighbors, due to familiarity, ease of communication, cross -mixing of media and population, and common valuesfl (Mooney 2001: 105). Such a search may also yield political information about the fibestfl venue to pursue policy change. Due to the potential for a regional clustering effect of venue shopping, I advance the following alternative hypothes is: 94 Geographic Neighbor Hypothesis: The likelihood of a state p icking a venue to pursue a policy increases as the proportion of contiguous neighboring states picking the same venue to pursue the policy increases. Importantly, I expect the influence of g eographic neighbors on venue shopping to be most substantial when neighbors share the same set of venue options. Policy actors™ choice of venue within a finon -direct democracy statefl may be influenced if neighboring states press for change via the legislatu re or legislative referenda. But this is less likely if direct democracy is the path for enactment. Moreover, while a regional clustering effect of venue selection is possible, Karch and colleagues™ (2016) recent article on the fipro -innovation biasfl in dif fusion research suggest that scholars may be overstating the existence of geographic diffusion. Examining the adoption of numerous interstate compacts by a handful of states to a plurality of states, they find that by only focusing on and modeling innovati ons that gain large traction may cause us to underestimate the role of learning and professional associations working across jurisdictions (Karch et al. 2016). Given these caveats, the impact of learning, both political and policy, should overshadow any geographic effect. Paralleling the policy learning mechanism in policy diffusion, actors and interest groups might select a venue simply because sever al other states have gone that route. Instead of actively weighing early movers™ success rate in a given forum (i.e., political learning) or looking to peer states, policy actors may copy the policy solution and most popular path taken by others. Policy learning and political learning are two different, albeit related, processes (May 1992; Mooney 2001; Rose 1991; Seljan and Weller 2011). Yet deconstructing these two concepts theoretically is a much easier task than parsing them empirically. While political learning encompasses the receipt of information about the success or failure of political strategies (e.g., venue selection) to pursue a policy solution, policy learning covers information about the societal problem and policy idea. A lthough I anticipate that learning about the number of other states going a route may have a positive effect picking a 95 path, it is also plausible for policy learning to no t affect venue choice. In fact, there may be occasions where political learning exists but policy learning does not. For example, where past research has failed to uncover policy learning as a mechanism driving diffusion Šperhaps due to a focusing event, punctuation (Boushey 2010), national attention, simplicity of a policy ( Nicholson -Crotty 2009), or another at tribute Špolitical learning may still occur with regard to venue selection and other processes even in the absence of policy learning. In effect, policy learning can act as a control to disentangle the se two related processes . With this in mind, I propose t he following hypothesis: Policy Learning Hypothesis : The likelihood of a state picking a venue to pursue a policy increases as the number of other states picking that venue to pursue the policy increases. The federal government can often encou rage or discourage policy adoption across states (e.g., Allen, Pettus, Haider -Markel 2004 ; Karch 2009, 2012; Shipan and Volden 2006, 2008; Welch and Thompson 1980). Interdependence is not only horizontal but also vertical. Accounting for the American feder ated structure, we would expect state -level activity to influence federal action (see Karch and Rosenthal 2015; Lowery, Gray, and Baumgartner 2011) and federal -level activity to impact policymaking in the states. Multiple scholars have demonstrated that po licy debates and policymaking in the national arena influence activity in the states. Karch (2006) finds that federal -level intervention on individual development accounts, family caps on welfare, and medical savings accounts had varying impact on states™ initiatives with these policies. Similarly, Karch (2012) shows that a nationally televised address by President George W. Bush on stem cell research, in conjunction with a broader debate over stem cell research legislation, increased the likelihood that states would act on the issue. Looking at partial -birth abortion bills, truth -in-sentencing laws, and hate crime legislation, Allen, Pettus, Haider -Markel (2004) show that federal incentives and penalties drive or discourage state -level activity. McCann, Shi pan, and Volden (2015) have offered compelling 96 evidence that even policy ideas not yet enacted at the national level can percolate down to the states, especially among states with legislative capacity and ardent interest group activity. Ultimately, t he federal government can signal to the states its preferences and potential for future national action. Such activity Še.g., congressional bill, presidential executive order, federal agency decision, Supreme Court ruling Šmay also affect the choice of venue wh ere states pursue innovations. While I do not predict the directionality of influence, it is possible that a federal bureaucratic regulation may cause state legislators to act, congressional legislation may inspire interest groups to pursue state constitut ional amendments, or a U.S. Supreme Court opinion may prompt a new plaintiff to seek change at their state high court. These example federal -level interventions can affect the diffusion of the policy as well as the choice of path to pursue that policy. In turn, I propose the following hypothesis to account for federal -level activities that may influence policy actors™ process of venue shopping. Federal Intervention Hypothesis: The likelihood of a state p icking a venue to pursue a policy increases / decreases as the federal government intervenes in the issue area. Related is the effect of the national political environment. Outside the scope of actual federal -government activity, national political fo rces can also influence the adoption of policies and may lead policy actors to pursue one venue over another. Several researchers have shown that the salience of an issue or policy can hasten action across multiple governmental units (Boushey 2010; Makse and Volden 2011; Nicholson -Crotty 2009). Indeed, a crisis or focusing event can force an issue onto the policy agenda. Such punctuations depart from the incremental change and learning that generally characterize the policymaking process (Baumgartner and Jo nes 1993; Boushey 2010). Moreover, the timing of certain national events Še.g., presidential elections Šmay also prompt policy change (Berry and Berry 1990, 1992; Mintrom and Vergari 1998) and the picking of one venue over another. Smith, DeSantis, and Kass el (2006), for example, investigated whether the 97 slate of anti -gay marriage ballot initiatives across states for the November 2004 presidential election was timed to increase voter turnout. While the authors found no evidence of heightened participation attributable to the ballot measures, accounting for national elections, events, and environmental context is critical to understanding another conceivable external factor influencing venue diffusion. I submit the following hypothesis: National Environment Hypothesis: The likelihood of a state picking a venue to pursue a policy increases / decreases as the national environment on the issue area ebbs and flows. In particular, I anticipate that greater prominence of an issue will cause policymakers in state legislatures to act. Since elected officials are electorally motivated (Mayhew 1974) and salient issues make it on the public agenda, I expect that state legis latures are the most obvious and least institutionally constrained venue to press for action. While issue salience should influence the pursuit of change via state legislatures, it should have less impact on the other institutional arenas. This may be due to greater institutional hurdles (e.g., signature requirements for a ballot initiative, litigant pressing forward a legal case to state high court) (Lupia et al. 2010; Lutz 1994) or fewer incentives for actors in those venues to address pressing issues. Internal Factors Driving Venue Choice These factors above Špolitical learning, institutional and political similarities, geographic neighbor, policy learning, federal intervention, and the national political environment Šare the external forces that could dictate policy actors™ venue shopping process. These are the external factors that have often been ignored by the venue choice literature because venue shopping has mostly been treated as an internal process. However, venue choice may indeed be an internal choice, as current scholarship implies. As such, in addition to accounting for these external forces, I will also 98 control for key internal determinants of venue choice, including state -level political, economic, institutional, demographic, and inter est -group factors. Important political factors within a state that could drive venue choice include the competitiveness of elections, party control of the governing bodies, the ideological predispositions of officials, or public opinion regarding the poli cy issue, among others. Close races between legislators may compel them to pursue a policy innovation in the legislature or ask the state electorate to vote directly on the issue (Barrilleaux, Holbrook, and Langer 2002; Holbrook and Van Dunk 1993) . Control of the state legislature or governor™s office by one political party over another may force out -group policy actors to pursue a more amenable path (Calvert et al. 1989). Indeed, Hinchliffe and Lee (201 6) find that increased party competition (i.e., greate r structure in roll call voting by party members and a decline in ficrossing -the -aislefl) of state legislatures to control governing institutions has contributed to greater political polarization at the state -level. Increased polarization can lead to gridloc k in the legislature ( Hetherington and Rudolph 2015; Shor and McCarty 2011), and may cause policy actors to press for change in alternative venues. Equivalently, in picking an institutional setting, policy advocates may also consider the ideological direct ion of venue actors, and whether their liberal or conservative predispositions will make them more responsive to the issue (Butler et al. 2015; Desmarais, Harden, and Boehmke 2015; Holyoke, Brown, and Henig 2012 ; Volden 2015) . Finally, policy actors may al so factor in where the state™s citizens stand on the policy issue. Public opinion may not only spur the adoption of an innovation (Enns and Koch 2013; Erikson, Wright and McIver 1993; Pacheco 2012; Wright, Erikson, and McIver 1987 ), but also influence the choice of venue (Baumgartner and Jones 1993; Kingdon 1984). Equally plausible is that a state™s own institutional arrangements will dictate venue choice. Rather than paying attention to institutional similarities with earlier adopters, policy actors within a state may consider their institutional hurdles and settings. Evidence abounds that institutions affect 99 policy diffusion ( Füglister 2012; Gilardi and Wasserfallen 201 9) and policymaking. For example, Lewis, Schneider, and Jacoby (2015) effe ctively demonstrate that institutional characteristics directly influence state policy outputs. They find that the finet effectfl of five institutional components Špower of the state house speaker, legislative professionalism, governor™s control over state bu dget, and term limitedness of legislators and governor Šcan move states to consider spending more on collective goods (e.g., education, transportation, natural resources, public safety) or particularized benefits (e.g., healthcare, welfare, corrections). A state™s institutional characteristics should also affect venue shopping. A more professionalized legislative body, equipping policymakers with the capacity and resources to understand societal problems and propose adequate solutions, can spur the adoption of innovations (Shipan and Volden 2006; Shipan and Volden 2014; Volden 2015). Greater legislative expertise may persuade or dissuade policy actors from pressing for change via the state legislature. And it may lead to competition for power between differe nt venues (Dilger, Krause, Moffett 1995; Miller, Ringsmuth, and Little 2015). Likewise, the professionalism of the state™s highest court may also condition the venue choice of policy actors. Yates, Tankersley, and Brace (2010) show that in more liberal -policy states, the greater judicial professionalism and accountability to citizens (through elections), the higher the uncertainty of outcomes and the more likely citizens will seek out arbitration through the judiciary. Effectively, the institutional structu re and professionalism of a state™s court send a signal to litigants of their chances of winning at trial. Lastly, the degree of difficulty in pursuing an amendment to the state™s constitution may also influence a policy actor™s venue selection (Fay and Wenger 2015; Lupia et al. 2010; Lutz 1994). The more hurdles to achieve an amendment (e.g., ranging from the number of signatures required for the petition to a supermajority of support from the voters) may encourage or discourage advocates. In this story 100 about the importance of picking among insti tutional arrangements, an essential part of the narrative is considering the settings within one™s state. Policy actors may also weigh internal economic and demographic conditions when deciding on the right venue. As Peterson (1995: 90) put it, fi[o]ne s hould not ignore the political meaning hidden in demographic and economic variables. For example, the taxable resources of a state are not simply an economic factor–the variable also measures the public™s demand for public servicesfl (as quoted in Schneider and Jacoby 2014). Socioeconomic factors may serve as a proxy for citizens™ policy preferences (Dye 1966; Hofferbert 1974). Measures of state wealth and resources correlate highly with more innovative states (Walker 1969). Furthermore, a wealthier and more educated polity tends to be more engaged in the political system (Leighl ey and Nagler 2013; Verba , Schlozman, and Brady 1995). Racial and ethnic diversity may also matter (Hero and Tolbert 1996). Nicholson -Crotty (2006) shows that diversity within a state helps legislators to be more representative of the citizenry and rebuff direct democracy (i.e., the tyranny of the majority). These types of economic and demographic variables may steer policy actors to choose one venue over another. Not least of the pot ential internal forces on venue selection is the role of state - and local -level interest groups (Gray and Lowery 1996). In selecting a venue, interest groups weigh their resources and capacities, as well as their opponents™ resources and capacities ( Holyok e, Brown, and Henig 2012; Ley and Weber 2015; Pralle 2003; Sabatier and Jenkins -Smith 1993). They also consider their competitive political, legal, and technical advantages relative to their opponents™ strengths (Ley and Weber 2015; Sabatier and Jenkins -Smith 1993; Jenkins -Smith et al. 2014). Current research also suggests that fidirect -democracyfl states will witness an increase in the number of organized groups as well as higher volatility in the groups™ entry into and exit from the political arena (Boehmke 2002, 101 2005, 2008). 38 Therefore, I will control for state -level interest groups on both sides of a policy issue, accounting for the capacity, resources, or membership size of both friends and foes. Beyond merely capturing pressure groups™ presence and stre ngth, however, I also account for a movement™s past success in a venue, as well as the countermovement™s past success. We know from the venue shopping literature that policy actors prefer venues that they are familiar with and already engaged in (Holyok e et al. 2012). It is reasonable for policy actors and interest groups to stick with the venue they know, especially if they have been previously successful. As a result, if policy actors or organized interests have already pursued and achieved policy change in one venue, they should be less likely to select another arena. Of course, groups and actors must also weigh their opponents™ policy successes in different venues. (Holyoke, Brown, and Henig 2012; Ley and Weber 2015). A movement™s success in one venue ma y force the countermovement to compete in a different venue. Thus, I account for both actors™ prior policy successes in a venue and opponents™ policy successes. Collectively, there is a host of internal and external forces that drive the venue -shoppin g process. Figure 4.3 below illustrates the principal competing and complementary stimuli that policy actors weigh in deciding to take one path over another. My central argument is policy actors, given their limited cognition, resources, and information , engage in satisficing, looking to entrepreneurs and actors in other states that have already pursued the solution to a common societal problem. These policy actors not only learn about the policy but also the political processes taken to enact the 38 Boehmke™s argument is that ballot initiatives provide an additional route for interest groups to influence p ublic policy. As a result, state legislators Šout of concern that the policy will be too far from their ideal point or result in punishment at the polls for not supporting the policy Šwill vote in line with the median voter , thus mitigating the need for a citizen initiative (Gerber 1996). And since the opportunit ies for accomplishing one™s goals are more plentiful in fidirect -democracy fl states, Boehmke contends that this will produce a greater number of interest groups in these states , especially citizen organizations championing and advocating on behalf of underrepresented groups. Moreover, the initiative process allows citizen groups to mobilize quickly around an issue, bringing it to the voters; if they fail, the group may fa de away, or they may disband if the issue is approved and no longer salient. This contributes to heightened entry and exit volatility for groups in these states. 102 policy in the other governmental units. Although policy actors weigh their own political, institutional, economic, demographic, policy, and interest -group factors, along with competing external forces, they are also influenced by the successful routes previously tr od by others. Figure 4.3: External and Internal State -Level Forces Influencing Venue Choice Implications of Venue Diffusion Evidence of venue diffusion for a policy innovation across U.S. states is of theoretical import for at least five reasons. First, exploring patterns of venue choice goes beyond traditional diffusion research, which tests if a policy spreads. Here, I am int erested in seeing if the choice of venue in one state to attempt a policy influences the venue shopping process in subsequent states for the same policy. Delving further into the dynamics of diffusion is exactly what several leading scholars have been clam oring for (Berry and Berry 2014; Gilardi 201 6; Graham, Shipan, and Volden 2013; 103 Howlett and Rayner 2008; Shipan and Volden 2012). This research attempts to answer these calls by fusing our knowledge of policy transfer with studies emphasizing venue shoppin g. Second, t he overwhelming focus of the diffusion scholarship on the transfer of policies from one legislative body to another legislative body does not square with the reality of policy activity in a complex system of multiple institutions. E vidence fo r the diffusion of venue selection could elucidate a fundamental shift from conventional policymaking via state legislatures to policy activity in alternative institutional venues. This shift may only be evident for some types of policies or may be depende nt on certain policy attributes. Nonetheless, t he patterns , locations, and speed in which policies are adopted across states may be changing. Moreover, these changes could have important implications for democratic responsiveness and accountability. Legisl ators may increasingly shirk their responsibilities, citizens may progressively bypass their elected surrogates to vote directly on different policies , state supreme court justices may actively weigh in on public policy, or governors may increasingly push for more considerable influence over policy. The pursuit of a policy via specific forums may even enhance the public™s perception of the policy™s or venue™s legitimacy. Augmented interaction among the multiple layers of government may have an impact on policymaking and decision -making processes, as well as American democracy writ large. Third, the choice of venue not only has long -term consequences for policy outcomes (i.e., the rejection or longevity of the policy solution) but also short -term influences o n the policy™s design, implementation, and the constituencies developed around it (Boehmke, Gailmard, Patty 2006; Karch 2009; Ley and Weber 2015; Lubell 2013; Maltzman and Shipan 2008). Political institutions and processes condition policy outcomes and new policies make new politics (Schattschneider 1935). Hence, the diffusion of venue selection might similarly influence the design , implementation, evaluation, and survival of policies. This one decision has serious implications for policy existence and policy entrenchment. It is the product of prior decisions made in multiple 104 ‚policy games™ and will influence future decisions ( Boehmke, Gailmard, Patty 2006; Karch 2009; Lubell 2013). Rather than focus on one policy venue at a time, like the overwhelming majority of the current scholarship, I recognize and empirically account for the fact that fipolicies are the product of multiple decisions being made in multiple venues over long periods of timefl (Ley and Weber 2015: 705). Venue choice matters. And examini ng venue choice explores a key component of the policymaking process that may spur or stymie diffusion. Fourth, the integration of the policy diffusion and venue shopping literature sheds further light on venue shopping. Current venue shopping literature conceptualizes the choice of venue one of three ways: (1) as a matching exercise of the policy image to an amenable venue (Baumgartner and Jones 1993); (2) as a strategic selection where policy advocates desire a competitive advantage, seek ideological con gruence, or develop a long -term preference (Holyoke, Brown, and Henig 2012; Lubell 2013; Pralle 2003; Sabatier and Jenkins -Smith 1993); or (3) as a balance of their own political, legal, and technical strengths against their opponents and against venue acc essibility (Ley and Weber 2015). While some of this scholarship emphasizes internal learning by policy actors about their own past successful or failed venue choices, the literature ignores potential external learning about venue shopping. Evidence for ven ue diffusion would suggest that the choice of venue may be less strategic and insular than previously thought. Political learning from other policy entrepreneurs and actors that have already taken action may carry weight in making venue shopping decisions. Finally, i f there is evidence that the choice of venue diffuse s, then policy actors™ learning may not stop with policy solutions or possible paths, but may also extend to the other policy stages : policy design, policy winnowing, agenda setting, implemen tation, evaluation, the proposal of parallel policies, etc. Moreover, evidence for political learning driving venue selection would suggest that learning between policy actors does not stop with innovative solutions, but also extends to tactical 105 strategies about how best to bring those innovations to market. Our current understanding of learning may underestimate the extent to which state actors are interdependent. Ultimately, evidence of venue diffusion may address existing gaps in the diffusion scholarshi p; shed further light on policy diffusion processes and dynamics; offer additional insights for the venue -shopping literature; and raise new questions for policy change in the filaboratories of American democracyfl (Brandeis 1932 ). Moreover, the theory of ve nue diffusion in the U.S. state context may also be generalizable to other dyadic relationships within the U.S. (e.g., cities to cities) as well as the comparative context (e.g., nation -states to nation -states). Conclusion The operating theory here is that if a government™s decision to adopt an innovative policy is conditioned on prior governments™ decision to adopt the policy (i.e., policy diffusion), then a government™s choice of means to pursue a policy may also be influenced by the prior venue choic es of other governments pursuing the innovation (i.e., venue diffusion). I charge that venue diffusion is a function of political learning. I anticipate that those individuals or organizations searching for a policy solution will find both policy entrepren eurs™ innovation and choice of venue useful information as the proportion of states successfully pursuing a policy via the same arena accrues. As innovative states and early adopters follow a given successful path, remaining states will likely follow suit. I turn now to test this theory of venue diffusion , as well as political learning™s role in the spread of policy innovations and venue shopping processes in the subsequent chapters. 106 CHAPTER 5: POLITICAL LEARNING AND THE DIFFUSION O F GAY MARRIAGE POLICIES On June 26, 2015, U.S. Supreme Court Justice Anthony Kennedy writing for the 5 -4 majority in the Obergefell v. Hodges case penned that fiNo union is more profound that marriage, for it embodies the highest ideals of love, fidelity, devot ion, sacrifice, and family–[Same -sex couples] ask for equal dignity in the eyes of the law. The Constitution grants them that right.fl 39 The ruling obliged states to issue marriage licenses and guarantee benefits to gay couples under the 14 th Amendment™s equ al protection clause. And it deemed state statutes and amendments prohibiting same -sex marriage unconstitutional. Gay marriage was finally the law of the land. The Obergefell decision was the climax of a decades -long struggle for civil rights by LGBT advocates, on the one hand, and traditional family values by religious conservatives, on the other hand. Few public policies evoked such passionate debate about morality and equal ity, spurred such stark evolutions in opinion, and captivated such public attention as same -sex marriage. From the national government™s fisexuality regimefl and Lavender scare in the 1940s -1950s to root out federal workers and the 1969 Stonewall riots by LG BT individuals to collectively assert their right to be gay, to the AIDS epidemic and the passage of anti -discrimination and hate -crime legislation, the pursuit of LGBT rights under the law has been long and arduous. And the quest remains unfinished. 40 Yet the watershed moment in the timeline for gay rights came in 1993 when the Hawaiian Supreme Court remanded the Baehr v. Lewin case 41 involving three same -sex couples that were 39 Obergefell v. Hodges, 576 U.S. 28 (2015) 40 For a comprehensive narrative and review of the historical struggle for LGBT rights, see Hirshman (2012 ); Mucciaroni (2008 , 2011); Smith (2008 ); or Valelly (2012 ). For a jurisprudential overview of gay rights and marriage equality, as well as the legal protectio ns that are still necessary for LGBT individuals, see Engel (2016 ). To learn more about gay rights advocates™ current efforts, read Mezey (2017). Finally, for information about the struggle for same -sex marriage in other parts of the Western Hemisphere and world, see Pierceson, Piatti -Crocker, and Schulenberg (2010) and Pierceson (2013) . 41 Baehr v. Lewin , 74 Haw. 530, 852 P.2d 44 (1993) originally, although renamed Baehr v. Miike in 1996 when the Lawrence H. Miike became the new State Director of Health for Hawaii . 107 denied marriage licenses back to the trial court (Dorf and Tarrow 2014; Fetner 20 08; Gallagher and Bull 2001; Hollander and Patapan 2016; Hume 2011; Keck 2009; Lewis 2011; Pierceson 2013; Smith 2008; Stone 2012). Rather than denying the appeal, the justices called on the state to explain the compelling interest it had in restricting ma rriage only to heterosexual couples. The Baehr case was not the first state case where LGBT individuals sued to marry. 42 But the ruling was the first time a state court of last resort had left open the possibility of equal marriage rights for gays and lesbi ans. What ensued was a swift and tactical countermobilization against gay marriage by the religious right and conservatives via state legislatures and ballot initiatives. Evoking threats to America™s cultural and familial fabric and garnering broad public support, opponents were able to achieve statutory and constitutional bans on same -sex unions in two -thirds of the states by the mid -2000s. Congress even passed the Defense of Marriage Act (DOMA) in 1996, defin ing marriage for federal purposes as a union between a man and woman and allow ing states to deny same -sex unions performed in other jurisdictions. Meanwhile, gay marriage proponents continued to methodically pursue equal rights in state and federal courts and legislatures. LGBT groups achieved early wins for gay marriage in Vermont, Massachusetts, Connecticut, and Iowa, followed by a cascade of success in federal court following the Supreme Court™s 2013 decision in United States v. Windsor overturning DOMA. In the end, gay rights advocates prevailed. The movement for and countermovement against gay marriage offer prima facie evidence for my principal arguments that: (1) the diffusion dynamics of a policy vary when the innovation spreads across multiple ven ues; and (2) the venue choice to pursue a policy in one state influences the venue shopping process in subsequent states. This chapter is dedicated to disentangling the first claim. The next chapter tackles the second assertion. Recall from prior sections that policy scholars 42 Minnesota was the first state in 1971 to decide a suit from same -sex couples denied a marriage license. Legal challenges from gay couples followed in other states too in the 1970s and 1980s , includin g in Washington, Kentucky, Alaska, Florida, Hawaii, Illinois, Iowa, New Hampshire, South Dakota, and Utah (Ha ider -Markel 2001 b; Soule 2004 ). 108 have mainly mapped the patterns of policy diffusion in one venue: state legislatures. Even where researchers have modeled policy transmission in other venues (e.g., cities, courts, bureaucracies, nation -states), the focus has still bee n on one arena. Therefore, we know little about how new policy ideas propagate across competing institutional venues. You may also remember that prior research has identified policy learning , whereby policy actors facing too little or too much information learn about solutions already adopted in other jurisdictions, as a central mechanism driving the diffusion of innovations. Because of the myopic focus on a univariate avenue, little has been documented about the political learning that occurs in the polic y diffusion process. I contend that policy actors also draw lessons from the tactical choices policy entrepreneurs and early movers made to pursue a new idea and then rely on that information to make their own political choices. For example, a critical fac tor in pressing for policy change is the optimal institutional venue to achieve the desired results. I theorize that policy actors satisfice, learn from and follow the previous institutional paths successfully taken in other states. In short, as knowledge about the successful paths increases, following states should also be more likely to change policy via those venues. Relying on the policy case of gay marriage, I leverage the spread of anti - and pro -gay marriage policies across multiple state venues to unpack the diffusion dynamics of these policies across competing institutions. The empirical results establish political learning as a central predictor of states outlawing and legalizing gay marriage. Policy actors learning about the successful paths purs ued in other states increases a state ™s likelihood of prohibiting marriage equality via the legislature by 3.9 percentage points and via legislative referendum by 33.8 percentage points. Similarly, political learning raises a state ™s risk of allowing same -sex unions via the fipeople™s branchfl by 4.9 percentage points. Because a state ™s average risk of changing policy in any given year is under two percent, these marginal effects are both substantive ly large and meaningful. Political learning ™s 109 effect on poli cy change via citizen initiatives, state courts, and federal courts is less clear, but the empirical results point to a positive, but limited effect. Collectively, political learning ™s impact on policy adoption is more substantial compared to other known e xternal mechanisms driving policy diffusion, including policy learning, regional effects, federal government involvement, and the national environment. Beyond establishing political learning ™s role in the diffusion process, however, this chapter also make s the case for and employs a repeated -events, competing -risks multinomial logistic regression model to better test the external and internal factors affecting policy change within and across different venues. This modeling approach provides not only a supe rior fit of Event History data than standard logistic regression where multiple venues are involved, but also better maps the underlying policy process unfolding across competing venues. Ultimately, t he results clarify our understanding from prior scholarship on the diffusion of same -sex marriage bans, while also offering original results for the dissemination of pro -gay marriage policies. I provide evidence that policy actors do not restrain themselves to one venue but utilize the available avenues that will maximize the chances of policy success. To help determine the most favorable avenue, policy advocates consider their institutional arrangements, political contexts, interest group pressures, past policy activity on related topics, and even opponents ™ policy successes. But they also take external cues , especially political learning, into account. Rather than making these decisions in isolation, policy actors rely on outside information about the success of other state s to help achieve policy change in their own states. Why the Policy Case of Gay Marriage? The fight over gay marriage is especially illustrative of my broader theoretical claims for three reasons. First, to explore the diffusion dynamics of a policy in multiple venues, you need a 110 policy that has been pursued in more than one forum . Variation on the dependent variable is required (King, Keohane, and Verba 1994). The case of gay marriage provides such variation , whereby o pponents pressed for bans on same -sex unions by route of state legislatures, legislative referendums, and citizen initiatives. Proponents fought for marriage equality in state and federal courts, and state legislatures. Moreover, multiple policy changes occurred within states in different venues, with several jurisdictions adopting but later reversing a ban on same -sex marriage. 43 I can leverage the multidimensionality of this policy area to examine whether policy diffusion diverges depending on the venue in consideration. Second, the adopt ion of anti -gay -marriage policies ha s been extensively documented and empirically analyzed by many excellent scholars: Barclay and Fisher 2003, 2008; Camp 2008; Haider -Markel 200 0, 2001a, 2001b ; Hume 2011; Lewis 2011 ; Lupia et al. 2010; Soule 2004; Taylor et al. 2012 . These studies , however, narrowly focused on one venue (e.g., legislature, constitutional amendments, courts) ( Haider -Markel 2001 a, 2001b ; Hume 2011; Keck 2009; Lupia et al. 2010; Soule 2004) or failed to account for competing venues in their m odels (Barclay and Fisher 2003). Only Barclay and Fisher (2008) and Lewis (2011b) consider the spread of same -sex marriage bans via multiple arenas. But both stop short of concurrently modeling the pursuit of these policies via different institutional sett ings. Furthermore, none of the current articles explore or model the transmission of pro -same -sex marriage policies, despite the known effects of countermovements on policy adoption ( Meyer and Staggenborg 1996). As a result, gaps in our understanding of policy 43 Consider the example of Hawaii, where in 1993 the state supreme court left open the possibility of same -sex cou ples marrying . Following this in 1998 , citizens approved a constitutional amendment via a legislative referendum granting the legislature the authority to prohibit same -sex marriage. However, most recently in 2013, the Hawaiian legislature passed a bill ov erturning their prior ban and legalizing gay marriage. 111 diffusion for this policy area remain. 44 And offering evidence of different diffusion dynamics and venue diffusion in a previously and broadly researched policy area further validates my claims here. Third, gay marriage is a helpf ul lens because of its policy attributes. Both anti - and pro -gay marriage laws are technically -simple, highly salient, morality policies involving cross -cutting cleavages. Past scholars have demonstrated that these types of policies diffuse differently com pared to other policy types (Boushey 2010; Hollander and Patapan 2016; Makse and Volden 2011; Mooney 2001; Mooney and Lee 1999; Mooney and Schuldt 2008; Nicholson -Crotty 2009; Pierce and Miller 1999 ). Morality policies tend to spread rapidly due to competi tion over societal and cultural values rather than due to any incremental learning (Mooney and Lee 1999). Thus, this policy area may also be the most challenging to find evidence of political learning in the policy adoption and venue shopping processes in other states. Even limited evidence of political learning™s effect may underscore its existence in different policy domains. Mobilization for and Counter -Mobilization Against Gay Marriage The fight over gay rights pre -dated the Hawaiian Supreme Court™s Baehr v. Lewin ruling in 1993. From the 1920s through the 1960s and beyond, gays and lesbians fought for the freedom to publicly associate and congregate (Valelly 2012). Throughout the 1970s to 1990s, gay rights groups around the country pushed for anti -dis crimination policies, overturning sodomy bans, and adopting hate crimes legislation. For example, LGBT college students at Michigan State University in East Lansing, Michigan helped pass the first nondiscrimination ordinance for the city that included sexu al orientation in 1972 (Fetner 2008). Slowly but surely the LGBT community was achieving social 44 In some ways, the pursuit of pro -gay marriage policies is the equivalent of a policy reversal (Eyestone 1977; Lowry 2005). The diffusion dynamics of reversals are different from adoptions: they rarel y embody geographic patterns, the speed is more gradual, and different institutional and interest -group factors drive the diffusion of these disinnovations. 112 change, predominately at the local level by dint of municipalities, universities, and corporations™ employment and human resource practices, but occasionally at the state level too (Fetner 2008). But as the gay rights movement gradually gained traction, the anti -gay countermovement materialized more swiftly. In 1977, Anita Bryant Šfamed singer, television celebrity, and former beauty queen, turned anti -gay activi stŠfounded fiSave Our Childrenfl and successfully repealed an anti -discrimination ordinance protecting LGBT individuals in Miami -Dade County. Her celebrity status helped catapult the campaign against gay rights to the national spotlight (Fetner 2001, 2008). The nascent anti -gay countermovement quickly integrated into the fiMoral Majorityfl and then the religious right, dwarfing the gay rights movement (Fetner 2001, 2008; Smith 2008). According to Tina Fetner (2008: xiv -xv): fiThe size of the religious right, wh ether measured in membership, size of organizations, revenue, or other resources, was dramatically greater than that of the lesbian and gay movement. If opposing movement activism were a head -to -head battle of strength, the religious right would have crush ed the lesbian and gay movement outright.fl The Christian right pressed for keeping anti -sodomy laws; overturning newly passed domestic partnership laws; restricting the rights of people with AIDS; and limiting LGBT -individuals™ ability to retain custody of their children, serve as foster parents, or adopt (Conger 2009; Donovan, Wenzel, and Bowler 2000; Fetner 2008; Stone 2012; Wald et al. 1996). The U.S. Supreme Court™s Bowers v. Hardwick 45 ruling in 1986 that fithere [was] no such thing as a fundamental ri ght to commit homosexual sodomyfl further emboldened the religious right. Groups such as Focus on the Family, Family Research Council, Concerned Women for America, Christian Voice, and Traditional Values Coalition pursued restrictions on gay rights at the local and state levels via multiple venues (Conger 2009; Fetner 2008; Green 2000; Haider -Markel 45 Bowers v. Hardwick, 478 U.S. 186 (1986) 113 2000; Stone 2012; Wald et al. 1996). The religious right was especially successful in making their case directly to the voters through popular referendums and ci tizen initiatives (Fetner 2008; Stone 2012). For example, in 1991, Colorado for Family Values, a conservative Christian organization, sponsored Amendment 2, a ballot measure that eliminated existing and future gay rights laws in the state. Proponents argue d Amendment 2 was necessary so LGBT individuals did not acquire fispecial rights.fl The amendment passed although the U.S. Supreme Court later overturned it in Romer v. Evans .46 In 1992, similar groups in Oregon lobbied for Ballot Measure 9, which required th e firing of LGBT public school teachers and outspoken allies of the gay community, along with the removal of all books from government -funded libraries that discussed homosexuality (Stone 2012). Gay rights groups were able to defeat Ballot Measure 9 in Ore gon, but the religious right spread similar tactics and initiatives to other states and localities (Fetner 2008). Yet, it was the Hawaiian Supreme Court™s Baehr ruling that was the tipping point for both social movements. While not an affirmative ruling, the decision was an opening for gay marriage in Hawaii and across the states. It caught the attention of LGBT rights groups and activists, many of whom previousl y believed pursuing equal marriage rights was antithetical to the movement or a fool™s errand (Fetner 2008). The court ruling galvanized the religious right and conservative lawmakers (Gallagher and Bull 2001). For them , it was further proof the gay rights movement was eroding traditional family values and morality in America. Fundamentalist Christian churches and conservative religious groups amped up their mobilization against gay rights, especially same -sex marriage (Fetner 2008). By 1994, one -third of t he religious right™s voter guides directly mentioned the fight against gay rights. Anti -gay groups even tried to paint LGBT rights as the right to pedophilia (Gallagher and Bull 2011). 46 Romer v. Evans, 517 U.S. 620 (1996) 114 Due to pressure from these interests and backed by popular support, ele cted officials across the country acted quickly (Haider -Markel 2000, 2001). Republican -led, and even some Democratic -led, state legislatures were fearful their states would have to recognize same -sex unions performed in Hawaii due to the U.S. Constitution™ s and their state constitutions™ full faith and credit clauses. Politicians began considering bans on gay marriage performed in their states and the recognition of same -sex unions solemnized in other states (Gallagher and Bull 2001). By the early 2000s, th irty -five state legislatures had passed statutory language prohibiting same -sex marriage. Conservatives and the Christian right were also successful at the federal level, convincing Congress to pass and President Bill Clinton to sign the Defense of Marriag e Act (DOMA) into law in 1996. DOMA defined marriage for federal purposes as a union between fione man and one woman,fl simultaneously allowing states to disregard the full faith and credit clause of the U.S. Constitution in recognizing same -sex unions perfo rmed in other states. But the opposition groups did not stop with state and federal statutory language forbidding equal marriage rights for gays and lesbians. Capitalizing on early public support, they also proactively pressed for constitutional amen dments via legislative referendum and citizen initiatives prohibiting gay marriage, civil unions, domestic partnerships, and anything akin to marriage for same -sex couples. Doing so circumscribed legislative and judicial efforts by the gay rights movement to overturn these restrictions. The only way to annul these so -called fisuper -DOMAsfl was to reverse public opinion and return to the ballot box (Stone 2012) or head to federal court. By 2008, more than thirty states enshrined the ban on gay marriage in thei r constitutions by passing legislative referend a or ballot measures. 47 47 The religious ri ght even had hopes for a U.S. constitutional amendment fito protect the institution of marriage.fl Delighting these groups, President George W. Bush called for such an amendment in his 2004 State of the Union Speech (CNN 2004). Although a federal ban was nev er adopted, the right™s overall proactive opposition was widely successful. 115 Figure 5.1: Adoption of Gay Marriage Bans by U.S. State by Venues, 1995 Œ 2010 Notes: State maps display adoption of anti -gay marriage statutes and constitutional amendments by venue in 1995, 2000, 2005, and 2010. fiLegfl = Legislature, fiLeg. Reffl = Legislative Referendum, fiCIfl = Citizen Initiative. States adopted bans on same -sex unions vi a the legislature, legislative referendum, citizen initiative, or multiple of these venues at different points in time. See Tables D.1 and D.2 in the Appendix for a full chronology of anti - and pro -gay marriage policies pursued in every state. Figure 5 .1 better displays each state™s adoption of gay marriage bans by venue type from 1993 Œ 2015 . 48 Showing four snapshots in time Š1995, 2000, 2005, and 2010 Šthe maps illustrate how dozens of states pressed for prohibitions on same -sex unions via state legislatures, legislative referenda, and citizen initiatives. By 2000, 33 state legislatures had favora bly adopted statutory language against marriage equality, while two state legislatures (Hawaii and Alaska) also asked voters 48 As discussed in Chapter 2 and Chapter 3 , not all states have the same set of venues available to pursue policy change. Delaware, for example, is the only state that does not allow the legislature to put forward a referendum to its electorate. Only 24 states allow their citizens to directly appeal to the voters on statutory or constitutional matters. See the Appendix for a full chronology of the anti -gay -marriage policies pursued and adopted in every state. 116 via referenda to enshrine fitraditional marriagefl into their constitutions, and two other states (California and Nebraska) sought to protect the status quo via citizen initiatives. Just five years later, most of the states that had circumscribed gay marriage statutorily also circumscribed it constitutionally via legislative referenda and plebiscite. By 2010, the overwhelming majority o f states had prohibited same -sex unions via the legislature, legislative referendum, citizen initiative, or via multiple of these avenues. The religious right™s vast and sweeping countermovement against gay marriage at the subnational and federal levels wa s hugely successful. Before the Baehr ruling, same -sex marriage had not been a top priority for the gay rights movement. Many in the LGBT community eschewed marriage as a fipatriarchal, heterosexual institutionfl (Fetner 2008). They saw any fight for gay mar riage as fiassimilationistfl to heterosexual culture. Still, for others in the movement, same -sex marriage was an equal rights issue. Given the internal disagreement, most gay rights groups sideline the issue and instead focused their attention on pressing f or anti -discrimination and hate crime laws (Fetner 2008). The Baehr decision and massive countermovement by the religious right changed that. Gay rights activists and groups such as the Federation of Statewide LGBT Political Organizations (later known as Equality Federation), Freedom to Mar ry, Gay and Lesbian Alliance Against Defamation (GLAAD), National Gay and Lesbian Taskforce, and Lambda Legal pursued an incremental strategy. They filed lawsuits in other states, methodically selecting courts they belie ved would be receptive to their cause (Andersen 2005; Rayside 2005). Policy actors pursued subsequent litigation in Alaska, Vermont, Massachusetts, and California, among other state courts. However, the right™s countermobilization forced LGBT interests to fight in multiple venues, including state capitals and at the ballot box. 49 Playing defense, gay allies were sometimes forced to support 49 Still, a wide disparity in resources existed between the religious right and gay rights movem ent. According to Linda Hirshman (2012: 344), fionly 3.4 percent of all gay and lesbian adults contribute more than thirty -five dollars to any 117 discordant legislation to mitigate the potential damage caused by more severe policy action. For example, gay rights lea ders and Democratic party officials in Washington voted to override the governor™s veto of a gay marriage ban so the bill would not end up as a ballot measure and further hurt Democrats on the ticket (Haider -Markel 2000). Playing offense, following legal v ictories in Vermont and Massachusetts, gay rights groups lobbied legislatures in California, Connecticut, New Hampshire, and Oregon. Even though only the latter three extended rights to same -sex couples, and only by way of domestic partnerships and civil u nions, state legislatures became a viable venue to press for policy change proactively .50 Although a rarer route, LGBT activists also pursued same -sex marriage via legislative referendum (Maryland) and citizen initiative (Maine). 51 As more state courts and capitols authorized same -sex unions and as public opinion shifted, gay rights groups finally turned their attention to the federal courts. In particular, the U.S. Supreme Court™s 2003 5 -4 ruling in Lawrence v. Texas , overturning the 1986 Bowers v. Har dwick precedent and states™ bans on sodomy, made the federal courts a more attractive venue. Similarly, the Court™s 2013 United States v. Windsor decision invalidating the federal Defense of Marriage Act (DOMA) precipitated a flood of federal lawsuits and opinions in favor of same -sex marriage. Although all states had legislatures, state high courts, or federal courts at their disposal to try to protect minority rights, the diffusion of pro -gay marriage policies w as not as sweeping compared to the religiou s right™s activities. LGBT interest groups were not as successful. Sixteen state legislatures considered adopting civil unions or same -sex marriage, with fourteen following through, identifiable gay cause. The ten largest anti -gay organizations ŠFocus on the Family and the like Šhave twice the $50 0 million in revenues of all the gay organizations put together.fl 50 For coding purposes, I treat civil unions as synonymous with same -sex marriage since the adoption of civil unions were innovative in that they guaranteed the right of gay persons to legall y codify their relationships and access state services and benefits. That said, I recognize the controversial distinction between these policy prescriptions. 51 It is worth noting that gay rights groups did not limit their mobilization to state legislatu res, state and federal courts. They also augmented their voices via blogs and social media outlets, organized meetings and marches, and engaged in civil disobedience. Gay rights groups also used their clout to shame corporations and private law firms that maintained or defended anti -gay policies (Hirshman 2012). Likewise, the religious right utilized its extensive network of churches, members, and media outlets to gain the ear of elected officials and the mass public (Fetner 2008). 118 the best success rate of any of the avenues pursued. Meanwhile, gay rights groups litigated in seventeen state supreme courts, with slightly more than half of those courts ruling in favor of marriage equality. Finally, mostly following the U.S. Supreme™ Court™s 2013 United States v. Windsor ruling against the national DOMA law, some thirty states filed suits in federal court with two -thirds of those states prevailing. The gay rights movement did achieve success in most of the venues it pressed for change, but their effort was more gradual and concentrated. Figure 5.2 illustrates each state™s successful adoption (or lack thereof) of pro -gay marriage policies via state legislature, state supreme court, or federal court at four points in time from 2000 to 2015. Only the first state venue where marriage equality was suc cessfully achieved is represented. For example, civil unions for same -sex couples were achieved in Vermont via the state supreme court in 1999, but Vermont™s legislature passed full marriage equality ten years later. The state maps only record Vermont™s fi rst successful venue. In stark contrast to the anti -gay marriage movement™s success, by 2005 only Connecticut, Massachusetts, and Vermont allowed same -sex couples to codify their relationships legally . The Connecticut legislature granted civil unions while both Massachusetts and Vermont™s state high courts mandated state action to provide equal protection for gay couples. Several subsequent state courts and legislatures took action , and five years later, one -fifth of the states allowed same -sex unions in so me capacity. By early 2015, just a little over two decades after Hawaii™s Baehr v. Lewin decision, and through further state legislative, state court, and federal legal action, an overwhelming majority of Americans lived in states permitting gay marriage. Later that same year, the gay rights movement™s fight for marriage equality culminated in victory following the U.S. Supreme Court™s Obergefell v. Hodges decision. 119 Figure 5.2: Adoption of Pro -Gay Marriage Policies by U.S. State by Venues, 1995 Œ 20 10 Notes: State maps display adoption of pro -gay marriage policies by venue in 2000, 2005, 2010, and 2015. fiLegfl = Legislature, fiLeg. Reffl = Legislative Referendum, fiCIfl = Citizen Initiative, fiSt. Courtfl = State Court, and fiFed. Courtfl = Federal Court. S tates pursued same -sex unions via the legislature, legislative referendum, citizen initiative, state courts, federal courts, or multiple venues at different points in time. Only the first venue in a s tate where marriage equality was successfully achieved i s represented here. See Tables D.1 and D.2 in the Appendix for a full chronology of anti - and pro -gay marriage policies pursued in every state. This historical account of the movement for and countermovement against gay marriage by way of multiple institutional venues documents the spread of anti - and pro -gay marriage policies. But homophily in policy adoption is not the same as policy diffusion. The latter involves external forces at play even after considering internal forces. The next section det ail s what other scholars have found regarding the spread of anti -gay marriage policies, how accounting for political learning and countermobilization efforts might matter, and how we can further leverage our understanding the dynamics of diffusion by model ing policies™ propagation via multiple arenas. 120 Past Diffusion Research of Gay Marriage Policies Multiple scholars have studied the adoption of gay marriage bans across U.S. states (Barclay and Fisher 2003, 2008; Camp 2008; Haider -Markel 200 0, 2001a, 2 001b ; Hume 2011; Lewis 2011 ; Lupia et al. 2010; Soule 2004). Not surprisingly following the narrative above, Haider -Markel (2001) found that national - and state -level conservative religious groups drove the transmission of statutory gay marriage bans. This conclusion challenged the thinking at the time Šthat states primarily looked to and pursued the policies adopted by their geographic neighbors (Berry and Berry 1990 ; Walker 1969 ). Rather, Haider -Markel™s results coincided with the concurrent findings that the diffusion of morality policies behaves differently relative to other issue areas (Mooney 2001; Mooney and Lee 1995, 1999). Instead of following an incremental learning or even geographic process, policies involving competition around values appear to s pread more rapidly (Boushey 2010) and are more dependent on public opinion than outside forces. Barclay and Fisher (2003) attributed the passage of legislative bans to their timing during election years, the percentage of residents with a college educati on, the number of localities in the state that provided domestic partnership coverage, and the strength of an LGBT presence in the state. Similarly, Soule (2004) showed that a state™s adoption of a statutory ban on same -sex marriage was mostly due to a sta te™s prior policy activity on sodomy bans and hate crime legislation, citizen ideology, interest group activity by the religious right and gay rights movement , and Congress ™s passage of DOMA in 1996. Meanwhile, Barclay and Fisher (2008) were the only scholars to test (albeit indirectly) and offer some evidence of policy learning. Hume (2011), Lupia et al. (2010), and Lewis (2011 , 2013 ), however, offer institutional explanations for states™ adoption of anti -gay marriage policies, especially constitutional amendments outlawing same -sex unions. Hume (2011) attributes that the adoption of constitutional amendments 121 banning gay marriage to t he capacity of state high courts . Policy actors in states with more professionalized courts of last resort that are more likely to protect minority rights appear to press for constitutional bans to curb the judiciary™s power. Lupia et al. (2010) find that a state™s degree of difficulty in amending its constitution largely dictates whether a state enshrines a traditional definition of marriage into its canon. Direct democracy states, for example, were much more likely to pursue and adopt such amendments, com pared to non -direct democracy states. Lewis™ s (2011 , 2013 ) research further supports this conclusion. Depending on different measures for a state™s level and use of direct democracy, Lewis shows that states permitting plebiscitary action were indeed more l ikely to adopt laws banning gay marriage. Taken altogether, past research trying to parse the internal and external forces driving anti -gay marriage policies has identified the following factors: national and state interest groups, federal -level influence from Congress or the Supreme Court, and the ease of amending state laws or constitutions, among other internal institutional factors. While these certainly comprise many of the pieces, the puzzle remains incomplete. I believe four key p arts are still miss ing. Four Missing Pieces to the Puzzle First, scholarship on this policy has largely treated the adoption of a statute, constitutional amendment via legislative referendum, or constitutional amendment via citizen initiative as equivalent. Researchers™ fo cus was on the propensity to adopt rather than the arena where the policy was adopted. Typical of past diffusion scholarship, the policy output Šgay marriage ban Šrather than the input Švenue to achieve such prohibitions Šwas the star. Because institutional se ttings are crucial to policy change and the policy process, the variation on venue should also be accounted for in our models. For those policy ideas only pursued via one forum , institutional variables specific to that venue would suffice. But for the numb er of innovations pursued over multiple paths, the 122 variation across and competition between institutional arrangements need to be accounted for to correctly capture the external and internal mechanisms at play. Incorporating the different, competing arenas in our modeling strategies allows us to determine which factors matter for achieving policy change in a venue. This richer understanding will further help unpack the remaining black box of the policy process. The second piece absent from o ur understanding of the fight for marriage equality is accounting for pro -gay marriage policy successes in our models. Knowing the power and role of interest groups in the policy process (Balla 2001; Gray and Lowery 1996; Mintrom 1997; Nownes and Lipinski 2005 ; Wolak et al. 2002), earlier research did control for the presence, size, and capacity of both the religious right and gay rights interest groups. In fact, n umerous different measures over the years have been employed to capture the religious right an d gay rights movements' state -level interest group presence. To operationalize the religious right, scholars have relied on percentage of state population that identifies as Evangelical Protestant or Roman Catholic ( Fleischmann and Moyer 2009 ; Haider -Marke l 2007; Thomas and Hrebenar 2008), the number of state -affiliated Focus on the Family Councils (Soule 2004), the ratio of religious right interest groups to total groups in the state (Conger and Djupe 2016), among other approaches. To measure the gay rights movement, researchers have used the membership rate per capita in national LGBT groups (Haider -Markel 2001a, 2001b), estimates of proportion of state residents identifying as LGBT (Taylor et. al 2012), the number of openly -gay state legislators (Hai der -Markel 2007), the number of Gay and Lesbian community centers or pro -gay groups in the state (Kane 2003; Soule 2004), or the proportion of LGBT groups to total interests in the state (Conger and Djupe 2016) , to name a few . Despite these excellent atte mpts to account for the role of interest groups in the diffusion process of anti -gay marriage policies, most of the research failed to account for the gay rights movements™ policy successes on the marriage front. Lewis (2011) is the only article to my know ledge 123 that includes whether states had allowed civil unions or domestic partnerships in some capacity. Lewis finds that states with such laws were more likely to adopt bans on same -sex marriage since they had provided an fialternativefl for same -sex couples, although it is unclear whether these fialternativesfl preceded or succeeded the bans. Still, despite recognizing the reciprocal nature between social movements and countermovements (Austen -Smith and Wright 1994; Meyer and Staggenborg 1996), we do not know h ow the successful adoption of pro -gay marriage policies affected the spread of anti -gay marriage policies. Paralleling the second omission in our understanding of the policy fight over gay marriage, the third missing piece of the puzzle is that no scholar has examined the diffusion of affirmative same -sex marriage policies. This is somewhat surprising because it was the possibility of same -sex unions in Hawaii that galvanized the countermobilization on the right. And although the transmission of pro -gay ma rriage innovations was more gradual than the acute diffusion of anti -gay marriage policies, they were innovations, nonetheless. Scholars likely overlooked the adoption of pro -gay marriage policies because most wins occurred so late in the cycle of the move ment, between 2005 and 2015. But researchers may have also marginalized the fight for marriage equality because success happened initially via state supreme courts rather than in state capitols and at the ballot box. As Chapter 3 pointed out, save for a fe w dozen articles, diffusion scholars have largely dis counted policies that spread outside the legislative context. The fourth piece missing from the puzzle in our understanding of the policy activity around gay rights is accounting for the political lear ning that happens. Recall from Chapter 4 that May (1992) identified two fundamental types of learning in the policy process: policy learning and political learning . Policy learning involves drawing lessons from the fisocial construction of the policy proble ms, the scope of policy, or policy goals ,fl or even about the viability of the policy innovation, its design, and implementation (May 1992: 332). Since the diffusion process is treated as an incremental learning 124 process, scholars have proposed policy learni ng as a key mechanism for the spread of new ideas (Gilardi 2010, 201 6; Shipan and Volden 2008). Researchers have modeled policy learning as the cumulative number or proportion of states successfully adopting the innovation, the degree to which the policy a chieved its intended outcome (Volden 2006), or the policy™s impact (Gilardi 201 6; Shipan and Volden 2014 ). But diffusion scholars™ prioritization of policy learning has sidelined the political learning that also occurs. Political learning is the gaining o f information about how to fimaneuver within and manipulate the policy process to advance an idea or policyfl (May 1992: 340; see also Freeman 2008; Rose 1991). Political learning it is part of the pre -contemplation and knowledge -gathering stage articulated by Rogers (1962). As policy actors learn about available solutions to a common problem, they also learn how to feasibly pursue the policy proposal (May 1992). It is possible that political learning matters as much or more than policy learning in explaining the spread of policy adoptions across governmental jurisdictions. Only one scholarly work to my knowledge has tried to test empirically the impact of political learning. Seljan and Weller (2011) found evidence that the failure of states to adopt tax and e xpenditure limits via direct democracy affected neighboring states™ decision s to pursue the policy. Policy actors in states drew lessons from early movers about the feasibility and political process, thus affecting their choice whether to seek a policy cha nge. As Chapter 4 laid out, a n essential aspect of political learning is gaining information about the most favorable venue to pursue policy change. Although policy actors certainly consider their capacities, institutional settings, and political environ ments in shopping for a favorable venue (Holyoke, Brown, and Henig 2012; Sabatier and Jenkins -Smith 1993 ), they also might look to and learn from others (Ley and Weber 2015; Pralle 2003) . Policy actors, facing limited time, attention, and resources, are bo undedly rational (Lubell 2003). Thus, they satisfice and decide which path to take by considering, at least partly, where other policy entrepreneurs and actors have successfully 125 achieved policy change. Given the vast interdependencies in the policy process, it is more likely that policy actors learn from the venue sho pping processes in previous states. I mainly expect political learning to have a positive effect on the venue shopping process of others. Nonetheless, Pacheco (2017) reminds us that policy actors also learn from failures and that policy diffusion is not al ways a positive feedback process. For example, the pursuit of a specific venue or passage of a policy in one state may yield positive spillover effects in neighboring states, thus inducing free -rider dynamics (Franzese and Hayes 2006; Pacheco 2017) and no need for neighboring states to pursue or adopt the given policy. Therefore, political learning may also produce a negative effect on venue choice or policy enactment. Because venue shopping is such an integral part of the agenda -setting process, I opt t o model political learning as the cumulative proportion of states that were successful when they pursued a policy innovation via a given venue in a particular year . While policy learning involves seeing how many other states adopt a new idea irrespective o f the venue, political learning involves seeing which paths are most favorable. I believe political learning has been sidelined in the past because scholars have focused on how an innovation has spread from one legislative context to another legislative co ntext. When only one forum is involved, variation in the dependent variable does not exist and empirically parsing political learning (success via a given venue) from policy learning (success overall) is near impossible. Importantly , political learning sti ll occurs in these myopic contexts but is too entangled with policy learning to tell them apart. Therefore, when an innovation is pursued in multiple arenas, we can leverage this variation to determine the effect of both learning processes. 52 52 It is worth pointing ou t, however, that political learning could be captured in other ways. Although, I measure political learning as the proportion of states that pursue a new policy and successfully adopt the policy via a given venue, the theory of political learning involves more than simply looking for the most successful path. For example, states may strategically look to others similar along an institutional dimension. Policy scholars have shown that direct democracy states behave differently than non -direct democracy state s (Boehmke 2005; Bowler and Donovan 2004; Lewis 2011, 2013). And diffusion scholars find that policy actors look to states with similar degrees of legislative professionalism, supreme court professionalism, and difficulty in amending state constitutions (F ay and Wenger 2015; Hume 2011; Lupia et al. 2010; Shipan and Volden 2006; Yates, Tankersley, and Brace 2010). These actions imply a purposive search of 126 Since past research missed these four pieces of the gay marriage diffusion puzzle Šaccounting for adoption via multiple venues, acknowledging opposition success, mapping the spread of pro -gay marriage policies, and considering political learning, I revisit these elements in the succeeding sections. Specifically, I retest the prior conclusions reached by scholars on the dissemination of anti -same -sex marriage policies by including pro -gay marriage policy successes and political learning in the analyses. I als o model the spread of anti - and pro -gay marriage policies across multiple institutions, leveraging the variation afforded the different arenas. Finally, I explain how these results better inform our understanding of policy diffusion in the context of gay m arriage and beyond. The Diffusion of Gay Marriage Policies : Expectations Building on past research regarding the internal and external forces that drive policy change, and adding the missed pieces discussed above, I lay out revised expectati ons for why anti - and pro -gay marriage policies may have spread across U.S. states. As discussed in the pr evious section, political learning Šdrawing lessons from the successful tactics and venues pursued in earlier adopters Šmay drive the successful adoptio n of both anti - and pro -same -sex union policies. In the context of gay marriage, I anticipate that as conservative actors and fundamentalist Christian organizations successfully enact a ban on gay marriage via a venue in one state, subsequent actors will l earn from others™ calculations in picking a venue to achieve triumph in their state when they pursue similar bans. Likewise, as more gay rights groups effectively achieve marriage equality by way of a venue, actors in other states will learn from these tac tical maneuvers and be more successful in pursuing pro -gay marriage policies in their states. I expect that political knowledge gained from the institutional information to aid in picking a venue. Looking to one™s institutional peers may also sugge st political learning. Thus, political learning may merit an even broader measurement strategy. 127 successful venue shopping strategy in one subunit will transfer to and make policy actors more effective in achi eving that policy in their state . Successful policy adoption should be more likely precisely because subsequent actors largely follow the venue choice of like -minded predecessors. As such, I propose the following central hypothesis: H1: Political Learning: The likelihood of a state adopting an anti -gay (pro -gay) marriage policy increases as the proportion of states successfully pursuing that policy via a given venue increases . However, prior scholarship has identified plausible, alterna tive external mechanisms that may also influence the adoption of gay marriage policies. In facing a common societal problem, policy actors may satisfice and look for available solutions, regardless of how or where those solutions got adopted. As initial states adopt such policies, others will learn about the available solutions and follow suit (Shipan and Volden 2008; Karch et al. 2016). Nonetheless, since the fight over same -sex marriage involves a morality policy area, few scholars expect an incremental l earning process (Mooney 2001; Mooney and Lee 1995, 1999). Instead, scholars expect an acute response driven by contagion and public opinion. Still, Barclay and Fisher (2008) indirectly test the policy learning mechanism and do find some evidence that state s are more likely to pursue a ban on same -sex unions as the number of other states considering such bans increases. As such, I offer the following hypothesis: H2: Policy Learning: The likelihood of a state adopting an anti -gay (pro -gay) marriage policy increases as the number of other states adopting the policy increases . Another possibility is that policy actors may emulate the policy adoption of peer states. Perhaps states look to and copy the policies adopted by their contiguous geographic neighbors (Berry and Berry 1990; Berry and Baybeck 2005; Cohen -Vogel and Ingle 200 7). In the search for potential solutions, it may be easiest to look next door. Although Haider -Markel (2001) did not find 128 evidence of a regional diffusion effect, Hume (2011) and Lewis (2011) observed an effect in their models. To test this mechanism, I p ut forward the following hypothesis: H3: Geographic Neighbor: The likelihood of a state adopting an anti -gay (pro -gay) marriage policy increases as the proportion of contiguous neighboring states adopt ing the same policy increases . The federal governme nt can often encourage or discourage policy adoption across states (e.g., Karch 2006, 2012; Shipan and Volden 2006, 2008). Pertinent for my examination of same -sex marriage policies, Congress passed the Defense of Marriage Act (DOMA) in 1996 defining marri age for federal purposes as a union between a man and woman and allowing states to reject same -sex unions performed in other states. Moreover, the Supreme Court™s 2003 ruling in Lawrence v. Texas struck down the states ™ anti -sodomy law s, while the High Court™s 2013 ruling in United States v. Windsor invalidated DOMA. Federal government involvement in the fragmented American system could influence the spread of both anti - and pro -gay -marriage policies across sub -national units. Hence, I h ypothesize: H4: Federal -Level Intervention : The likelihood of a state adopting an anti -gay (pro -gay) marriage policy increases /decreases as federal -level intervention in the issue area occurs . Related to direct federal -government activity on the issue, the national political environment can also affect the adoption of policies. Multiple researchers have demonstrated that the salience of an problem or policy can hasten or hinder action across governmental units (Boushey 2010; Makse and Volden 2011; Nicho lson -Crotty 2009). Crisis or focusing events can force and sustain an issue on the policy agenda (Baumgartner and Jones 1993). As national attention increases, policymakers and interest groups may be more or less likely to enact a particular policy. Becaus e state legislators are electorally motivated and the least institutionally constrained to press for policy change, I anticipate that national attention will increase or decrease state legislatures™ propensity to act. In addition, the timing of certain nat ional events, such as presidential elections, may also prompt policy 129 action (Berry and Berry 1990, 1992; Mintrom and Vergari 1998) Although Smith, DeSantis, and Kassel (2006) do not find evidence that anti -gay marriage ballot measures increased turnout dur ing the November 2004 presidential election, the timing of the propositions could have influenced the constituency that turned up at the poles. Given the potential for the national context to increase or decrease the adoption of such policies, I submit the following hypothesis: H5: National Environment : The likelihood of a state pursuing a policy via a given venue increases / decreases as the national environment on the issue area ebbs and flows. Prior research has also identified how interest group ac tivity can affect policy adoption (Boehmke 2005; Mintrom and Vergari 1998). As the narrative above recounting the mobilization for and countermovement against gay marriage explained, conservative and religious right networks and gay rights groups were pivo tal to the passage of anti - and pro -gay marriage policies (Barclay and Fisher 2003; Haider -Markel 2000, 2001; Haider -Markel and Meier 2003 ; Lewis 2011; Soule 2004 ). Although past scholars accounted for the size, capacity, or resources of these organized interests, they failed to account for the success of the opposing side. Numerous studies have documented how one side ™s actions can spur a response by the opposi te side ( Austen -Smith and Wright 1994; Conger and Djupe 2016; Kane 2010; Meyer and Staggenborg 1996; Stone 2012). Even the fear or threat of action by the opposition can produce fianticipatory countermobilizationsfl (Dorf and Tarrow 2014). As such, in addition to controlling for the size of state -level interests in the fight over gay marriage, I also account for the opposition™s policy wins regardless of the venue where they occurred . In the context of anti -gay marriage policies, this implies that the enactment of marriage equality in some states should influence the policy action by the religious right. In contrast, the passage of policies curtailing same -sex unions should affect the LGBT epistemic community™s efforts to press for policy change. I propose: 130 H6: Opposition Success: The likelihood of a state adopting an anti -gay (pro -gay) marriage policy increases as the number of opposition policy successes increase. In summary, I expect that political learning, gaining information about successful paths taken by early movers, will affect the adoption of anti - or pro -gay marriage policies in subsequent states. However, other external forces --- including, how many other states pass the policy, whether state neighbors enact, federal government involvement and national environment contexts, and policy successes by the opposition --- could also drive subnational policy activity on gay marriage. Still, state internal factors should also influence policy change. Although I do not list out separate hypotheses for thes e internal forces, based upon pr evious policy diffusion research, I anticipate that a state™s institutional arrangements, political context, interest group pressure, prior policy adoptions, and demographic determinants could explain why it prohibits or per mits same -sex unions. I turn now to a systematic analysis of my hypotheses. Data and Methods Data In this section, I re -examine the policy diffusion of anti -gay marriage policies by resolving the four missing pieces of past research. To do so, I const ructed the relevant data universe of anti - and pro -gay marriage policies pursued and adopted via multiple institutional venues across U.S. states. Unlike most policy diffusion studies that model binary policy adoptions irrespective of venue, my claims rest on accounting for successful and failed policy attempts via multiple venues . For example, California™s legislature failed to pass a ban on same -sex unions in 1997, but a Californian citizen initiative proscribing gay marriage did succeed in 2000. Standard diffusion models would only record California™s adoption of the innovation in 2000, ignoring the rich information from California™s legislative attempt in 1997. Because policy actors not only learn about new innovations but also draw 131 lessons from the fruitful and foiled political tactics (including choice of venue) employed in earl ier states to pursue those innovations, accounting for this in the data is crucial to telling the story. As such, I searched for and compiled successful and failed anti - and pro -gay marriage policy attempts via manifold forums in all U.S. states since the early 1970s. I relied on and triangulated data from myriad sources, including Freedom to Marry (2015) , Haider -Markel ( 2000, 2001), Hume (2011), Keck (2009), Lewis (2011), National Conference of State Legislatures (NCSL) , the National Gay and Lesbian Task Force (2013) , Pinello (2015), Stewart (2015), and Thompson (1994 ). See Tables C.1 and C.2 in the Appendix for a full chronology of the anti - and pro -gay marriage policies pursued by venue type in each U.S. state. 53 For my purposes here, states enter the ri sk set of adopting an anti - or pro - gay marriage policy following Hawaii Supreme Court™s Baehr decision in 1993 and exit the set before or on 2015 when the Obergefell ruling settled the issue . Although other state courts adjudicated similar cases decades before the Baehr ruling and even though Hawaii™s high court later ruled against same -sex marriage, the opinion was the first genuine opening for gay marriage in America. 54 Furthermore, i f a state passed a statutory or regulatory restriction before 1993, it i s also not included in the risk set. 55 53 A few additional notes regarding the anti - and pro -gay marriage events included in the chronology and analyses: I only incorporate civil unions and domestic partnerships that extended governmental and legal benefits to same -sex couples mirroring the benefits provided via marriage. Only those state court cases appealed to and taken up by a state™s highest court are included in the analyses. Court cases, legisl ation, or executive orders extending rights of divorce to same -sex couples are also not included as events since those policies did not affirm a right to a union for gay couples. In the same vein , although multiple court cases and policies over the decades dealt with other gay rights issues, the policy events here explicitly pertain to marriage equality. Finally, legislat ive votes to convene a state constitutional convention, with the possibility of voting on gay marriage issues , are also not considered in the analyses because conventions rarely occur (e.g., once every ten years for many states) and open the door for policy considerations in other issue areas. 54 Minnesota ™s Supreme Court was the first state court to uphold a definition of marriage a s between one man and one woman in 1971 (Baker v. Nelson ), with at least ten other state courts issuing similar rulings: Washington, Kentucky, Alaska, Florida, Hawaii, Illinois, Iowa, New Hampshire, South Dakota, and Utah (Haider -Markel 2001b; Soule 2004) 55 Several states adopted direct or indirect statutory o r regulatory language denying same -sex couples the right to marry prior to the 1993 ruling in Hawaii. I document that eleven states had such laws on the books or in their family code: Arizona in 1975; Florida in 1977; Indiana in 1986; Louisiana in 1988; Ma ryland in 1973; New Hampshire in 1987; Oklahoma in 1975; Texas 1973; Utah in 1977; Virginia in 1975; Wyoming in 1977 (Freedom to Marry 2015; National Gay and Lesbian Task Force 2013; Soule 2004; Stewart 2015; and Thompson 1994). California™s legislature pa ssed similar legislation in 1977 but was never signed into law by the governor. Nonetheless, even accounting for the states that had previously adopted such policies with a dummy variable, the key findings in this chapter remain. 132 The key dependent variable in this section is whether a state adopted an anti -gay marriage policy via the legislature, legislative referendum, or citizen initiative in a given state -year from 1993 Œ 2015. 56 Accordingly, the unit of analysis is state -venue -year, where units take on a value of 0 until states adopt a prohibition on same -sex unions via the given forum , when those state -venue -years take on a value of 1. Since states are at risk of passing an anti -gay marriage policy via three venues at the same point in time, this affords a maximum of 50 3 23 =3,450 . To account for the risk of policy adoption via each arena, I pool the observations from each venue into one dataset. Those states that do not permit citizen initiatives are controlled for in the models since they are not at risk of adopting an anti -gay mar riage policy via a citizen ballot measure. In Event History Analysis, the traditional approach to modeling policy diffusion, units depart the risk set after experiencing the event of interest. Here, states may leave the risk set for one venue upon adop ting a policy in that venue but remain in the risk set for the other venues until they enact policies in those venues. Importantly, however, because several states continued to pursue additional bans via the same venue even after adopting an initial ban (e .g., Idaho, Texas, Utah, Virginia via legislature), there is a possibility of repeated events in the dataset. That is, states may experience additional unordered events after initial adoptions in a given venue. Therefore, states exit the risk set after the y have successfully adopted an anti -gay marriage policy via a given forum , or after enacting additional policies via those venues if they were possible (Boehmke 2009 a; Box -Steffensmeier and Jones 2004; Box -Steffensmeier and Zorn 2002; Buckley and Westerlan d 2004; Jones and Branton 2005). 56 Two states ŠAlabama and M ississippi Špursued limited same -sex marriage bans via gubernatorial executive orders in 1996. This venue option is not included in the pooled models because the events were rare; because both states enacted bans via their legislatures in 1998 and 1997, res pectively; and because adding this additional venue could inflate the number of zeros and potentially overleverage the ones in the dataset (Boehmke 2009b). 133 For instance , the Idaho Legislature amended its marriage laws in 1995 defining marital union as between a man and a woman. Idaho™s legislature passed further language prohibiting the recognition of same -sex unions performe d in other states in 1996. As a result, Idaho does not drop out of the dataset for the legislative venue units until 1996 due to the repeated event, even though it successfully passed its first bill in 1995. As a point of clarification, successful policy e nactment implies the passage and implementation of the policy. If a state legislature adopted statutory language prohibiting same -sex unions but the governor vetoed the legislation, this event would not be coded as a success. Variable Operationaliza tion The Political Learning hypothesis (H 1) holds that as the proportion of states that successfully pursue an anti -gay marriage policy via a given venue increases, other states will be more likely to adopt a similar policy. I operationalize my main indepe ndent variable ŠPolitical Learning Šas the total number of states that successfully adopted the policy via the given venue at time , divided by the total number of states that pursued the policy via that venue at time . Fundamentally, the political learning variable is a proportion variable capturing the cumulative success rate in each distinct venue by a given year .57 To illustrate furthe r, consider political learning ™s numerical value s for states pursuing prohibitions on same -sex unions via legislative action at three points in time: 1993, 1994, and 1997. In 1993 , following the Hawaiian Supreme Court™s Baehr decision , no state leg islature pushed to proscribe gay marriage that year , so the political learning variable takes on a value of 0 for those fifty state -venue -year units. By 1994 , the Hawaiian legislature was the first to pass specific statutory 57 = 134 language banning same -sex marri age. Because the success rate in that arena in 1994 was one hundred percent (with only Hawaii trying and successfully adopting the bill) , the political learning variable takes on a value of 1.0 for those fifty state -venue -year units. At that point, all the states (i.e., Hawaii) that tried to pass statutory language outlawing gay marriage via the legislature succeeded. However, by 1997, only 27 out of 33 state legislatures had successfully banned gay marriage. Therefore, political learning takes on a value of 0.818 for those fifty state -venue -year units in 1997 . Still, political learning™s values for observations associated with legislative referenda or citizen initiatives take on a value of 0 for 1993, 1994, and 1997 because action in these venues did not occur until 1998 and 2000 for legislative referenda and citizen initiatives, respectively. Importantly, p olitical learning™s values for states are contingent on and thus vary by the venue under consideration. I anticipate a positive coefficient fo r political learning across all venues. As more states successfully alter the status quo via a given site , policy actors in other states will learn from this, thus increasing their propensity for success in their states. Nonetheless, depending on the succe ss rate via a given venue, political learning may be attenuated. Moreover, a negative coefficient could still offer evidence of political learning, perhaps pointing to a complicated policy process in leader states convincing laggard states to shy away from that arena . The second mechanism articulated in H 2 is Policy Learning, where a state™s potential for adopting an anti -gay marriage policy is partly a function of other states™ decisions to pass a ban on same -sex unions. I measure Policy Learning in two di fferent ways. In one approach , I capture policy learning as the cumulative number of states opting to prohibit gay marriage irrespective of venue. For a different approach, I operationalize policy learning as three separate variables: Policy Learning from Legislature , Policy Learning from Legislative Referendum , and Policy Learning from Citizen Initiative . Each variable captures the cumulative number of states enacting a gay marriage ban by the respective venue type . Regardless of the measurement approach, I 135 anticipate that as more states adopt an anti -gay marriage policy, other states will learn from these fipolicy solutionsfl and be more likely to follow. To test the regional effect hypothesis (H 3), I created the Geographic Neighbor variable as the proporti on of geographically contiguous neighbors that had adopted an anti -gay marriage policy regardless of venue. I rely on Berry and Berry™s (1990) classification of geographic neighbors, with one exception: I treat Alaska and Hawaii as a neighbor pair. I expec t that as more of a state™s neighbors prohibit same -sex marriage, the state will also be more likely to block equal rights for gay couples. I expect a positive coefficient for this variable. The Federal -Level Intervention hypothesis (H 4) predicts that fed eral government involvement in an issue will lead to an increase or decrease in state -level policy activity depending on the type of engagement . In the context of efforts to curtail same -sex marriage, I capture federal -level intervention with two variables . First, Federal Government DOMA takes on a value of 1 for the years following the Congress™ s passage of the Defense of Marriage Act in 1996. I suppose that the national government™s action to outlaw the federal recognition of gay marriage would spur states to take similar action. However, it is also conceivable that early and massive state -level action snow -balled, convincing the federal government to act (Shipan and Volden 2006). Therefore, a negative relationship between the DOMA variable and policy adoption is also possible. Second, Lawrence v. Texas Supreme Court Decision takes on a value of 1 for the years following the U.S. Supreme Court™s 2003 ruling declaring sodomy bans unconstitutional. Following this at -the -time ficontroversialfl decision and the religious right™s response (Smith 2008), I anticipate that states will be more likely to adopt an anti -gay marriage policy, especially state constitutional bans via legislative referendum or plebiscitary action. To test the National Environment Hypothesis (H 5), where heightened national attention around an issue should spur or stall polic y change, I rely on two variables. To capture national 136 salience around marriage equality, I construct a NYT Issue Salience variable providing the cumulative number of New York Times ™ stories on gay marriage during the year. I am agnostic as to whether such heightened attention will have a positive or negative effect on a state™s policy adoption. On the one hand, greater issue awareness could spark a backlash, pushing policy actors and the public to call for further bans. On the other hand, increased focus o n the issue and LGBT community could slow restrictions of minority rights. Presidential elections also tend to put contentious issues in the spotlight and provide opportunities to mobilize fellow partisans around a cause. I suggest that states will be more likely to adopt an anti -gay marriage ban during presidential election years. I depend on Presidential Election Year , where a value of 1 represents a national election in that calendar year. To test the last external mechanism, opposition policy suc cess, I construct a Pro -Gay marriage Counter variable which comprises the cumulative number of pro -gay marriage policies adopted across the country by year. Following the Opposition Success hypothesis (H 6), I predict greater opposition success and adoption of pro -gay marriage policies will foster an increase in countermovement efforts to adopt anti -gay marriage policies. In turn, I anticipate a positive coefficient for this variable. External factors, however, are not the only drivers of policy change. Internal institutional, political, interest group, policy environment, and demographic factors may also affect whether a state adopts an anti - or pro -gay marriage policy, regardless of what other states do. For a state™s institutional attributes, I include three key variables previously found to influence the policy process: legislative professionalism, state supreme court professionalism, and ease of amending the state constitution. We know that legislatures with more considerable resources and cap acity ar e more likely to act (Squire 2007; Bowen and Greene 2014). More professional legislatures may try to preempt policy activity in other venues to achieve an outcome more in line with legislators ™ 137 preferences (Boehmke et al. 2015; Boehmke and Shipan 2015; Dum as 2017; Gerber 1996). Or, depending on how contentious the issue, more astute and electorally mindful legislators may fipass the buck fl to other venues, a llowing the electorate to decide via legislative referendum or an interest group™s citizen initiative. I rely on Bowen and Greene™s (2014) first dimension measure of Legislative Professionalism , where higher values indicate a more professional state house and sen ate. Overall, I assume legislative professionalism will have a positive effect on policy adoption. The professionalization of a state™s court of last resort may also influence a policy™s success or failure (Squire 2008; Yates, Tankersley, and Brace 2010 ). Hume (2011) finds that supreme courts with higher capacity are more likely to adopt constitutional amendments banning gay marriage since opponents of same -sex unions feared the more professionalized judiciary would use its power to curb any legislative a ction on the issue. I concur with Hume™s (2011) expectations and use Squire™s (2008) measure of State Supreme Court Professionalism , where higher values indicate a more resource -ready and qualified judiciary. The degree in difficulty in amending a state™s constitution may also help or hinder policy adoption, especially constitutional bans to prohibit gay marriage ( Dinan 2018; Fay and Wenger 2015; Hume 2011; Lupia et al. 2010; Lutz 1994 ). I follow Lupia et al. ™s (2010) operationalization of Difficulty Amendi ng State Constitution as a range from 1 for states that only require enough signatures for an amendment to make it on the ballot to 4 for states that require both legislative approval via multiple sessions and a voter supermajority to modify the constituti on. Essentially, direct democracy states that allow direct or indirect citizen initiatives score lower on the scale, while non -direct democracy states score higher on the scale capturing difficulty in amending the state constitution. I suppose that as the institutional hurdles to achieve policy change increase, the likelihood of adopting an anti -gay marriage policy will decrease. Turning to political considerations, I include three variables. Because the partisan control of legislative and executive branc hes of government could explain the enactment of pro - or anti -gay 138 marriage policies (Calvert et al. 1989 ; Camp 2008; Goggin et al. 1990; Hinchliffe and Lee 201 6), I control for State Government Party Control . The covariate takes on a value of 0 for unified Republican control, 0.5 for bipartisan control, and 1 for unified Democratic control of both state legislative chambers and governor™s mansion. I expect that states with bipartisan or Democratic control of state government will be less likely to adopt a b an on same -sex unions. Similarly, the ideological direction of actors within a venue might make them more responsive to an issue (Brace and Hall 2001; Butler et al. 2015; Desmarais, Harden, and Boehmke 2015; Holyoke, Brown, and Henig 2012 ; Volden 2015). In turn, I control for State Supreme Court Ideology using Bonica and Woodruff™s (2015) measure, where positive scores indicate a more conservative judiciary. Since more liberal state supreme court justices may have a greater penchant for pr otecting minority rights, I predict a positive coefficient for this variable. Successful adoption of a same -sex marriage bill or ban may also depend on public attitudes toward gay marriage ( Enns and Koch 2013; Erikson, Wright and McIver 1993; Lax and Phill ips 2009; Pacheco 201 1, 2014; Wright, Erikson, and McIver 1987). Relying on Lewis and Jacobsmeier ™s (2017) new state -level estimates of Public Support for Gay Marriage from their MRP analysis , I anticipate more favorable attitudes will make the adoption of a ban less likely. Given national, state, and local religious right organizations™ role in advancing bans on same -sex unions throughout the U.S. and given the counter efforts by national and subnational LGBT networks, I also control for state -level in terest groups™ influence on policy change. Despite scholars offering different measures to capture interest -group presence and pressure, no current measures are the same for the religious right and gay rights groups, nor do they cover the full time period. As such, I follow other scholars™ lead (Colvin 2004; Fleischmann and Moyer 2009 ; Lax and Phillips 2009; Lewis 2011) and rely on the percentage of a state ™s population that identifies as Evangelical Christian or member of the Church of Jesus Christ of Latt er -day Saints as a surrogate for the 139 conservative religious groups .58 I proxy gay rights groups ™ presence in a state as the percentage of the population that identifies as LGBT. Both measures come from estimates provided by Taylor et al. (201 9). I suppose a positive coefficient for the Evangelical Population variable and a negative coefficient for the LGBT Population variable. Because policy makes mass politics (Campbell 2012), a state™s prior policy adoptions related to gay rights may also drive a state™s propensity to prohibit or permit same -sex unions. I consider whether a state adopted and had in place a ban on sodomy, which were frequently used to target gay couples engaged in consensual sex. Sodomy Ban takes on a value of 1 if a state still had a ban on the books. 59 Much like Soule (2004), I suppose states with sodomy bans are more likely to adopt an anti -gay marriage policy . I also include whether a state passed LGBT Hate Crime Law to increase penalties for crimes committed against individuals based on sexual orientation. I expect a negative coefficient for this variable since states receptive to seeking justice for gays as a protected class should be less likely to deny marriage equality to LGBT individuals (Earl and Soule 2001). Finally, following past advice on known determinants of policy change related to gay marriage, I also include three state -level demographic contro ls. First, I include the percentage Racial / Ethnic Minority Population for a state since African Americans and Latinos were less supportive of gay marriage than their white counterparts (Colvin 2004; Lewis and Gossett 2008). Second, I consider the percent age of a state™s residents 25 -years old and older that have earned a college degree. As a state™s Population with College Degree increases, I suppose a decrease in the propensity to adopt a ban on gay marriage since higher education breeds greater toleranc e (Barclay 58 Although members of the Mormon Church (Church of Jesus Christ of Latter -day Saints) are not directly correspondent to conservative Evangelical Christians, leaders from both groups oppose gay marriage. Hence, I include both groups in the Evangelical Popul ation measure. By omitting Mormons from the measure, conservative religious influence would be minimized in many Western states. 59 Following the 2003 U.S. Supreme Court Lawrence v. Texas decision, sodomy bans were declared unconstitutional. States with b ans in the dataset prior to 2003 retain their 2003 values through 2015. Even though the Court ruling deemed those policies unconstitutional, those values remain in the dataset as they are indicative of the state™s prior policy context and propensity with r egard to the issue of gay marriage. 140 and Fisher 2003; Fleischmann and Moyer 2009 ; Haider -Markel and Meier 1996, 2003). Third, I control for the natural log of State Population , since more populous states tend to protect minority rights and thus less likely to pass anti -gay measures (Donovan and Bowler 1998). I expect a positive coefficient for the first covariate and negative coefficients for the following two variables. As an additional note, if variables were missing an observation for a given year, linear interpolation was used t o fill the missing value. See Table C.3 in the Appendix for these variables™ descriptions, summary statistics, and sources for the anti -gay marriage models. Methods Early diffusion studies (e.g., Eyestone 1977; Gray 1973) mapped the spread of policies by modeling the cumulative proportion of states adopting the policy at time . The method produced the characteristic sinusoidal curve charting the ratio of states that adop ted the policy, with steeper slopes suggesting a more rapid diffusion and flatter curves indicating a gradual transmission across states. While this approach captured the propagation of policies across jurisdictions, it did not account for the mechanisms t hat might cause policy adoption. Berry and Berry™s (1990) article employing Event History Analysis (EHA) to examine the spread of lottery policies across U.S. states changed that. EHA has now become the tool of choice for documenting and analyzing policy diffusion . It is useful because it accounts for policy adoption in each state in each year (state -year unit of analysis), allowing covariates to distinguish between internal (e.g., state resources, politics, institutional settings, opinion ) and external ( e.g., policy learning, geographic neighbors, federal pressure ) factors (Blossfeld, Golsch, and Rohwer 2007; Box -Steffensmeier and Jones 2004; Buckley and Westerland 2004; Volden 2006). EHA is akin to survival or duration analysis where the model determines the fihazardfl or firiskfl rate of a state pursuing an 141 innovative policy at a given point of time. Once the state has adopted a policy, it is no longer at risk and drops out of the dataset. The flexible approach still produces the distinctive s -curve for the cumulative frequency of policy adoption across the states reminiscent of early studies. And this approach simultaneously controls for internal and external predictors modeling the mechanisms driving the policy diffusion while appropriately treating censored data. Many policy diffusion scholars have relied on logistic regression to estimate this type of discrete -time data, calculating a unit™s likelihood of policy adoption in a given year. For purpose s of comparison with prior diffusion research on this topi c, I start with logistic regression to estimate the EHA data. Logistic regression gets its name because a logit link function is used to specify parameters in terms of the log -odds ratio of the probability of the event occurring to it not occurring. The co efficients then are interpreted relative to the log -odds of the event occurring. But, because the interpretation of log -odds is not always straightforward, I provide predicted probabilities and odds -ratios where appropriate. The probability equation for lo gistic regression is as follows: Pr(=1|)= exp ()1+exp (), where the probability of a state adopting a policy, Pr (=1), is a function of the covariates, , and the coefficients, , are expressed as exponentiated logit parameters for each covariate ( Box -Steffensmeier and Jones 2004; Long 1997). Results from the logistic regression models for the adoption of gay marriage bans are presented below in Models 1 Œ 3. Although standard EHA an d logistic regression are useful fihammers,fl their ubiquity has resulted in scholars treating all cases of policy diffusions as nails. Boehmke (2009 a: 229) put it best that the field fihas reached a point of diminishing marginal returns from the standard EHA model.fl Much of the past diffusion research o n anti -same -sex marriage policies either focus es solely on policy adoption in one venue (e.g., legislature, legislative referendum , and citizen initiative), or treats 142 all gay marriage bans as equal, regardless of the path pursued. Yet the external and internal forces influencing policy change could behave differently for each institutional arena. And there may even be inter -venue dynamics at play affecting the spread of these innovations. 60 To account for this, I rely on multinomial, rather than binary, logistic regression. The multinomial logistic model is a fiseries of ‚linked™ logit modelsfl ( Box -Steffensmeier and Jones 2004; Long 1997). For anti -gay marriage policies, there are four separate avenues ( ) a state could take: (0) No Policy Adoption, (1) Adoption via Legislature, (2) Adoption via Legislative Referendum, and (3) Adoption via Citizen Initiative. Thus, the model estimates three separate equations ( 1 ) which ar e then referenced to a chosen baseline category. In this case, the reference category is fiNo Policy Adoption.fl Although three separate fistand -alonefl logit equations are estimated with the same baseline, the advantage of the multinomial logit in this contex t is that it models the competing risk of states enacting a policy via one available venue over another venue , allowing different covariate estimates for each forum (Boehmke 2009 a; Box -Steffensmeier and Jones 2004; Long 1997) . I prioritize uniqueness over parsimony (Boehmke 2009 a). This approach should unpack the external and internal factors driving policy change in one 60 There have also been more recent adaptations and advances in modeling policy diffusion. Regarding estimators, Cox proportional hazards, complementary log -log, or rare events logit models make fewer assumptions about the functional form of the data generating process (Box -Steffensmeier and Jones 2004). Borrowing from the international relations conflict literature that models disputes between pairs of countries, Volden (2006) introduced the policy diffusion community t o a dyadic approach. Treating states as dyads provides a richer specification of the diffusion process by more fully examining the relationship between states. But the nature of the dyadic data structure also increases the number of zeros in the dataset (G ilardi and Füglister 2008), potentially producing fiapparent emulationsfl in the data even if they do not exist in reality (Boehmke 2009 b). Given this potential risk, and because I am more interested in the adoption via a particular venue (rather than direct interdependence between states), I opt for a monadic data structure. Furthermore, I employ a multinomial logistic regression estimation strategy. Interpreting coefficients for state dyads across multiple venues may prove too challenging. Other scholars ha ve run separate analyses for each policy component, comparing the determinant covariates across the different models. Taylor et al. (2012), for instance, use a two -stage seemingly unrelated regression model to account for the spread of fourteen different L GBT antidiscrimination policies. In the first stage, they estimate separate EHA models for each policy. In the second stage, they rely on join parameter estimates and covariance matrices to calculate the standard errors for the covariates. Using Chi -square d tests, they can check the difference in covariates by policy. This approach allows them to see how each law ™s complexity and content affect its diffusion relative to the other policies. See Boehmke (2009a) for several excellent recommendations for when to perform separate or pooled analyses. 143 venue, as well as help uncover inter -venue dynamics within states. 61 This modeling strategy could be applied to other policy contexts invol ving multiple policy components or venues. 62 I present the probability equation for multinomial logistic regression: Pr(=|)= exp ()1+exp (), where the probability of a state adopting a policy via a given venue , Pr (=1,2,3,4), is a function of the covariates, , all relative to the baseline of fiNot Adopting a Policy.fl The coefficients by venue, , are expressed as exponentiated lo git parameters for each covariate ( Box -Steffensmeier and Jones 2004; Long 1997). Although displaying and interpreting multinomial logistic regression results is more challenging than for dichotomous logistic regression models, maximum likelihood can be use d to estimate the model and the coefficients are interpretable as logit coefficients, though relative to the baseline category. Again, because log -odds coefficients are not intuitive and comparisons across venue categories is necessary, I provide predicted probabilities, odds ratios, or marginal effects where appropriate to ease the interpretation of key covariates. Before proceeding to the results from the binary and then multinomial logistic regression models, a brief discussion about the potential for de pendency in the duration data is required. A state™s passage of a policy via a venue in a given year may generate dependencies across the other venues within the year since states remain at risk in the other forums . To account for this potential 61 Recall that the key independent variable of interest, political learning, as a venue -specific variable takes on different values for the respective venues under consideration and thus different values for the separ ate logit equations. 62 A key assumption of multinomial logit models is that the possible choices are independent of one another: Independence of Irrelevant Alternatives (IIA). Here, for example, that means that the venue choice of legislative referendum is independent from the v enue choice of citizen initiative. I believe there are strong theoretical reasons to treat the choices as separate venues, especially given the separate institutional contexts for each venue. Because of the IIA constraint (and the potential for its violati on), some scholars prefer a multinomial probit model, which does not depend on the IIA assumption, but is susceptible to estimation challenges (Dow and Endersby 2004; Kropko 2008). However, Kropko™s (2008) computer simulations show that multinomial logit m odels provide more accurate results than multinomial probit models fieven when the IIA assumption is severely violated.fl And Dow and Endersby (2004) contend that multinomial probit™s penchant for weak identification can produce misleading findings and sugge st employing multinomial logit. For these reasons, I use multinomial logit models. 144 heterosked asticity within a state -year, I cluster all standard errors by state -year ( Box -Steffensmeier and Jones 2004; Primo et al. 2007). And to correct for temporal dependence Šthat the probability of adoption by a state in one year is related to its likelihood of passage in previous years Šthat may exist, I include a time counter variable (see Beck, Katz, Tucker 1998; Buckley and Westerland 2004). Without this time variable, I would be assuming that the probability of a state adopting a policy in a given year does n ot change over time, which is highly unlikely. Controlling for time has the added benefit of accounting for the possibility that cohort replacement (i.e., the substitution of older generations with younger generations) drives changes in public opinion and support for gay marriage policies (Harrison and Michelson 2017; Lax and Phillips 2009; Lewis and Gossett 2008). Results for Anti -Gay Marriage Policies Table 5.1 contains the results for three separate binary logistic regression models to examine the spre ad of anti -gay marriage policies across the U.S. states from 1993 Œ 2015. As a reminder the, dependent variable for all three models is the likelihood of a state adopting a ban on gay marriage in any given year, regardless of venue. Attempting to replicat e the base findings from prior work on the diffusion of anti -gay marriage policies, Model 1 (fiStandardfl) provides the point estimates for a typical model in the diffusion literature. Essentially, Model 1 analyzes the event history data without tackling the four gaps that I identified from earlier scholarship in this area. 63 From the results in the first column, we see the only external factors that appear to influence a state™s propensity to adopt a statutory or 63 There are, however, three important differences in these models compared to preceding work. First, I include far more external and internal explanations for policy adoption than any prior article. I do so for theoretical motivations and to incorporate pr ior works™ specific contributions in telling the underlying story. Second, in contrast with earlier work that only examined the spread of anti -gay marriage policies via one venue or without regard to venue, the dataset here is comprised of pooled observati ons for each state -venue. Units only exit the venue subset if a state adopts a ban via that venue and remains in the other venue subsets until enacting a ban in those alternative venues. Although venue specific information is included in the data, I am not leveraging this information yet. I will do so in the subsequent analyses. Finally, my data also included repeated events where states may have added additional statutory bans. 145 constitutional ban were the U.S. Supreme Court ™s Lawrence v. Texas decision and presidential election years. Looking at internal factors, state governments controlled by the Democratic Party and with higher public support for same -sex unions were less likely to adopt a gay marriage ban. Of note, neither the policy learni ng nor geographic neighbor variables were statistically reliable at -Markel 2001a). To investigate further the potential role of policy learning, whereby states adopt an anti -gay marriage policy as more and more states enact bans, I rely on the second measurement approach for the Policy Learning Hypothesis. Instead of capturing policy learning as the aggregate number of bans passed by year, I break those bans out by venue, providi ng the cumulative number of bans adopted by venue -year. Model 2 (fiStandard +fl) includes these three separate policy learning covariates along with the pro -gay marriage counter variable, which controls for the gay rights movement™s policy successes. These m ore refined policy learning variables indicate that states were more likely to adopt a gay marriage ban as the number of states enacted bans via legislative referendum and the legislature (although the latter misses statistical reliability at conventional levels). 64 Political party control of state government still seems to matter, with Democratically controlled states still less likely. But the effect from other external and internal factors is not statistically distinguishable from zero. Neither of Model 1 nor Model 2, however, account for the role of political learning, where policy actors gain information about the venues that prior states have used to impede marriage 64 A standard deviation increase (about 7 states) in the number of states ado pting a constitutional ban via legislative referendum increases a state™s odds of adopting an anti -gay marriage policy by a factor of 7.93, holding all other variables constant. For policy learning from legislative action, an increase of nearly 13 states a dopting statutory language outlawing gay marriage increases a state™s odds of adopting a ban by nearly six - 146 Table 5.1: Policy Diffusion of Anti -Gay Marriage Policies using Binary Logistic Regression Explanatory Variables Model 1: Standard Model 2: Standard + Model 3: Standard + Pol Learn Political Learning [+] --- --- 3.652* (0.371) Policy Learning [+] 0.076 (0.046) --- --- Policy Learn from Leg [+] --- 0.137ƒ (0.081) 0.223* (0.083) Policy Learn from Leg Ref [+] --- 0.274* (0.132) 0.283* (0.144) Policy Learn from Cit Init [+] --- -0.123 (0.234) 0.825 (0.585) Geographic Neighbor [+] 0.644 (0.553) 0.790 (0.551) 0.825 (0.585) Federal Government DOMA [ -/+] -0.124 (0.961) -0.465 (1.030) -0.911 (1.243) Lawrence v. Texas Sup. Ct. Decision [+] 1.517* (0.729) 1.749 (1.381) 3.700 (2.00) NYT Issue Salience [+] -0.001 (0.006) 0.007 (0.008) -0.002 (0.010) Presidential Election Year [+] 1.014* (0.459) 0.849 (0.517) 0.964 (0.674) Pro -Gay Marriage Counter [+] --- 0.033 (0.085) 0.207* (0.085) Legislative Professionalism [+] 0.023 (0.110) 0.006 (0.112) 0.018 (0.118) State Supreme Court Professionalism [+] 1.166 (1.210) 1.114 (1.223) 1.016 (1.300) Difficulty Amending Constitution [ -] -0.193 (0.164) -0.183 (0.162) -0.171 (0.169) State Gov. Party Control [ -] -.883* (0.375) -0.953* (0.389) -1.013* (0.406) State Supreme Court Ideology [+] 0.022 (0.295) 0.205 (0.311) 0.215 (0.327) Public Support for Gay Marriage [ -] -0.077* (0.037) -0.034 (0.043) -0.017 (0.047) Evangelical Population [+] 0.008 (0.018) 0.021 (0.019) 0.027 (0.020) LGBT Population [ -] 0.228 (0.365) 0.164 (0.376) 0.151 (0.394) Sodomy Ban [+] 0.137 (0.327) 0.187 (0.326) 0.246 (0.346) LGBT Hate Crime Law [ -] 0.353 (0.373) 0.374 (0.389) 0.412 (0.407) Racial/Ethnic Minority Population [+] 0.012 (0.013) 0.012 (0.012) 0.014 (0.013) Population with College Degree [ -] -0.013 (0.039) -0.017 (0.040) -0.024 (0.043) State Population (Ln) [-] -0.192 (0.216) -0.186 (0.216) -0.208 (0.233) Constant -0.567 (1.912) -1.524 (2.094) -2.032 (2.316) N 2451 2451 2451 2 (20), (23), (24) / Log Likelihood 102.12 * / -276.99 102.22 * / -272.5 7 338.79 * / -246.19 AIC / aROC 595.98 / 0.817 593.14 / 0.827 542.39 / 0.881 , 5, two tailed. Dependent variable is likelihood of adopting anti -gay marriage polic y (irrespective of venue) . Statistically significant logistic regression coefficients are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. 147 equality successfully . Model 3 (fiStandard + Political Learningfl) includes the political learning variable, along with the opposition success and policy -learning -by-venue variables added in Model 2. Political learning™s effect is both statistically significant and substantively large. Increasing political learning by one standard deviation (a 40% rise in success via a given venue) increases the probability of a st ate enacting a gay marriage ban, on average, by 6.8 points. Put differently, moving from an environment where all states fail via one venue to an environment where all states succeed in that venue maximizes political learnings™ marginal effect on policy ad option by 9.4 percentage points. 65 On its face, this effect may not seem impressive. But because the probability of adopting a ban in any given state in any given year is under 3 percent, an increase of nine points is comparably large and substantively mean ingful. In addition to political learning™s effect, Model 3 also shows that the opposition™s policy successes (i.e., permitting same -sex unions) in other states increases the risk of a backlash in other states. For every additional state that allowed gay marriage, the risk of another state prohibiting gay marriage, on average, rose by 0.6 percentage points. Both results suggest that policy actors in one state learn from and react to policy actions in other states perpetrated by their own network and by the opposition. 66 65 This is the Average Marginal Effect (AME) , where the marginal effect of political learning is calculated as the difference in t he two probabilities of all observations at their current values in a state of no political learning (i.e., no state successfully adopts policy via the given venue) and a state of perfect political learning (i.e., all states successfull y adopt policy via g iven venue). I prefer AMEs to mean marginal effects (MEMs) because it relies on all units™ variable values rather than the means of those values. 66 Figure C.1 in the Appendix plots the predicted probabilities of a state adopting an anti -gay marriage pol icy as political learning, policy learning via legislatures, policy learning via legislative referenda, and the cumulative number of enacted pro -gay -marriage policies increase. The figure in the top -left quadrant displays the predicted probability of a sta te adopting an anti -gay marriage policy via any venue in any given year as they learn from the successful paths taken in earlier states. The increase in probability appears linear until a success rate of 80 percent, which causes the slope of the predicted probabilities to take a marked upturn. This drastic change in slope is likely due to gay marriage bans ™ low failure rate , with most bans successfully adopted across the three competing venues. Nevertheless, the increase in probability of a state adopting a n anti -gay marriage policy in any given year rises to 2 percent as political learning spans its full scale. Policy learning via the legislature and legislative referendum (top -right and bottom -left quadrants, respectively) also have an effect. Both predict ed probability slopes are relatively flat until at least 20 states adopt a ban via the legislature and 10 states enact a ban via legislative referendum. Then the predicted effect on policy change appears more acute, with an increase in likelihood of laggar d states adopting a ban of .04 and .12 in any given year as the sizeable number of legislatures and legislative referenda adopt prohibitions , respectively. Finally, the predicted probability slope 148 The upshot of these initial three models is that policy actors appear to learn about the tactics used and paths successfully taken in early mover states, thus making them more likely to succeed in their state. Although states learn about ava ilable solutions adopted in other states (i.e., policy learning), they also become informed about how prior states were successful (i.e., political learning). Policy actors also try to counteract the opposition™s policy success. Including political learnin g and opposition policy success in the models reveals a richer and more accurate understanding of the external forces driving policy change across the states. Indeed, all the indicators of model fit point to Model 3 as being the superior model. 67 Noneth eless, it is worth noting that several of the external and internal factors that previous research found to affect states™ adoption of gay marriage bans do not appear to play a role here. For example, s tates do not appear to emulate their neighbors. Furthe rmore, legislative and state supreme court professionalism, degree of difficulty in amending the state constitution, prior policy activity on gay rights, and demographic controls have a limited influence on the spread of gay marriage bans. 68 Perhaps most su rprisingly, the proxies for the religious right and gay rights movements did not predict policy adoption. There are two po ssible reasons for this. First, it is possible these pressure groups™ influence is captured by the political learning, opposition poli cy success, or policy learning variables. Organized interests, especially national groups, are the most likely candidates to share the political and policy knowledge gained in one state with other states. And because there are no for pro -gay marriage policy success (bottom -right quadrant) follows the typical sinusoidal curve. As more states adopt pro -gay marriage policies, the pro pensity of adopting a ban also increases. However, causality should not be inferred from this relationship, as both anti - and pro -gay marriage policy wins increas ed over time. 67 McFadden™s pseudo R2 increases from 0.157 in Model 1 to 0.25 in Model 3. The Akaike information criterion (AIC) also drops across the models, from a high of 595.98 in Model 1 to a low of 542.39 in Model 3, indicating that Model 3 is a highe r quality model and may better represent the data -generating process. As further evidence, the area under the Receiver Operating Characteristic (aROC) curve improves in accuracy across the three models. 68 It is worth mentioning that even though these varia bles were not statistically significant, the signs of their coefficients were largely in the anticipated direction. Of course, there were a few exceptions to this including the LGBT population and hate -crime legislation variables. Congress ™s passage of DOM A also appeared to dampen rather than incite a state's propensity to adopt a gay marriage ban . Again, this may be because states ™ response s were more acute than the federal government ™s response. In fact, it is possible that the heightened degree of subnational activity had a snowball effect on the national government™s decision to act. 149 consistent longitudinal me asures for the religious right and gay rights movements at the national level, and because these groups are not monoliths, I do not control for their national -level influence. Second, as previously discussed, the binary logistic regression model is not the best strategy for estimating a dynamic policy process happening via multiple, competing venues. Modeling a state ™s propensity of adopting anti -gay marriage policies without regard to the venue where such policies are adopted is problematic. Policy actors can pursue bans on same -sex unions via multiple , competing institutional arenas Šstate legislatures, legislative referenda, and citizen initiatives Šin the same year. Thus, states are not merely at risk of enacting a gay marriage ban, but rather at ri sk of adopting a gay marriage ban via available venues at multiple points in time . Yet the current and typical EHA modeling strategy, binary logistic regression, fails to produce the respective probabilities and coefficients associated with adopting anti -gay marriage policies via competing arenas ( Blossfeld, Golsch, and Rohwer 2007; Box -Steffensmeier and Jones 2004; Buckley and Westerland 2004; Cann and Whilhelm 2011). As discussed above in the Empirical Estimation Strategy section, I believe the more appro priate and informative modeling scheme is to employ a multinomial model. Although not as parsimonious as estimating a single coefficient for each variable, I leverage the competing venues to estimate separate coefficients for each independent variable by arena . This shed s light on which variables matter in the diffusion processes by institutional avenue. Because the data include repeated events and states are at risk of selecting one venue over another (Cann and Wilhelm 2011), I use a repeated events compet ing risk multinomial logistic regression model to estimate the population parameters. In addition to estimating the data with multinomial logistic regression, I also add two new state -level controls. First, a prime assumption of multinomial logit is that e ach state is at risk of selecting each venue. To allow for this and acknowledge that 26 states do not permit direct or indirect citizen initiatives, I include a dummy 150 variable, Direct Democracy , to control for this difference and retain the non -direct -demo cracy states in the dataset. Moreover, it allows me to examine further whether and how direct democracy states may behave differently than their counterparts (Boehmke 2005; Bowler and Donovan 2004; Lewis 2011, 2013). I anticipate that direct democracy stat es will be more likely to pass a ban on gay marriage, especially via citizen initiative. 69 The second control that I include is whether a state adopted a prior anti -gay marriage policy via another venue. The Prior Anti -GM Policy variable is a running tally for the prior policy events on this issue in other venues. I anticipate that states that previously adopted a ban on gay marriage in other arenas will be less likely to enact in the respective venue. Both variables should better disentangle the inter -venue dynamics at play. Table 5.2 includes the repeated -events, competing risk multinomial logistic regression results for the adoption of gay marriage bans via state legislatures, legislative referenda, and citizen initiatives, relative to the reference catego ry of not passing an anti -gay marriage policy. 70 Although not reflected in the table, the average probability of a state adopting a gay marriage ban in any given year was quite low: 1.6 points for an enactment via the legislature, 0.7 for passing a ban via legislative referendum, and 0.6 for outlawing gay marriage via citizen initiative. However, as the results show, multiple factors increased or decreased a state™s propensity to adopt an anti -gay marriage policy via a particular venue. To aid in the interpr etation of the multinomial logistic regression coefficients, I present the Average Marginal Effects for the main predictors of states adopting a gay marriage ban in Figure 5.3. 69 An additional institutional variable could be included to control for the fact that Delaware is the only state that does not allow referenda referred by the legisla ture to ensure they are not at risk of adopting a ban via referendum. Including a Delaware dummy variable does not change the results in any of the models. 70 Likelihood ratio tests indicate that almost none of the venue choices should be combined, exce pt for the possible combination of legislative referendum and citizen initiative when compared to each other. And despite known problems with IIA tests and their irreproducibility (Allison 2012; Cheng and Long 2007; Dow and Endersby 2004), I carry out seve ral IIA tests. By and large, the results suggest there are no violations of the IIA assumption. We can also have a great deal of confidence in the model fit given a McFadden™s pseudo R 2 of 0.497, and an aROC curve statistic of 0.928, suggesting a very high model classification. 151 Table 5.2: Policy Diffusion of Anti -Gay Marriage Policies using Mult . Logistic Regression Explanatory Variables Legislature Leg. Referendum Citizen Initiative Political Learning [+] 4.001* (0.392) 61.930* (13.628) 1.277 (1.386) Policy Learn from Leg [+] -0.010 (0.608) -1.965* (0.594) 2.279 (1.466) Policy Learn from Leg Ref [+] -0.569 (0.495) 0.811* (0.374) 0.359 (0.553) Policy Learn from Cit Init [+] -0.308 (1.841) -0.438 (0.796) -0.227 (0.895) Geographic Neighbor [+] 0.531 (0.899) 3.006ƒ (1.803) - 9.907* (2.970) Federal Government DOMA [ -/+] 2.876 (8.001) 71.933* (17.616) -12.232 (19.876) Lawrence v. Texas Sup. Ct. Decision [+] -3.231 (8.132) 7.950ƒ (4.224) -3.408 (4.103) NYT Issue Salience [+] 0.089 (0.120) 0.019 (0.029) 0.014 (0.025) Presidential Election Year [+] -0.635 (1.588) 1.078 (0.615) 2.312 (1.454) Pro -Gay Marriage Counter [+] -1.957 (2.699) 0.050 (0.217) -0.382 (0.762) Legislative Professionalism [+] -0.149 (0.179) 0.160 (0.405) -0.397 (0.611) State Supreme Court Professionalism [+] 4.313ƒ (2.247) 4.847 (5.194) -8.200 (7.703) Difficulty Amending Constitution [ -] 0.177 (0.682) 0.022 (0.584) -20.892* (4.753) Direct Democracy [ -/+] 0.778 (0.657) -0.694 (1.142) 17.256* (3.427) State Gov. Party Control [ -] -1.752* (0.682) -0.203 (1.077) 4.184ƒ (2.316) State Supreme Court Ideology [+] 0.070 (0.446) 1.753ƒ (1.056) -5.470* (2.077) Public Support for Gay Marriage [ -] -0.146 (0.103) -0.137 (0.151) -0.122 (0.112) Evangelical Population [+] 0.004 (0.035) 0.107ƒ (0.064) -0.485* (0.125) LGBT Population [ -] -0.479 (0.948) 0.133 (0.852) -4.718* (1.846) Prior Anti -GM Policy [ -] -2.324* (0.968) -0.847 (0.573) -6.126* (1.282) Sodomy Ban [+] 0.569 (0.575) 0.577 (1.006) 0.741 (1.519) LGBT Hate Crime Law [ -] 0.278 (0.844) 1.130 (1.076) -0.408 (1.403) Racial/Ethnic Minority Population [+] 0.013 (0.027) 0.085* (0.042) -0.090 (0.914) Population with College Degree [ -] 0.016 (0.074) 0.187 (0.130) -0.316ƒ (0.179) State Population (Ln) [ -] 0.171 (2.344) -1.358 (0.839) 1.638 (1.456) Constant -6.243 (6.022) -77.813* (14.245) -39.161* (19.689) N 2451 2 (78) : 2126.34 * AIC / aROC 566.00 / 0.928 Log Likelihood: -202.00 Repeated -events competing risks model estimated using multinomial logit model. D V is likelihood of adopting anti -gay marriage policy by venue. D V has four categories ; baseline category is not adopting an anti -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state -year, are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted due to space considerations. The hypothesized direction of th e IV effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. 152 Recall from Chapter 3 that a verage marginal effects for discrete variables can be interpreted as the difference between being in one hypothetical state (e. g., pre -Lawrence decision) and being in another hypothetical state (e.g., post -Lawrence decision), with all the other covariates held at their same values. And average marginal effects for continuous variables can be interpreted as the instantaneous rate o f change in the dependent variable following a small (i.e., unit) increase in the explanatory variable. The central predictor of interest here, political learning, the process of gaining tactical knowledge from other states about the most favorable venue t o achieve policy change, influenced a state™s likelihood of enacting a gay marriage ban. The average marginal effect of political learning on a state™s likelihood of prohibiting same -sex marriage via the legislature was 3.9 percentage points. Political lea rning appears to have an even larger impact on states adopting proscriptions via legislative referendum , with an astounding marginal effect of 33.8 percentage points . The gap in political learning ™s effect between legislatures and legislative referendum is not surprising since legislators may have required more political information from prior states to pass a constitutional amendment than statutory language. As expected, the political learning variable for the citizen initiative venue is positive , but it is not statistically reliable , and its marginal effect is not distinguishable from zero . Perhaps, because fewer than half of the states permit citizen -driven ballot measures, direct democracy states are less able to look to and learn from other direct democracy states. It also may be that, given the varying requirements across direct democracy states to get an initiative on the ballot, the transfer of political knowledge regar ding this venue is attenuated. As we can see from Figure 5.3, political learning™s influence on policy adoption outweighed all other external factors, except for the post -DOMA variable . Political learning appears to play an even larger role than policy le arning and other known determinants of policy diffusion. For 153 Figure 5.3: Average Marginal Effects for Key Anti -Gay Marriage Policy Predictors predicted probability plots of political learning™s impact on the adoption of anti -gay marriage policies acros s the three venues, see Figure C.2 in the Appendix. Turning to the other external factors, policy learning matter ed for adopting a ban by way of legislative referenda. Some states were less likely to prohibit gay marriage via legislative referendum as more state legislatures enacted statutory bans. Perhaps those states may have been willing to pass statutory language prohibiting gay marriage but may have been less willing to enshrine such language in their constitutions. Still, other states were 0.5 percent age points more likely to adopt a constitutional ban via legislative referendum in any given year as the number of other states going that route increased. States also took cues from their neighbors, at least when considering whether to pass a constitution al ban on gay marriage. A 40% increase in the proportion of neighbors enacting a ban via legislative referendum raised a state™s propensity to adopt a similar measure by one percent in a given year. Direct democracy states, however, were less likely to pur sue a ban via 154 citizen initiative as their direct -democracy neighbors did so, to the tune of 0.6 percentage points in a given year with the 40% shift . Federal government activity also appears to have spurred state -level activity in some arenas. Congress ™s passage of DOMA in 1996 increased a state™s marginal probability of adopting a ban via legislative referendum by 39.4 percentage points. While the U.S. Supreme Court™s Lawrence decision further augmented a state™s risk of passing a ban via legislative referendum. States also were more likely to adopt a legislative referendum during presidential election years too. Considering the internal factors influencing policy adoption, s tates ™ own institutional arrangements played a role. Direct -democracy states with higher hurdles to amending their constitutions were less likely to adopt a constitutional ban via citizen initiative. A one -unit shift on the four -point difficult y in amending constitution scale (with 1 being direct democracy states with only signature requirements and 4 being passing a legislative majority and voter supermajority), decreased the likelihood of a state adopting via citizen initiative by 0.6 points in a given yea r. States with more professionalized supreme courts also saw greater activity in their legislatures; perhaps policymakers passed statutory bans to signal their policy preference to the judiciary (Barclay and Fisher 2008). Internal political factors similarly help explain a state ™s decision to oppose gay marriage. Democratically -controlled states were less likely to outlaw same -sex marriage via the legislature than Republican -controlled state governments. As a result, states with Democrats in charge of the state capitol were more likely to see a citizen initiative adopted. Even shifting from a bipartisan controlled government to a Democratically -controlled government increased the risk of passing a ban via citizen initiative by 0.9 points in a given year. Interestingly, the results also indicate that states with a more conservative supreme court were less likely to pursue a ban via citizen initiative , but more 155 likely to pursue one via legislative referendum . Knowing their supr eme court was more conservative, legislators may have felt more confident putting a ban up for a vote to the state electorate. State interest group strength and capacity also comes into clearer focus with the multinomial model. S tates with a higher Evangel ical and Mormon population s were more likely to adopt a constitutional ban via legislative referendum than citizen initiative. And states with a greater number of residents identifying as LGBT were also less likely to enshrine a traditional definition of m arriage into their constitutions. A one percent increase in a state ™s proportion of LGBT residents decreased its probability of adopting a ban by 0.6 percentage points in a given year. A state ™s pr evious policy activity on the issue also determined subsequ ent policy action. States that had adopted a statutory ban via the legislature were less likely to do so again (despite states like Idaho, Texas, Utah, and Virginia passing multiple pieces of legislation). And direct -democracy states that had previously passed a statutory ban lowered their marginal probability of adopting another ban via citizen initiative by 2.5 percentage points. That said, the gay rights movements™ cumulative policy successes appear to bear little on the religious right™s successes in th e state, at least when adoption is broken down by venue. 71 Overall, by l everaging multinomial modeling , I have refine d our understanding of the diffusion process for anti -gay marriage policies. The main reason for mixed findings from past studies on this t opic is largely because venue type was omitted from the models. Once the arena is accounted for, we can better see which factors drive policy change in these respective venues. Regardless, the upshot from these results is that policy actors are more likely to adopt a policy in a given venue as the y learn about other states successfully adopting the policy in that venue. Such 71 A state ™s demographic context similarly helps explain a policy change. As a stat e™s proportion of racial and ethnic minorities increased, the state was more likely to adopt a ban via legislative referendum. This could be due to low er levels of support for gay marriage among African Americans and Latinos, or this variable could serve as a proxy for southern states. Education also appears to play a role, with more educated states less likely to adopt a ban via citizen initiative (alth ough the coefficient just misses the 156 influence occurs even when considering alternative external factors and states™ own institutional, political, interest group, policy, and demographic contexts. Diffusion of Pro -Gay Marriage Policies The previous empirical results offe r initial evidence that political learning affects states™ likelihood of passing anti -gay marriage policies. But what explains the spread of pro -gay marriage policies across U.S. states? Although more gradual and later than the religious right™s countermob ilization against gay rights, the pro -gay marriage movement pushed for equality via the state courts, state legislatures, and finally in the federal courts. This section is dedicated to unpacking the external and internal forces driving the legalization of same -sex unions. The same hypotheses detailing the external mechanisms driving anti -gay marriage policies mostly apply here too, although for many of the hypotheses I anticipate a reverse outcome (e.g., Democratically -controlled states more likely to p ass same -sex unions, greater LGBT population more likely to permit gay marriage). Likewise, I rely on many of the same variables from the anti -gay marriage models for the pro -gay marriage models , except for the following four changes. First, I opt to combi ne policy learning into one variable for the sake of parsimony . Second, for the federal -government involvement (H 4) hypothesis, instead of DOMA and Lawrence variables, I rely on a binary U.S. v. Windsor Sup Ct. Decision variable because the 2013 precedent -setting ruling declared a portion of the federal DOMA unconstitutional. 72 This case likely encouraged more federal courts to grant gay marriage in states filing lawsuits and may have spurred action in other venues. Third, I employ the Prior Anti -GM Policy v ariable to capture opposition policy success and 72 I do not include the 2003 U.S. Supreme Court™s Lawrence v. Texas and presidential election year variables in the analyses. Both variables perfectly predict dozens of observations mak ing it challenging for the model to estimate parameters™ standard errors and the model™s overall likelihood -ratio test. Given this, I opt to exclude them from the analyses. Nevertheless, including them in the model does not affect the key takeaways. 157 a Prior Pro -GM Policy to account for the gay rights movements™ prior policy wins in a state . Lastly, I control for federal court ideology within a state. Since states are at risk of enacting gay marriage via the federal courts, I also control for the mean District Court Ideology of federal district -court judges, using Bonica et al. ™s (2017) aggregated measure at the state level. More positive scores indicate a more conservative district court. In turn, I anti cipate that states may be less likely to see same -sex marriage with more conservative federal courts. 73 Table C.4 in the Appendix provides descriptions, summary statistics, and sources for the variables included in the pro -gay marriage models. To empirically test the diffusion of pro -gay marriage policies, I similarly rely on a pooled dataset of state -years by venue, where states are at risk of allowing same -sex marriage (or the equivalent via civil unions) through the legislature, state courts, or federal courts from 1993 to 2015. Again, as is typical of event history data, a state takes on a value of zero until it successfully enacts gay marriage in a given venue, when those state -venue -years convert to a one and drop out of the dataset. 74 However, states remain in the dataset for other available forums , and if they pursue additional gay marriage policies (e.g., change from civil unions to full marriage equality) via the same venue. Followi ng the modeling logic for the diffusion of anti -gay marriage policies, I similarly rely on a repeated -events, competing -risks multinomial logistic regression model to estimate the external and internal coefficients for the spread of marriage equality polic ies across venues. 75 73 I do not control for the difficulty in amending the state ™s constitution or whether states permit direct -democracy since those are not current venues under consideration. That said, robustness checks that add those variables do not alter the overall findings. 74 Success here is defined as successful enactment (not just pass) of pro -gay -marriage policy. If a federal district court ruled in favor of gay -marriage, but the circuit court stayed the case and never allowed its implementation, the unit remains coded a z ero. The one exception is for the 1993 Baehr v. Lewin case in Hawaii; although that case did not result in the successful enactment of same -sex marriage, the partial success from the Hawaiian Supreme Court (remanding it back to the trial court) led to prec ipitation of pro - and anti -policy activity across the states. This unit takes on a value of 1 in 1993, but a 0 in 1999 for the repeated event when the Hawaiian Supreme Court rules against gay marriage because the electorate adopted a constitutional amendme nt prohibiting same -sex unions. 75 As is good practice, I also estimate robust standard errors clustered on state -year and include a time counter variable to guard against heteroskedasticity of error within state -years and temporal dependency ( Beck, Katz, Tucker 1998; Box -Steffensmeier and Jones 2004; Buckley and Westerland 2004 ; Primo et al. 2007). 158 Results for Pro -Gay Marriage Policies Table 5.3 contains the results from the repeated -events, competing -risks multinomial logistic regression model of pro -gay marriage policies via state legislatures, state courts, and federal courts , with fiNo Policy Adoptionfl as the baseline category. 76 Although not displayed in the table results , the probability of a state adopting a pro -gay marriage policy via any venue in any given year is quite low. The average risk of a state adopting via the leg islature or federal courts in a given year is 0.006, while the probability via state courts is only 0.003. This is not shocking since policy inaction is the status -quo. As before with the spread of gay marriage bans, political learning plays a central rol e in the diffusion of same -sex -union policies. A one -standard -deviation increase (36%) in states™ success rate via legislatures increases a subsequent state™s likelihood of allowing gay marriage via the legislature by 4.5 percentage points in a given year. The same positive shift in the success rate for the federal courts augments a state™s propensity to adopt a pro -gay marriage policy by 0.1 percentage points in a given year. Although suggesting a positive relationship between political learning and legali zing gay marriage in state courts, the coefficient is not statistically reliable. It is possible that given the low success rate in prior state courts (only 53% overall), subsequent states may have been less sure of their chances via their judiciary. Figur e C.3 in the Appendix provides the predicted probability plots 76 As with the anti -gay marriage multinomial model, I test the Independence of Irrelevant Alternatives (IIA) assumption. The test results suggest that the separate venues to pursue pro -gay marriage policies are not independent and the categories should be combined. However, there are strong theoretical reasons to treat the choices as separate venues, especially given the separate institutions for each venue. Furthermore, there are known problems with IIA tests and their irreproducibility (Allison 2012; Cheng and Long 2007; Dow and Endersby 2004), thus I put little weight on these test statistics. I believe we can have a great deal of confidence in these resul ts as well. Nearly 54% of the variance in the dependent variable ( R2 159 Table 5.3: Policy Diffusion of Pro -Gay Marriage Policies using Mult . Logistic Regression Explanatory Variables Legislature State Court Federal Court Political Learning [+] 14.643 * (4.994 ) 1.215 (0.837) 0.882ƒ (0.463) Policy Learn [+] -0.192* (0.08 4) -0.029 (0.084) 0.337 * (0.0 97) Geographic Neighbor [+] -1.781 (2.058) -3.735* (1. 673) 2.966* (1.517) U.S. v. Windsor Sup. Ct. Decision [+] 3.040 (1.350) 2.812 ƒ (1.699) -1.502 (2.328) NYT Issue Salience [+] 0.004 (0.010) 0.006 (0.008) 0.0 37* (0.0 12) Prior Anti -GM Policy [ -] -0.028 (0.635) -0.510 (1.400) 0.642 (0.674) Legislative Professionalism [+] 0.015 (0.623) 0.367 (0.366) 0.416 (0.582) State Supreme Court Professionalism [+] 3.473 (5.539) 0.659 (5.174) -5.768 (4.190) State Gov. Party Control [+] 0.512 (1.439) -0.287 (1.916) 1.289 (1.529) State Supreme Court Ideology [ -] -2.217 (1.822) -0.912 (1.000) -1.935 ƒ (1.140) District Court Ideology [ -] 0.795 (1.082) -1.327 (0.911) 1.391* (0.672) Public Support for Gay Marriage [+] 0.070 (0.073) 0.000 (0.066) 0.220* (0.060) Evangelical Population [ -] -0.078 (0.080) -0.155 (0.147) 0.128 * (0.0 59) LGBT Population [+] 1.193 (0.736) -0.748 (0.655) -0.533 (0.633) Prior Pro -GM Policy [ -] -3.259* (0.878) -1.503 (1.115) -3.700* (1.022) Sodomy Ban [ -] -14.102* (1.110 ) 0.689 (1.509) 1.105 (0.906) LGBT Hate Crime Law [+] 16.539 * (1.177 ) 2.017 (1.242) 0.445 (1.089) Racial/Ethnic Minority Population [ -] -0.039 (0.044) -0.009 (0.044) -0.000 (0.028) Population with College Degree [+] 0.086 (0.139) 0.074 (0.151) 0.010 (0.100) State Population (Ln) [+] -0.448 (0.881) -0.718 (0.539) 0.583 (0.606) Constant -38.673 * (7.041 ) -3.322 (6.151) -21.486* (5.805) N 3253 2 (63) : 4745.14 3* AIC / aROC 410.12 / 0.741 Log Likelihood : -139.06 . Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of adopting pro -gay marriage polic y by venue. Dependent variable has four categories, baseline category is not adopting a pro -gay marriage polic y. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. for enacting pro -gay marriage policies by forum as political learning increas es. 77 77 The risk of adopting via state legislatures is relatively flat until 80% of states are successful that route when the probability takes a drastic turn upward. The probability of adopting via state and federal courts is almost linearly related to other states™ success. As political learning increases, states are more l ikely to adopt via those venues. 160 From Figure 5.4, we can see political learning™s marginal effect on the probability a state allows same -sex unions. Although political learning exhibits a relatively smaller effect compared to the anti -gay marriage context, its impact is still substantively meaningful. The average marginal effect on enactment via state and federal courts is 0.2 and 0.3 points, respectively (although both just miss gay marriage via the state legislature is 4.9 percentage points. The only variable with a slightly larger marginal effect was whether a state had previously adopted an LGBT hate -crime law. Looking at the other external factors, states were additionally susceptible to the cumulative adoption of pro -gay marriage policies across the U.S., especially in state legislatures and federal courts, although the marginal effect was relatively small. 78 Meanwhile, states also paid attention to Figure 5.4: Average Marginal Effects for Key Pro -Gay Marriage Policy Predictors 78 For an increase of one additional state allowing same -sex unions via the legislature, subsequent states were 0.1 percentage point less likely to also adopt via the legislature. States were, however, more likely to legaliz e gay marriage via the federal courts as more states granted same -sex unions. Twelve additional states permitting gay marriage (a one -standard -deviation shift) increased a state™s propensity of guaranteeing LGBT minority rights via federal courts by 5.7 percentage points. This policy learning effect is likely due to the U.S. Supreme Court™s Windsor decision declaring the federal DOMA law unconstitutional. Subsequent district and circuit courts relied on that precedent to rule in favor of marriage equality. Given this federal intervention, policy actors may have seen the federal courts as a more favorable venue or legislators may have felt less pressure to respond to interest group activity as another viable outlet emerged. 161 their neighbors™ policy activity. State judiciaries were 0.2 percentage points less likely to permit same -sex unions as a quarter of a state™s neighbors allowed gay marriage via the state courts . Perhaps judges witnessed the political fallout for having allowed gay marriage in neighboring states (e.g., recall of judges in Iowa following state supreme court™s affirmative decision in 2009), and, consequently , were fearful of succumbing to the same fate. The geographic neighbor variable for the federal courts is also positive and significant. This result is likely an artifact of multiple states belonging to the same federal circuit. Where circuit courts upheld or overturned lower federal -distri ct -court decisions, these rulings impacted multiple states within the same circuit. Thus, the regional effect of looking to and learning from neighbors may be inflated here. Interestingly, the U.S. Supreme Court 2013 Windsor decision made state supreme cou rt™s slightly more likely to legalize gay marriage, while the federal courts were less likely to grant marriage equality. It is possible the variance from the Windsor decision is correlated with the policy learning and geographic neighbor variables, thus d isguising the ruling™s effect on the federal courts. 79 Turning to internal factors, states™ institutional settings appear to explain little in the successful enactment of pro -gay marriage policies. But states™ political contexts offer more leverage. Broade r state public support for gay marriage is predictive of success in the federal courts; a ten percent increase in the public™s approval of gay marriage increases a state™s likelihood of enacting same -sex unions via the federal courts by nearly two points. Unexpectedly, it appears that a state™s probability of successfully enacting gay marriage via the federal courts increases as the district courts become more conservative . Importantly, this is relative to the baseline category of not enacting gay marriage, rather than compared to the other venues. Still, this result may be due to the broad measurement strategy using the aggregate of federal district judges™ ideologies in the state, estimated 79 with the Windsor ruling. Although states were unlikely to achieve success via the federal courts because of heightened national attention, the increase in salience is ass ociated with a higher probability of success via the federal courts. 162 by Bonica et al. ™s (2017). Or it could be due to the model not acc ounting for the corresponding circuit court™s ideology, where district court rulings were upheld or overturned. Regarding the role of state interest groups, states with higher Christian Evangelical and Mormon populations were less likely to see pro -gay marriage activity via the state legislature and state high court. As a result, policy actors in those states appear to have turned to the federal courts. A ten percent increase in a state™s Evangelical population augmented a state™s risk of legali zing gay marriage via the federal courts by one percentage point in a given year. Essentially, where gay rights activists perceived state -level venues foreclosed to them because of greater countermobilization, they turned to federal -level venues that may h ave been more receptive, especially following the Windsor decision. Yet, because the other coefficients for Evangelical Population and LGBT Population are not statistically significant does not imply organized interests did not play a role in the diffusion of gay marriage policies. On the contrary, it is possible state - and national -level interest groups™ role is partially captured via the political learning, policy learning, and prior pro -gay marriage policy variables. 80 Finally, a strong predictor of a st ate ensuring marriage equality via its legislature is the state™s prior policy activity on anti -sodomy and LGBT hate crime laws. States banning consensual gay sex had a 4.7 percentage point lower average marginal risk of enacting gay marriage than states w ithout sodomy bans. At the same time, states that had previously passed an LGBT hate crime law had a 5.5 percentage point higher marginal probability of adopting same -sex marriage compared to states without such measures. Taken as a whole, the adoption of pro -gay marriage policies across the U.S. states was due to a combination of internal and external factors. Although not as evident as in the 80 The latter variable controls for the number of pro -gay marriage policies adopted by the state in other venues (rather than the one under consideration). It turns out that as states have legalized some aspect of gay marriage (e.g., civil unions) in one venue, they are less likely to do so in the others. For every additional pro -gay marriage policy adopted by a state in another venue, a state ™s probability of adopting an equivalent policy via the legislature, state court, and federal courts decreases by 0.5, 0.2, and 0.6 points, respectively. This is to be expected since prior adoption arguably makes future adoption unnecessary. 163 anti -gay marriage models, political learning played a major role in the adoption of pro -gay marriage policies across, especially in state legislatures and federal courts. Robustness Checks Are these results robust to different modeling strategies? Given the discrete nature of the dependent variable and uncertainties about the exact parametric relationship between the variables , I estimated the population™s coefficients usin g a complementary log -log model and Cox proportional hazards model (Box -Steffensmeier and Jones 2004 ; Buckley and Westerland 2004 ). As explained in Chapter 3, the complementary log -log allows for the estimation of rare events, while the Cox model makes no assumptions about the functional form of the hazard rate. Results for the anti -gay marriage and pro -gay marriage models are in the Appendix (Tables C.5, C.6, C.7, C.8). The results are mostly consistent with the findings presented here. 81 Still, p erhaps the se results are due to the measurement decisions for several of the discrete or time -varying explanatory variables. As a further robustness check, I used different operationalizations for dozens of the determinants and re -ran the main multinomial logistic r egression models. 82 None of the different operationalizations nor the new 81 Some readers may also be concerned about the clustering of e rrors to account for potential heteroskedasticity at the state -year level rather than the state level (Cameron and Miller 2015). Typically, such a narrow clustering would result in a cluster of one observation (which makes the clustering irrelevant). But i n this case, one cluster represents four possible policy options (i.e., no action, action via the legislature, action via legislative referendum, and action via citizen initia tive) for a state in a given year. Because there is variation within a state year to make these decisions, I opt to cluster at the state -year level, following the lead of prior research (Karch et al. 2016; Makse and Volden 2011; Shipan and Volden 2006). Nonetheless, Tables D.9 and D.10 in the Appendix reveal that clustering the models™ standard errors at the state level do not lead to substantively different results. 82 For example, replacing Bowen and Greene ™s (2014) first dimensional measure of legislative professionalism with their second axis measure; using Shor and McCarty ™s (2011 ) state house and state senate chamber ideological measures instead of party control; including different measures of citizen ideology (Berry et al. 2010; Enns and Koch 2013); swapping Bonica and Woodruff ™s (2015) state supreme court ideology measure for Windett et al.™ s (2015); depending on different proxies for religious right and gay rights interest group strength (Button et al. 1997; CenterLink (2016); Conger and Djupe 2016; Equality Federation Institut e and Movement Advancement Projects; Family Re sea rch Council; Taylor et al. 2019); and employing different operationalizations for effectiveness of direct democracy (Bowler and Donovan 2004; Lewis and Jacobsmeier 2017) or difficulty in amending state const itution (Lutz 1994) did not lead to different takeaways. Furthermore, the inclusion of other internal forces also known to affect policy change, such as House speaker power (Mooney 2013), state term limits (Miller et al. 2018; Sarbaugh -Thompson 2010), legi slative polarization (Conger and Djupe 2016 ), electoral competitiveness (Ranney 1976); election of state high court judges (Hume 2011); and policy innovativeness (Boehmke and Skinner 2012) did not al ter the overarching findings. 164 measures yielded substantively different conclusions. In short, I believe these findings are robust to various modeling and measurement strategies. Is Political Learning Simply Policy Learning? A key question is whether the operationalization of political learning is simply a surrogate for policy learning. Policy research has paid much more attention to policy learning, the gaining of information about the policy problem, solut ions, and implementation, than political learning, the drawing of lessons about how best to work within and pursue change via a policy process. Scholars have operationalized policy learning as the cumulative number or proportion of total states adopting th e innovation ideas (Gilardi 2010, 201 6; Shipan and Volden 2008) , although later research has prioritized the effectiveness of an innovation (Gilardi 201 6; Shipan and Volden 2014; Volden 2006). Parsing political learning from policy learning is a challenge since both involve some aspect of the success of the policy process: success rate of those that attempted vs. success rate of all units at risk. When a policy is pursued in only one institutional venue, disentangling political and policy learnings™ effect from each other could prove problematic. In fact, I expect that political learning™s effect would be largest when a policy is pursued in only one venue. Fortunately, when a policy is pursued across multiple venues, we can leverage this structure to establ ish empirically each components™ contribution to policy diffusion. In examining the correlation between political learning and policy learning in the spread of anti -gay marriage policies, we see a strong association between the two variables with a Pearso n™s -gay marriage correlation between the two variables in both policy are as, these Pearson correlations also suggest that I am tapping into different, albeit related, latent concepts. Modeling the competing venues 165 further helps separate these variables™ effect across arenas . And hopefully, by controlling for other external and internal factors, any initial shared variance between these mechanisms can be allocated to their respective components. Conclusion This chapter demonstrated the utility of mapping the spread of policies across multiple venues, which is a cornerstone of American federalism and is occurring with higher frequency. I also underscored how multinomial logistic regression can help trace this process across such venues. Not only is multinomial logit est imation a more appropriate modeling strategy from a theoretical perspective as it captures policy change across competing institutions , but it also provides a more complete understanding of the diffusion dynamics occurring within each arena. This chapte r also established political learning™s role in the policy diffusion process, at least in the fight over marriage equality. Opponents and proponents of same -sex marriage, alike, learned from the tactics used and paths successfully taken in prior states to achieve successful policy adoptions in their states. Indeed, the effect from political learning is as substantively large as, if not larger than , other external forces driving the spread of gay marriage policies across U.S. states. What is more, political learning™s effect was largest in the legislative context where elected officials pursued either statutory or constitutional bans on gay marriage or affirmative bills to legalize same -sex unions. We should expect elected officials to be particularly politic ally and electorally conscious (Mayhew 1974). Despite establishing political learning™s impact on successful policy adoption via a given venue, it remains unclear whether learning about the successful paths taken in prior states makes subsequent states mo re likely to pursue the same forum . That is, political learning may help states achieve policy adoption in a venue, but does political learning also increase states™ propensity to 166 pursue the policy change via the same venue? Political learning aids in the spread of public policies, but does political learning also aid in patterns of venue shopping across states? Relying on the same policy case of gay marriage, the next chapter attempts to answer these questions. 167 CHAPTER 6: THE DIFFUSION OF V ENUE CHOICE The last chapter established two considerations for future policy diffusion research. First, policy diffusion research should account for and model the spread of innovations across multiple, competing venues. Doin g so can yield new insights into the dynamics of policy change. 83 Second, future policy diffusion research should consider political learning™s role in policy change. In the context of same -sex unions, states appear more likely to prohibit and permit gay ma rriage as they learn about the successful campaign tactics and venue shopping decisions made in prior states. Indeed, political learning™s effect held even after accounting for states™ internal institutional, political, interest group, and demographic char acteristics. And political learning™s influence on policy adoption outweighed the impact of other established diffusion mechanisms. In short, learning how to advance a policy solution successfully is as important as learning about available solutions . Policy actors appear to consider the policy solutions and the political processes other states use to realize those solutions. While learning about the successful routes previously taken makes states more likely to adopt a policy, does it make them more likely to pick the same venue to pursue the policy? That is, do states choose a venue to press for policy change because prior states have chosen that venue ? This chapter seeks to answer that question. Based upon my new theory of venue diffusion , whereby a state™s choice of venue to pursue an innovation influences subsequent states™ venue choice to pursue the policy, I believe the answer to the question is fiyes.fl If the fundamental premise of policy diffusion is that one state™s adoption makes ens uing states more likely to adopt, then it stands to reason that one state™s venue shopping may condition subsequent states™ venue shopping processes. 84 I charge that political 83 Although modeling the di ffusion of anti -gay marriage policies via state legislatures, legislative referenda, and citizen initiatives did not overturn past findings. It did help clarify our understanding, resolving prior inconsistent results. P olicy learning and federal government involvement were vital factors in the propagation of legislative referend a, while states ™ internal contexts were chief predictors in the spread of plebiscitary action. Meanwhile, there was less support for regional diffusion effects and interest group var iables. 84 Policy scholars have been alluding to, but not necessarily testing, this phenomenon for some time. Haider -Markel (2001b) indicated that fipolicy innovation and diffusion may be a part of a larger, strategic move toward gaining political 168 learning is the primary driver of venue diffusion. Learning about the successful venue other states have used to enact a policy should make subsequent states more likely to pick the same venue to pursue the policy (as this chapter will demonstrate) and more likely to adopt the policy in that venue (as the last chapter showed ). Continu ing with the policy case of gay marriage, this chapter offers both qualitative and quantitative evidence of venue diffusion. The qualitative evidence points to policy entrepreneurs and interest groups spreading successful strategies, including venue choice , to their respective camps across U.S. states. Although national -level organizations certainly played a prominent role in disseminating these tactical repertoires, treating the anti - and pro -gay rights movements as top -down monoliths is mistaken. Local an d state -level groups innovated and pushed the envelope even when national groups advised against it. Thus, the full epistemic communities for the religious right and gay rights movements facilitated the diffusion of venue shopping processes across states. The quantitative evidence offers further support of venue diffusion and political learning™s role. Opponents of gay marriage were much more likely to prohibit same -sex unions via the legislature and legislative referenda as other states successfully did s o. Meanwhile, proponents of marriage equality were more likely to press for change via state legislatures as prior advocates found success via that route. However, political learning™s impact on some venues fluctuated over time. Early success via citizen i nitiative, state courts, and federal litigation predicted other states following suit, but as success waned in those venues, policy actors either switched to alternative routes or waited for a final ruling from the U.S. Supreme Court. Empirical evidence fo r venue diffusion holds advantage by expanding the scope of the conflict to other institutional venues that may be predisposed to more favorable policy decisions.fl And social movement scholars have envisaged a much broader view of diffusion than policy scholars for some time. Researchers i n this vein have acknowledged the inputs Šsuch as protest tactics, interpretative frames, web of actors and interest group networks Šas much as the outputs (i.e., social change) (see Givan et al. 2010; Tarrow 2005). Because policy scholars have been ficirclin g the wagonsfl about its possibility, I test the notion of venue diffusion in this chapter. 169 even after accounting for alternative external forces and states™ internal forces that also predict venue choice. Not only were policy actors more likely to learn about and follow the successful paths taken by early movers, but the y also were more likely to emulate the way taken by institutionally - and politically -similar states. Policy actors prioritized information from peer states over information from all sources. Past research largely treats venue shopping as policy actors™ aut onomous evaluation of the best route to press for or impede policy change based upon a state™s institutional setting and political context. Instead, I show that policy actors™ venue choice is a product of both internal and external considerations. Ultimate ly, the fusion of the policy diffusion and venue shopping literatures improves our understanding of the factors driving patterns of venue choice across states. Venue Diffusion and Political Learning Decades of research have shown that venue shopping is a fundamental part of the agenda -setting process and in achieving policy change. Selecting the most favorable venue helps policy pioneers advance their goals (Baumgartner and Jones 1993; Kingdon 1984; Sabatier and Jenkins -Smith 1993) as well as foster new i nstitutions and constituencies that can further entrench the policy in the status quo (Karch 2009; Lubell 2013; Maltzman and Shipan 2008; Pralle 2003). Gone are the days that policy actors and interest groups solely focus their lobbying efforts in the hall s of Congress or state legislatures. Instead , policy actors and organized interests are capitalizing on the fragmented nature of U.S. federalism and pressing for change across horizontal and vertical venues. Venue shopping is on the rise (Miller 2009; Piot t 2003; Smith and Tolbert 2004, 2007), with actors turning to state legislatures, state courts, gubernatorial executive orders, ballot measures, state agency rules, and federal -level equivalents. Although competing in multiple venues impl ies multiple veto points, it also afford s multiple opportunities for policy change, especially if policy actors encounter failure in 170 other arenas (Lubell 2013; Pralle 2003). Individuals and groups can reduce the risk of failure through diversification (Boehmk e, Gailmard, and Patty 2013; Constantelos 2010; Jourdain, Hug, and Varone 2017 ). The venue shopping literature has primarily focused on internal factors and contexts driving venue choice. Recall from Chapter 4 that p olicy actors and organized interests ma y opt for a venue given their resources, capacities, prior experience, past success, and competitive advantage (Ley and Weber 2015; Lubell 2013; Lubell et al. 2010; Pralle 2003; Sabatier and Jenkins -Smith 1993). Or these change agents may weigh their oppon ents ™ size and influence in a given venue (Holyoke et al. 2012; Ley and Weber 2015). Still, these actors may gauge the institutional or political hurdles of a given venue, or the ideological congruence between their cause and the citizens or officials in t hat venue (Baumgartner and Jones 1991, 1993; Holyoke et al. 2012; Ley and Weber 2015). These mechanisms driving venue choice, however, are mainly internal considerations, either organizational or intra -jurisdictional. While I do not doubt the power of thes e internal considerations, I believe policy actors also weigh external information when picking a venue. As Chapter 4 laid out, I believe that policy actors™ venue choice in one state influences like -minded policy actors™ venue choice in other states. T his theory of venue diffusion is a theory of interdependence. It is a story about policy entrepreneurs strategically selecting the most favorable channel to press for a new policy, and subsequently, policy actors considering these prior choices in picking their venue for policy activity. Given policy actors™ limited time, attention, and resources, and facing too little or too much information, they satisfice and look elsewhere for solutions to common problems. This search results in both policy and politica l learning. Policy actors actively learn from one another about the policy innovation (i.e., policy learning) and the best avenue to achieve that innovation (i.e., political learning). That is not to suggest that policy actors blindly copy the routes taken by others before them. I concur with past research that policy actors consider their 171 capacities and resources, and their own states™ institutional and political contexts. But I argue that policy actors also weigh the successful avenues taken by other stat es. Policy actors™ venue choice is a product of both internal and external considerations. In the context of gay marriage, I charge that policy actors representing the religious right or gay rights movements learned about the tactical strategies and deci sions made by like -minded policy entrepreneurs and early movers in states that previously pursued the same innovation. As such, I theorize that as conservative activists and fundamentalist Christian organizations sought a ban on gay marriage via a particul ar venue, subsequent opponents of same -sex unions learned from these successful strategies and were more likely to pursue a ban in the same venue. Likewise, I contend that as more gay rights activists and organizations successfully engaged in a specific ve nue to achieve marriage equality, subsequent proponents were more likely to follow the same institutional path. Opponents of gay marriage largely focused their efforts in state legislatures and at the ballot box via legislative referenda and citizen initi atives. 85 Table 6.1 shows the main institutional venues that policy actors picked to pursue anti -gay marriage policies. Capitalizing on public support against gay rights, the religious right and conservative groups moved quickly to convince 40 state legislatures (most Republican -controlled, although some Democratically -controlled) to p ass statutory prohibitions against same -sex unions. These same groups then persuaded 19 state legislatures to put forward legislative referenda, encouraging state electorates to pass constitutional bans on gay marriage. 86 These groups obtained enough signat ures in 16 direct democracy states to allow voters a direct say via citizen initiative whether to constitutionally prohibit same -sex unions. Table 6.1 also 85 This strategy contradicts the resource mobilization theory of venue shopping, which suggests that organized interests representing the status -quo position Šanti -gay rights groups in this case Šshould have an institutional advantage at the federal level. Although adversaries of marriage equality were successful in getting Congress to pass a federal definition of fitraditional marriagefl in the 1996 Defense of Marriage Ac t (DOMA), conservative and religious right groups prioritized action at the state level. 86 Minnesota was the only state to not pass its legislative referenda in 2012. 172 documents the high success rate in each of these venues, with 88% of citizen initiatives, 90% of sta tutory bills, and 95% of legislative referenda successful. Notes: Table displays the venues selected to pursue anti -gay marriage policies and the success rate for those venues. Success rate is calculated by the number of states adopting via given venue divided by the number of states pursuing via given venue. Delaware is the one state without legislative referendum for constitutional amendments . Not all venues are represented here. Two states pursued limited same -sex marriage bans via gubernatorial executive orders (Alabama and Mississippi) followed by legislative action in those states . However, given the rarity of these events via those venues, they are not included here . See Appendix D for a full chronology of anti - and pro -gay marriage policies pursued in every state. Although most civil rights movements attempt to effect change through more central channels like Congress and the federal courts, the gay rights movement initially turned to state court s and later to state legislatures (Werum and Winders 2001). 87 Table 6.2 presents policy actors ™ and interest groups ™ venue shopping to pursue pro -gay marriage policies and their corresponding rates of success. Seventeen states pursued marriage equality via the state supreme court, with a 53% success rate. 88 Although almost a third of the states picked the state judiciary to protect minority rights, policy actors and organized interests in 16 states made their case to state legislatures. Nearly nine out of ten state capitols allowed civil unions or same -sex marriage. Despite much activity at the 87 Save for a dozen and half works in the diffusion and venue shopping literatures (Anders en 2005; Barclay 2010; Barclay and Fisher 2008; Caldeira 1985; Cann and Wilhelm 2011; Canon and Baum 1981; Dear and Jessen 2007; Dorf and Tarrow 2014; Fiorino 1976; Glick 1992; Hinkle 2015; Hinkle and Nelson 2016; Lewis 2011; Oakley 2009; Parent 2010; Pope lier 2015; Werum and Winders 2001), however, scholars have largely ignored the judiciary as a venue for policy change. According to Keck (2009) and Ley (2014), this is a mistake. Relying on interviews and case studies from environmental conflicts in the Pa cific Northwest, Ley (2014) argues that the fijudiciary can be an institutional venue that enhances public input, can be more inclusive than other venues, and produces positive -sum outcomes when other venues cannot.fl Facing a low probability of winning, LGB T organizations tried to avoid the ballot box. When gay rights groups did engage that venue, it was largely to legally challenge the religious rights™ initiatives or counterpetition (Stone 2012). 88 Despite the perception that the judiciary is the most ardu ous and costliest venue to achieve change (Andersen 2005), Kane (2010) suggests that the gay rights movement was able to mobilize greater support for gay rights in other venues because they took the fight to the courts. Ke ck (2009) and Ley (2014) also make the case that for some issue areas, the judiciary may yield a higher rate of return compared to other arenas. Therefore, regardless of the specific success rate, action in the state courts may have had a positive spillover effect on policy activity in the other forums. Table 6.1: Venue Choice to Pursue Anti -Gay -Marriage Policies, 1993 Œ 2015 Institutional Venue No. of States with Venue No. of States Picking Venue No. of States Successful in Venue Success Rate via Venue State Legislature 50 40 36 90% Legislative Referendum 49 19 18 95% Citizen Initiative 24 16 14 88% 173 state -level, policy actors als o turned their attention to the federal courts later in the timeframe. While two states (i.e., California and Nebraska) did attempt to achieve marriage equality through the federal court system before the U.S. Supreme Court™s 2013 ruling in U.S. v. Windsor , most federal litigation trailed that decision. Some 30 states filed suits in federal court to advance gay marriage in their states, with two -thirds successful via that route. Notes: Table displays the venues selected to pursue pro -gay marriage policies and the success rate for those venues. Success rate is calculated by the number of states adopting via given venue divided by the number of states pursuing via given venue. Not all venu es are represented here. Two states successfully granted marriage equality via legislative referendum (Maryland) and ballot initiative (Maine), while another two permitted same -sex unions in a limited capacity via gubernatorial executive order (Rhode Islan d and Missouri). However, given the rarity of these events via those venues, they are not included here . See Appendix D for a full chronology of pro - and anti -gay marriage policies pursued in every state. Qualitative Evidence of Venue Diffusion in F ight for Gay Marriage Since much of the policy process remains a black box for policy scholars, finding evidence of political learning and venue diffusion could prove challenging for most issue areas. Fortunately, the struggle for gay marriage is well docu mented. Several detailed narratives exist about the principal actors and key decisions made at critical junctures during the fight for and against same -sex unions (Andersen 2005; Cole 2016; Conger 2009; Fetner 2008; Hirshman 2012 ; Pierceson 2013; Smith 2008; Solomon 2014; Stone 2012). Qualitative evidence abounds that policy actors learned from the tactics and innovations used in other states. This section details how pro -gay marriage advocates relied on the strategic venue shopping decisions made by policy entrepreneurs and early movers in other states when deciding on their arena to press for change. Table 6.2: Venue Choice to Pursue Pro -Gay-Marriage Policies, 1993 - 2015 Institutional Venue No. of States with Venue No. of States Picking Venue No. of States Successful in Venue Success Rate via Venue State Legislature 50 16 14 88% State High Court 50 17 9 53% Federal Court 50 30 20 67% 174 In the fight for marriage equality, Evan Wolfson was a policy entrepreneur. 89 At Harvard Law School, Wolfson wrote a pioneering paper, entitled fiSamesex Marriage and Morality: The Human Rights Vision of the Constitution,fl promoting same -sex unions. Later hired by Lambda Legal Defense Fund (Lambda Legal) in 1989, he pressed the organization to defend equal marriage rights for LGBT couples. Wolfson wanted L ambda Legal to take up Ninia Baehr, Genora Dancel, and the two other couples™ legal challenge in Hawaii in 1991. But many in the gay rights movement balked at spending resources on fighting for same -sex marriage, seeing the efforts as premature, futile, or assimilationist. Wolfson™s boss said no, although he later allowed Wolfson to work with counsel, Dan Foley, behind the scenes on the Baehr v. Lewin case 90 (Cole 2016). Contrary to the commonly advised strategy of raising as many legal arguments as possible, Wolfson recommended limiting the disputes to state claims only. He feared making federal constitutional arguments would result in a jurisdictional change to the feder al courts, where they perceived no chance of winning (Cole 2016). The tactic paid off. After the Hawaiian Supreme Court remanded the Baehr case back to the trial court, the judge ruled Hawaii had no compelling interest to deny LGBT couples the right to mar ry. The judge required the state to recognize same -sex marriage but stayed his decision pending appeal. Facing popular pressure, Hawaiian legislators passed a statute restricting marriage to opposite -sex couples. Legislators also put a legislative referend um before Hawaiian voters in 1998, authorizing the legislature to limit marital unions to one man and one woman and enshrining the policy in the state™s constitution. The referendum passed by a wide margin. When the Baehr case returned to the Hawaiian Supr eme Court in 1999, the court dismissed it because the issue was moot given the newly adopted constitutional amendment. 89 Many LGBT activists refer to Evan Wolfson as the fifatherfl or fiPaul Reverefl of the same -sex marriage movement (Cole 2016; Gallagher and Bull 2011). 90 Baehr v. Lewin , 74 Haw. 530, 852 P.2d 44 (1993) originally, although renamed Baehr v. Miike in 1996 becau se Lawrence H. Miike tool over as the new State Director of Health for Hawaii. 175 Regardless of the swift countermovement, gay rights advocates in other states followed Wolfson™s lead to press for change via state c ourts and to limit the arguments to state claims (Cole 2016). This precedent is why nearly all the pro -same -sex marriage suits filed before 2009 narrowed their claims to state issues (Cole 2016). Learning from Wolfson™s strategy in Hawaii, Jay Brause and Gene Dugan applied for and were denied a marriage license in Alaska in 1994. They sued, relying merely on state arguments. 91 The Alaskan court ruled in the couple™s favor, stating that the ban on same -sex marriage constituted sex discrimination (Pierceson 2013; Smith 2008). In response, Alaska™s legislature followed its Hawaiian counterpart, passing statutory language and asking voters to approve a constitutional amendment banning gay marriage. Similarly following Wolfson™s direction and learning from the sw ift popular and legislative backlash in Hawaii and Alaska, Susan Murray and Beth Robinson, two attorneys representing same -sex couples in Vermont, decided to build p ublic and political support in the state first before filing a lawsuit for marriage equalit y. Murray and Robinson turned to Vermont Coalition for Lesbian and Gay Rights (VCLGR) and Mary Bonauto from the National Gay and Lesbian Alliance Against Defamation (GLAAD) for help (Cole 2016; Solomon 2014). Together, they produced video testimonials to p ersuade the public why equal marriage rights were so important. They also lobbied Vermont legislators, asking them to let a lawsuit run its course before taking any legislative action and oppose any constitutional ban via legislative referendum if the lega l case succeeded. Legislative allies were delighted only to have to play defense (Cole 2016). Only once the group gained enough commitments to be able to defeat a constitutional amendment in the legislature did they file the suit. Like the Baehr and Braus e cases before it, the Baker v. Vermont case raised state claims only, including Vermont™s ficommon benefits clause,fl which mandated that state benefits must be made 91 Brause v. Bureau of Vital Statistics , 1998 WL 88743 176 available to all residents (Cole 2016; Pierceson 2013; Smith 2008). 92 The case won at the Ve rmont Supreme Court in 1999. The Court ruled that the state had not justified such discrimination on the basis of sexuality and ordered the legislature to come up with a fix. Although short of full marriage rights in purpose, in 2000, Vermont™s legislature approved civil unions for same -sex couples that gave them the same legal rights and obligations as marriage in practice. 93 Vermont became the first U.S. state to permit same -sex unions. Building off the success in Vermont, Mary Bonauto turned her attentio n to GLAAD™s home state ŠMassachusetts. Racing a countermobilization by the religious right to press for a constitutional amendment in The Bay State, Bonauto did the necessary fipolitical ground -workfl before filing Goodridge et al. v. Department of Public He alth on behalf of Julie and Hillary Goodridge and six other same -sex couples (Cole 2016; Solomon 2014). Bonauto teamed up with Evan Wolfson, who left Lambda Legal to found Freedom to Mar ry, to argue Goodridge et al. before the Massachusetts Supreme Judicia l Court (Solomon 2014). They won. Not only did Massachusetts™ high court rule in the couples™ favor, but they also decided that the fiseparate -but -equalfl civil -union compromise was inadequate. Massachusetts became the first U.S. state to issue marriage lice nses to same -sex couples. Still, not all gay rights strategies were as disciplined or coordinated. Following President George W. Bush™s 2004 State of the Union call for a U.S. constitutional amendment fito protect the institution of marriage,fl San Francisco mayor Gavin Newsom began issuing marriage licenses to same -sex couples. He did so irrespective of prominent LGBT groups™ concerns that it woul d spark a backlash (Hirshman 2012). The California Supreme Court ordered the city to halt issuing licenses 92 Baker v. Vermont, 744 A.2d 864 (Vt. 1999) 93 For coding purposes, I treat civil unions as synonymous with same -sex marria ge since the adoption of civil unions were innovative in that they guaranteed the right of gay persons to access state services and benefits available to married persons . That said, I recognize the controversial distinction and gap in rights between these policy prescriptions. 177 until a case could make its way through the court system. However, California™s high court also seemed to encourage the mayor to file a separate action questioning the constitutionality of the current marriage statutes (Cole 2016). Again, without consulting key gay rights leaders, the mayor accepted the court™s invitation and immediately filed a constitutional challenge. Later in 2004, the California Supreme Court ruled that San Francisco™s same -sex marriages were invalid. The mayor did n ot have the authority to usurp state law simply because he believed current marriage statutes were unconstitutional (Cole 2016; Hirshman 2012). Four years later, however, California™s high court ruled on the constitutional challenge it had invited from the city, deciding that the right to marry was fundamental regardless of sexual orientation. Nonetheless, the one step forward for gay rights in California resulted in two steps back. Following the 2008 ruling, the religious right pushed and narrowly won Prop osition 8, a citizen initiative that enshrined the traditional definition of marriage into California™s constitution (Cole 2016). Seeing the footsteps trod before them, gay -marriage proponents filed and appealed suits to a dozen more state high courts. 94 Results were mixed with only half of those state judiciaries siding on behalf of marriage equality. Importantly, regardless of success or failure, policy actors in other states took advantage of the lessons learned by Evan Wolfson in Hawaii and early movers in Alaska, Vermont, Massachusetts, and California (Cole 2016). Although subsequent actors likely considered their state contexts, their decision to pursue same -sex marriage via their state courts or alternat ive venues was not an isolated, independent choice. Instead, it was influenced by the venue shopping processes already done in pr evious states. This interdependence, however, is not unique to the pro -gay marriage effort that occurred via state courts. Similar narratives are t old about gay rights activists learning from policy 94 Connecticut, Georgia, Iowa, Kansas, Louisiana, Maryland, Montana, New Jersey, New Mexico, New York, Oregon, and Washington 178 entrepreneurs pressing for change in state legislatures (Cole 2016; Solomon 2014). Religious right organizations also learned from the policy actors™ paths pursued before them, whether by way of state cap itols, legislative referenda, or citizen initiatives (Conger 2009; Fetner 2008; Haider -Markel 2000). Notably , the religious rights™ political strategies and venue choices affected the gay rights movements™ tactic al and venue decisions (Conger and Djupe 201 6; Fetner 2008; Pierceson 2013; Smith 2008; Stone 2012; Meyer and Staggenborg 1996 ). In fact, t he right™s countermobilization spurred the growth, institutionalization, and capacity building of local, state, and national LGBT groups ( Conger and Djupe 2016; Fetner 2008; Smith 2008; Stone 2012). Role of State and National Interest Groups As both the gay rights™ and religious right™s grassroots movements grew, stronger national organizations emerged. These national groups helped share the campaign successes a nd failures with other local and state groups. fiNational organizations developed training programs for activists, sent staff members to work on local campaigns, and provided an institutional memory of past campaign tacticsfl (Stone 2012: xxiii). National gr oups institutionalized the social movements for the LGBT community and the religious right. They facilitated political learning across the respective networks and epistemic communities (Cole 2016; Fetner 2008; Solomon 2014; Stone 2012). For example, follo wing the successful defeat of Oregon™s Ballot Measure 9 (the one requiring the firing of LGBT teachers and banning of fihomosexualfl books) in 1992, the tactics used became the model campaign to rout similar anti -gay policies in other states (Stone 2012). Na tional gay rights leaders shared the successful strategies Šincluding issue framing and messaging, 95 95 LGBT organizations shared strategies on how to reverse the negative perception of gays and lesbians. Gay ri ghts groups™ early frames painted the fight for same -sex marriage as one for equal civil rights. Although this message gained traction with some individuals, it failed to seriously sway mass opinion. Later frames, however, emphasized liberty, highlighting how the status quo denied same -sex couples™ commitment and love for one another. Love and devotion were emotive appeals that resonated with most Americans. Much like venue shopping is an essential aspect of the agenda -setting process, policy framing Šhow is sues are portrayed by the policy entrepreneurs and actors involved Šis 179 fundraising, door -to -door canvassing, obtaining endorsements from political and religious allies, and coordinating volunteers Šwith their state and local netw orks (Stone 2012). National LGBT groups also disseminated the legal tactics used to disqualify citizen initiatives before they got on the ballot in other states (Cole 2016; Stone 2012). In the same manner, early movers in the anti -gay movement shared their political strategies, messaging, and ballot measure language with like -minded groups in other states. Indeed, between 1974 and 2008, the religious right attempted more than 245 popular referendums and ballot measures at the local and state levels to curta il gay rights (Fetner 2008). Both movements learned from their own and each other™s victories and defeats. While we should not minimize the role of national -level interest groups in explaining this process, nor should we overstate it. The fact that s ome national -level interest groups help ed disseminate new policy ideas and help ed coordinate the choice of venue does not imply state and local interests were sidelined in the diffusion process. LGBT and religious right groups and individuals were not mono lithic in their missions or approaches. For example, national gay rights organizations advised against the lawsuits in Hawaii and Alaska (Cole 2016). And national LGBT interests disagreed with the newly -elected San Francisco mayor™s decision to issue marri age licenses to same -sex couples (Cole 2016). Despite the progress, national groups feared these events would spark a backlash and lead to setbacks. Indeed, it was because of individual pioneers and local actors that the movement for marriage equality gain ed traction. We should be careful not to dismiss this process simply as fiinterest groupfl politics (Salokar 1997). Along these lines, Stone (2012: xxiii) finds that fi[m]ost tactical innovations occurred in local or statewide campaigns, and were then spread through connections between organizations and social also a key part of the early stages of the policy process. Just as the venue choice in one jurisdiction may influence the venue choices in other jurisdictions (i.e., venue diffusion), so too might the policy framing (and counter -framing) in one jurisdiction diffuse to other jurisdictions. This is yet another policy input ripe for transmission that should be explored further. Gilardi, Shipan, and Wueest (2019) offer an excellent start to t his vein of research. 180 networks between activists.fl Further countering the narrative that one or two national organizations are driving the policymaking on an issue in the states, Wolak et al. (2002) show that most interest gr oups hold unique state registrations rather than multi -state registrations. It appears that most state -level groups firemain strongly rooted within their statesfl even as some aspects of these groups have become more nationalized (Wolak et al. 2002: 551). In stead of national groups steering the diffusion process, diffusion helps scale u p the level of coordination across actors (Givan et al. 2010). Likewise, interest groups with parallel missions may pursue divergent courses of action. Engel (2007) highlights how two national LGBT organizations ŠNational Gay and Lesbian Task Force and Human Rights Campaign (HRC) Šfollowed different strategies in different institutional venues to press for expanded LGBT rights during the 1990s. The National Gay and Lesb ian Task Force relied more on grassroots and local efforts to advance gay rights, while the Human Rights Campaign prioritized federal venues to press for equality. Interestingly, Engel argues that these diverse paths resulted not from coordination between the two entities but rather from differences in organizational identity at their outset. Engel shows how the identity argument outperforms the classic resource or capacity explanations employed by scholars to account for the differences in organizational s trategies. Following the passage of several state -level constitutional amendments banning gay marriage, however, both groups augmented the number of venues where they actively helped press for policy change (Engel 2007). Thus, this is not a story of one o r two national groups blindly replicating an innovative approach that had previously been successful in one state in other states. In contrast, it is a story of numerous individuals and local, state, and national groups working together to share tactical repertories, including venue choice, to advance a common cause. The qualitative evidence shows that national -, state -, and local -level professional organizations were critical to the transfer of policy 181 ideas and campaign tactics around gay marriage (Cole 20 16; Fetner 2008; Solomon 2014; Stone 2012; Werum and Winders 2001). Venue Diffusion in Fight over Gay Marriage : Expectations The earlier narrative provides qualitative evidence of political learning and venue diffusion in the fight over gay marriage. Yet, do the data bear this out? I now turn to empirically testing whether policy actors consider the successful paths taken by oth er states when pressing for policy change in their states. Although this section offers several expectations that parallel the expectations from Chapter 5, the primary difference here is that I am laying out the prospects for venue diffusion rather than fo r policy diffusion . The focus now is picking a venue rather than adopting a policy in a given venue. The difference between fipicking a venuefl and fiadopting a policyfl may sound trivial, but the former tests venue diffusion while the latter tests policy diff usion. Following the theoretical contributions from Chapter 4, I propose political learning as the principal mechanism driving the patterns of venue choice across subnational units. Although political learning could involve drawing lessons about other stat es™ policy framing, policy winnowing, coalition building, I focus on a key aspect of the agenda -setting process: venue shopping. As policy actors receive signals about the venues that policy entrepreneurs and early movers used to achieve policy change succ essfully, they should be more likely to pick the same venue to upend the status quo in their state. This comports with prior venue shopping scholarship that suggests policy actors learn from their own triumphs and flops (Ley and Weber 2015; Pralle 2003). I f policy actors learn from their own actions, we should assume they learn from others™ successes and failures as well. Related to the policy case of interest, I theorize that as conservative activists and fundamentalist Christian organizations picked a pa rticular venue to pursue a ban on gay marriage, subsequent opponents of same -sex unions learned from these successful strategies and were more 182 likely to select the same venue to pursue a ban. Likewise, I contend that as more gay rights activists and groups successfully engaged in a specific venue to achieve marriage equality, subsequent proponents were more likely to follow the identical institutional path. In summary, my main hypothesis is: H1: Political Learning: The likelihood of a state picking a venu e to pursue a policy increases as the proportion of other states successfully pursuing the policy via the same venue increases . Paralleling this purposive search for information, policy actors may also assess their institutional or political similarity relative to the states that have gone a given route. Just like states look to peers for a policy solution (Shipan and Volden 2006, 2014; Volden 2006), states may also look to their peers when picking a venue to press for policy change. More recent experimental research indicates that policy actors do indeed seek out information and take cues from self -selected, likely homophilic, netw orks (Butler et al. 2015; Zelizer 2019). Returning to the Boston Marathon example presented in Chapter 1, whereby runners share information about the most favorable courses to qualify for the prestigious event , racers likely gauge the fastest courses overa ll as well as the fastest courses for runners like them (e.g., considering gender, age, experience). It makes sense then that states will evaluate how comparable they are to other states that have already gone down a particular path. If those states engage d in venue shopping with their institutional and political contexts in mind ( Baumgartner and Jones 1991, 1993; Holyoke et al. 2012; Ley and Weber 2015) , then subsequent states with analogous contexts should be more likely to opt for the same venue. The ab ove narrative even reinforces this point, as policy entrepreneurs and early activists strategically selected the next states (e.g., Vermont, Massachusetts) based upon comparable institutional and political settings ( Cole 2016; Solomon 2014) . In the context of gay marriage, potential dimensions of similarity include legislative professionalism, state supreme court professionalism, difficulty in amending state constitution, citizen ideology, supreme court ideology, 183 and district court ideology. An added bonus of accounting for these comparisons between units is that I can simultaneously track each state™s institutional and political hurdles relative to the hurdles in other states that have already taken action in a given venue . And venue accessibility is a know n determinant of venue choice (Baumgartner and Jones 1991, 1993; Ley and Weber 2015). Given this, I hypothesize that: H2: Institutional / Political Similarities: The likelihood of a state p icking a venue to pursue a policy increases as more institutionall y and politically similar states opting for the same venue increases. Nonetheless, there are other well -founded explanations for why policy actors choose a venue. A s they do from time to time for policy solutions, states may look to their contiguous geographic neighbors to aid in selecting the best avenue to press for an anti - or pro -gay marriage policy change (Berry and Berry 1990; Berry and Baybeck 2005; Cohen -Voge l and Ingle 2007). The tendency toward homophily may cause policy actors to look no further than their neighbors, emulating the paths taken in geographically proximate states to achieve policy change. Hence, I put forward the following hypothesis: H3: Geo graphic Neighbor: The likelihood of a state p icking a venue to pursue a policy increases as the proportion of contiguous neighboring states picking the same venue to pursue the policy increases . Akin to the policy learning mechanism in policy diffusion, activists and groups may also weigh the cumulative number of other states adopting relevant policies via a given venue when deciding on an appropriate avenue. Instead of prioritizing a venue™s succ ess rate, assessing similarities with other states taking a path, or following the lead of neighboring states, policy actors may learn about the number of states that have already successfully select ed a venue and consequently join the bandwagon. Therefore , I propose the following hypothesis: 184 H4: Policy Learning: The likelihood of a state p icking a venue to pursue a policy increases as the number of other states successfully picking that venue to pursue the policy increases . Policy actors™ choice of venu e, however, may not be immune to federal government involvement in the policy area (e.g., Allen, Pettus, Haider -Markel 2004 ; Karch 2009, 2012; Shipan and Volden 2006, 2008; Welch and Thompson 1980). As such, federal -level activity on the issue could lead s tates to pick one institutional arena over another to cement or circumvent national signals. Moreover, the domestic political environment, such as the timing of presidential elections, could also make one venue more accessible than others (Baumgartner and Jones 1993; Berry and Berry 1990, 1992; Ley and Weber 2015; Mintrom and Vergari 1998; Smith et al. 2006). Hence, I hypothesize the following: H5: Federal Intervention : The likelihood of a state p icking a venue to pursue a policy increases / decreases as the federal government intervenes in the issue area. H6: National Environment : The likelihood of a state p icking a venue to pursue a policy increases / decreases as the national environment on the issue area ebbs and flows. Still, the venue -shoppin g literature emphasizes how policy actors consider their opponents when selecting an arena to press for change ( Holyoke, Brown, and Henig 2012; Ley and Weber 2015). A movement™s success in one venue may force the countermovement to compete in a different forum . As such, opponents™ policy successes in a state may also explain venue choice. Given this, I suggest the following: H7: Opposition Policy Success : The likelihood of a state p icking a venue to pursue a policy increases / decreases as the number of o pposition policy successes increase. Beyond considering opponents™ policy successes, policy actors also weigh their resources and capacities, as well as their opponents™ resources and capacities ( Holyoke, Brown, and Henig 2012; Ley and Weber 2015; Pralle 2003; Sabatier and Jenkins -Smith 1993). Change actors try to balance their own political, legal, and technical strengths against their challengers™ strengths (Ley and Weber 185 2015). Policy actors try to determine their competitive advantage (Sabatier and Jenkins -Smith 1993; Jenkins -Smith et al. 2014). I expect a similar dynamic here, with the religious right and gay rights movements™ sizes (and thus strength of influence) affecting both sides™ venue shopping processes. Indeed, the qualitative evidence above pointed to these anticipatory countermobilizations (Dorf and Tarrow 2014; Stone 2016). For example, greater interest group size may make policy actors more likely to select the legislative arena to pursue change, since elected officials may be more r esponsive to a constituency base as its size grows. Or as opposition size grows, interest groups may be forced to take their fight to the courts to protect minority rights. Venue choice could ebb and flow as interest group and opposition group sizes change . As such, I hypothesize: H8a: Interest Group Strength : The likelihood of a state p icking a venue to pursue a policy increases / decreases as the size of the interest group increases. H8b: Opposition Interest Group Strength : The likelihood of a state p icking a venue to pursue a policy increases / decreases as the size of the opposition interest group increases. Finally, we also know from the venue shopping literature that policy actors prefer venues where they are already engaged (Holyoke et al. 2012). Change agents typically stick with the venue they know. Consequently, if policy actors have already pursued and successfully adopted a policy in a venue, they should be less likely to pursue a similar policy in a competing venue. I offer: H9: Prior Policy Success : The likelihood of a state p icking a venue to pursue a policy decreases as the state has already successfully pursued the policy in other venues. 186 Data and Methods Data To test the theory of venue diffusion, I rely on the same compiled dataset of U.S. states pursuing anti - and pro -gay marriage pol icies used in Chapter 5.96 Again, states enter the risk set in 1993 following the Hawaii Baehr case and exit on or before 2015 when the U.S. S upreme Court Obergefell decision settled the issue. Since states could pick among alternate venues to pursue a poli cy change, states are stacked in the dataset for by venue and year . Therefore, the unit of analysis is state -venue -year. 97 More specifically for anti -gay marriage policies, actors in states could select the legislature, legislative referendum, or citizen initiative to pursue a ban .98 State -venue -year observations take on a value of 0 until the state picks a venue to pursue a ban, when that state -venue -year takes on a value of 1. Because the pooled data include unordered repeated events, states remain in th e dataset if they are at risk of selecting alternate venues to outla w same -sex unions; if the state re -selects the same forum to pursue additional bans; or if the state was unsuccessful in their initial attempt and reattempts via the same path. Essentially , state -venue -year units stay in the dataset until they are successful in the arena of interest or if they repeat the same venue choice. 99 Similarly for pro -gay marriage policies, states were at risk of choosing the state legislature, state court, or federal courts to permit same -sex unions. In the pro -gay marriage dataset, state -venue -year units take on a value of 0 until the state picks a route , when the unit switches to a value 96 Appendix C provides the full chronology of the anti - and pro -gay marriage policies pursued by venue type in each U.S. state. 97 For anti - and pro -gay marriage policies alike, because all 50 states could pick among three venues in a 23 -year time span, the maximum number of observations is: 50 3 23 =3,450 . 98 Importantly, states that do not permit citizen initiatives are controlled for in the models via a direct democracy variab le since they cannot be at risk of pursuing plebiscitary action. 99 Success in this instance is defined as the policy being adopted and implemented. Statutory language restricting same -sex unions that passes the state legislature but is vetoed by the govern or, for example, would not be treated as success and would remain in the dataset for that venue until the policy was successfully enacted. 187 of 1. Despite picking a venue to press for gay marriage, however, the state may remain in the pooled dataset for the other competing venues, or if the state retries (if initially unsuccessful) in the same venue or selects the same forum to pursue additional pro -gay marriage poli cies. 100 For instance, Delaware, Hawaii, Illinois, New Hampshire, and Rhode Island remain in the dataset after choosing their state legislatures to allow civil unions because policy actors later returned to the state legislatur es in those states to pursue fu ll same -sex marriage rights. They also remain in the dataset for the other available venues (i.e., state and federal courts) since they remained at risk of taking action on gay marriage in those other forums. Readers may have a sense of déjà vu in that the dependent variable here (i.e., picking a venue) appears to be the same as the dependent variable in Chapter 5 (i.e., adopting a policy). But the dependent variable in this chapter is a state selecting a venue in a given year to press for policy ch ange instead of a state enacting a policy in a given year (as it is in Chapter 5) . To illustrate the difference , consider the Californian state legislature™s attempt to legalize gay marriage in 2005. California™s House and Senate passed legislation permitt ing same -sex unions, only to have the statute vetoed by Governor Schwarzenegger . For the models in this chapter, the dependent variable takes on a value of 1 for the legislative venue because policy actors selected the state legislature to press for policy change. However, in Chapter 5, the dependent variable took on a value of 0 for the legislative venue because the policy was not enacted by the state. 100 Success here implies successful enactment of civil unions or same -sex marriage. If a federal district court rules i n favor of marriage equality, but the circuit court stays the ruling, this is not treated as a success and the unit remains in the dataset for that venue until same -sex marriage is allowed and implemented. The one exception to this is the 1993 Baehr v. Lew in case in Hawaii because the partial success (the Hawaiian Supreme Court remanding the case back to the trial court rather than dismissing it outright) led to the flood of activity in this policy area. It is treated as a success until the subsequent rulin g in 1999 when the Hawaiian Supreme Court sided against marriage equality because of the constitutional amendment passed by the state electorate in 1998. 188 Variable Operationalization This section is devoted to my choices in operationalizing variables used to test the above arguments in the anti - and pro -gay marriage models. The central predictor, Political Learning , for both the anti - and pro -gay marriage analyses remains the same as in Chapter 5. As a reminder, I operationalize political learning as the cumulative success rate of the states picking a given venue and achieving their policy goals at time . Overall, I expect a positive relationship between political learning and venue choice for both models. However, a negative coefficient would not necessarily negate political learning™s role. Learning can occur from failure, too. And since some venues yielded more mixed success rates, especially in the pursuit to legalize same -sex uni ons, it is possible that political learning™s impact varies by venue and over time. Because many of the other explanatory variables are the same as the ones included in Chapter 5™s models, I turn your attention only to the five differences in the variables used her e to test the above hypotheses . First, because policy actors might prioritize venue shopping cues from institutionally and politically similar peers (H 2), I construct three new variables for the anti -gay marriage models: Similarity in Legislative Professionalism , Similarity in Citizen Ideology , Similarity in Difficulty in Amending Constitution . For the pro -gay marriage models, I include the similarity in legislative professionalism and similarity in citizen ideology variables, along with three add itional variables: Similarity in Supreme Court Professionalism , Similarity in Supreme Court Ideology , and Similarity in District Court Ideology . All six of these measures are constructed in the same way, where I calculate the Euclidean distance between a s tate™s position and the average position of the states that have already selected the venue of interest. Then, I multiply the value by -1 to reverse code it, so larger values point to greater similarity. The base component s for all of the variables, exce pt citizen ideology, are used and defined in Chapter 5. For the citizen ideology 189 variable, I use Berry et al. ™s (2010) measure which is an aggregate of fiCommon -Spacefl congressional ideology scores, where higher values indicate a more liberal electorate. These similarity measures do not simply evaluate how analogous a state is relative to other states along these dimensions. Rather these measures gauge how similar a state is to the other states that have already pursued the policy via a given venue of interest . Hence, t hese measures are different across state -years depending on the venue in question. 101 Essentially, these similarity variables help capture policy actors ™ determination of whether the policy would work in their own state if i t worked in other similar environments (Rose 1991; Shipan and Volden 2014) . I expect a positive relationship between each of these variables and the likelihood of a state picking a given venue, as states should emulate the paths already taken by their inst itutional and political peers. The second change to my measurement strategy relative to Chapter 5, is that both the policy learning (H 4) and geographic variables (H 3) are specific to the venue under consideration. Instead of capturing the cumulative num ber of states by year that adopted a gay marriage ban or legalized same -sex unions (regardless of venue), Policy Learning by Venue captures the cumulative number of states by year that picked a given venue to pursue the policy successfully . And instead of simply representing the proportion of neighbors that adopted the policy of interest (regardless of venue), Geographic Neighbor by Venue represents the proportion of geographically contiguous neighbors that picked a given forum to successfully pursue the po licy. I assume that increases in both variables will make policy actors more likely to choose the given arena .102 101 For example, after policy actors achieved civil unions by way of the Vermont Supreme Court, gay r ights activists looked for another state with similar judicial arrangements and that could be more receptive to a lawsuit. These actors identified Massachusetts, whose level of supreme court professionalization and citizen ideology were only 0.048 units (out of 1.0 possible units) and 4.69 units (out of 72 possible units) different , respectivel y, from Hawaii and Vermont. This points to the possibility that policy actors prioritize states with similar institutional and political environments. 102 For purposes of clarification , the following variables are venue specific variables, deviating acros s state -venue -year units depending on the forum under consideration : political learning, similarity in legislative professionalism, similarity in citizen ideology, similarity in difficulty in amending state constitution, geographic neighbor by venue, and p olicy learning by venue. 190 Third, although I did not include the Lawrence v. Texas Supreme Court Decision and Presidential Election Year variables in the pro -gay marriage models in Chapter 5 because the ir inclusion complicated model estimates and likelihood -ratio tests, I can control for th ose variables here. Fourth, I do not include the NYT salience measure, sodomy ban, LGBT hate crime law, racial/e thnic minority population, and college education variables in these models. While there was much theoretical support for these variables driving policy adoption, there is less theoretical support from the venue shopping literature to evince their role in v enue choice. As such, I omit them from the models. Lastly, for these pro -gay marriage models, I measure opposition success as Prior Anti -Gay Marriage Policy, which is the cumulative number of gay marriage bans adopted by each state by year. I elect a narro wer count by state (rather than cumulative number across states) because the opposition success was nearly ubiquitous and state specific . See Tables D.1 and D.2 in the Appendix for complete variable descriptions, summary statistics, and sources for the ant i- and pro -gay marriage models, respectively. 103 Methods As I described in detail in Chapter 5, traditional event history data is typically modeled using logistic regression, estimating the likelihood of an event (e.g., adopting policy vs. not adopting policy). Such an approach is unsatisfactory here because the coefficients would fail to explain why a given venue was selected at all or why one venue was picked over another. Given this, I opt for a modeling strategy that accounts for policy actors™ discre te (and sometimes repeated) choice among multiple, competing venues: multinomial logistic regression. 104 And , as I explained in Chapter 5, the 103 I should also note that if variables were missing an observation for a given year, I relied on linear interpolation to fill the missing value. That said, I made a point to use variables with observations for nearly all state -venu e-years. 104 Of course, I could re -estimate the model with a different baseline category, which would change the coefficients and interpretations since all results are relative to the baseline outcome. Although I only report the coefficients of picking the venues relative to picking no venue in the tables, I do recount additional comparisons of picking between venues where appropriate. An alternative modeling strategy that has sometimes been suggested for this type of data is the gap -time 191 models™ standard errors are clustered by state -year to reduce the potential for heteroskedasticity while a time cou nter variable is added to the models to account for temporal dependence. 105 Results for Anti -Gay Marriage Policies Table 6.3 exhibits the results for the repeated -events, competing risk multinomial logistic regression model of venue diffusion for anti -gay marriage policies. 106 Recall that the dependent variable is a state™s likelihood of picking an institutional arena Šstate legislature, legislative referendum, or citizen initiative Što pursue a ban on same -sex unions, relative to the baseline of not selecting any venue. Though not reflected in the table below, the overall probability of policy actors choosing the state legislature to pursue an anti -gay marriage policy in any given year is two percentage points. Likewise, the likelihood that actors will opt to pass a ban on gay marriage via legislative referendum or citizen initiative in any given year is 0.8 and 0.7 percentage points, respectively. On their face, these probabilities may seem low. However, stasis is the status quo; U.S. state institution s are intentionally designed to impede and slow policy change. Moreover, these values represent the likelihood that policy actors, on average, will select these venues among multiple available venues in any given year throughout the entire 23 -year time per iod. Considering that, the model for competing risks. Although attractive, the gap -time model assumes an ordered nature to the events, which is not the case here. Furthermore, the gap -time model assumes a proportional hazard across the discrete choices, much like the Cox-proportional hazards model. Th is is problematic as the risk for picking a venue may not be proportional across the venue choices or across time. Consequently, I opt for the multinomial logistic regression model. 105 Despite advice from Cameron and Miller (2015), I cluster units by state -year because dynamics within a state in a given year may affect policy actors™ choice between competing venues within that state -year. Cameron and Miller (2015) recommend against this because it usually clusters on one observation (which results in no clus tering at all), but in my dataset state -year clusters group on three observations (one for each venue). The variance within a state -year is the variance of interest because actors can pick between three discrete venues in a given year. Nevertheless, cluste ring only by state does not affect the key findings. See Appendix D (Table D.7 and Table D.8) for results for both anti - and pro -gay marriage policies clustered by state instead of state -year. 106 Overall, the model performs quite well. T he area under the ROC curve, a statistic indicating the accuracy of the model, is 0.975, while McFadden™s pseudo R2 value is 0.532. Checking the Independence of Irrelevant Alternatives (IIA) assumption, I carry out IIA and likelihood ratio tests, despite strong theoretical reasons to treat the venues as separate choices regardless of the results. The results suggest there are no violations of the IIA assumption and that none of the choices should be combined. 192 Table 6.3: Venue Diffusion of Anti -Gay Marriage Policies using Mult . Logistic Regression Explanatory Variables Legislature Leg. Referendum Citizen Initiative Political Learning [+] 2.038* (0.970) 46.327* (22.601) -3.170ƒ (1.882) Similarity in Legislative Professionalism [+] 0.364* (0.145) 2.306* (0.797) 0.267 (0.239) Similarity in Difficulty Amending Constitution [+] 0.669 (0.416) 1.156* (0.565) 4.347* (1.222) Similarity in Citizen Ideology [+] 0.064* (0.029) 0.102* (0.043) 0.033 (0.031) Geographic Neighbor by Venue [+] 0.239 (0.756) 1.071 (2.289) 2.102 (1.325) Policy Learn by Venue [+] 0.227* (0.041) 0.162 (0.122) -0.237 (0.153) Federal Government DOMA [+] -0.353 (1.133) 14.950* (2.783) 12.926* (1.266) Lawrence v. Texas Sup. Ct. Decision [+] 4.995* (1.276) 0.508 (1.095) -1.044 (1.276) Presidential Election Year [+] 0.739 (0.553) 0.832 (0.594) 2.226* (0.618) Pro -Gay Marriage Counter [+] 0.198* (0.075) 0.007 (0.098) -0.252 (0.153) Evangelical Population [+] 0.013 (0.026) 0.086 (0.059) -0.154* (0.051) LGBT Population [ -] 0.817 (0.698) 0.758 (0.634) -1.310 (0.876) Prior Anti -GM Policy [ -] -0.724 (0.579) 0.136 (.810) -1.650* (0.702) State Supreme Court Professionalism [+] 3.854* (1.863) 2.130 (3.185) 1.187 (4.133) State Supreme Court Ideology [+] 1.047* (0.407) 1.186 (0.799) -1.900* (0.942) Direct Democracy [ -/+] 1.202 (0.529) -1.262 (0.979) 19.415* (1.500) Public Support for Gay Marriage [ -] -0.064 (0.062) -0.018 (0.087) -0.110* (0.050) State Population (Ln) [ -] -0.213 (0.223) -0.403 (0.639) -0.298 (0.508) Constant -1.904 (2.750) -61.773* (22.972) -27.959* (3.606) N 2505 2 (57) : 4640.55 * AIC / aROC 557.04 / 0.975 Log Likelihood: -218.52 ƒ 5, two tailed. Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of picking a venue to pursue anti -gay marriage polic y. Dependent variable has four categories, baseline category is not picking a venue to pursu e an anti -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. risk of forum shopping is sizeable. 107 107 Based upon the model, some states™ observed values deviated from their predicted values. For example, Mississippi and Texas were expected to pursue an anti -gay marriage policy via some venue in 1996, but neither acted until 1997. Likewise, there was a high probability that Maine would adopt a constitutional ban via legislative referendum in 2004 but did not do so until via popular referendum in 2009. Oklahoma, and South Dakota were also expected to adopt a constitutional ban on gay marriage via citizen initi ative in 2004; Oklahoma did so in 2004 but via legislative referendum, 193 From the table, the results are clear: political learning affects states™ venue shopping (H 1). Because the coefficients are expressed as log -odds and thus arduous to interpret, I provide average marginal effects for key variables in Figure 6.1. As a remin der, for continuous variables, average marginal effects can be interpreted as the instantaneous rate of change in the dependent variable following a small (e.g., one unit) increase in the independent variable , holding the other predictors constant . A small increase in policy actors™ success via state legislatures augments subsequent states™ propensity to pick that venue over others by 2.6 percentage points. A similar increase in political learning for legislative referend a intensifies the effect of choosing that venue compared to others by 27.5 points. Since legislators referring a vote to the state electorate are probably less certain about the outcome, legislators may rely especially on the success rate in other states before moving forward. In contrast to the positive effect on selecting the legislature or legislative referendum to pursue policy change, political learning has a negative marginal effect of 2.3 percentage points on choosing a citizen initiative relative to other venues to press fo r a gay marriage ban (although the Citizen initiative was the arena with the lowest success rate out of the three venues. This reality appears to have made subsequent states less likely to press for anti -gay marriage policies via plebiscite. Regardless, political learning™s marginal effect on picking any of the three venues is larger than any of the other external and internal variables, ex cept instead, and South Dakota did not adopt its constitutional ban via plebiscite until 2006. Aside from inaction, other states had a low probability of acting via the legislature when they did. Hawaii, Idaho, and Utah™s early efforts via the legislature was unexpected, with a predicted probability under seven percentage points for each state. However, the religious right™s continued push via Massachusetts™ legislature in 2006 and 2007 (foll owing failed attempts in 2004 and 2005) was perhaps the least anticipated, with predicted probabilities at 0.03 and 0.01, respectively. Hawaii and Alaska, being the first to adopt constitutional bans via legislative referendum, also took the model by surpr ise. The predicted probabilities for both states lobbying for constitutional language outlawing same -sex unions in 1998 was under seven percent. Still, Wisconsin™s push in 2006, Arizona™s pursuit in 2008, and Minnesota™s attempt in 2012 via legislative ref erendum all deviated greatly from their predicted values of 0.02, 0.01, and 0.01, respectively. In similar fashion, several states pursue d gay marriage bans via citizen initiative much earlier or later than anticipated. Both California and Nevada had initi atives on the ballot in the early 2000s despite predicted values under eight percentage points of doing so. And Arizona (in 2006), Maine (in 2009), and Washington (in 2012) all followed the citizen initiative path when facing low probabilities of doing so, the highest being Arizona at only a ten percent chance of acting via plebiscite. 194 for Congress™ s passage of DOMA and the U.S. Supreme Court Lawr ence ruling. See Figure D.1 in the Appendix for predicted probability plots of picking each venue as political learning for the given venue increases. Comparing between venues, a state™s odds of picking the legislature over citizen initiative as political learning in the fipeople™s branchfl increases one standard deviation (42%) is a factor of 8.82. The same increase in political learning for legislative referenda swells a state™s risk of picking a referendum over citizen initiative by a factor of 9.6 x 10 8. Therefore, policy actors are much more likely to choose the legislature or legislative referenda over direct citizen action as those routes™ success rates rise. This is not terribly surprising since conservative and religious right interest groups had the institutional and public opinion advantage against the gay rights movement. Given this, Figure 6.1: Average Marginal Effects of Key Variables on Venue Diffusion for Anti -GM Model 195 Figure 6.2: Political Learning™s Effect on Venue Choice Over Time for Anti -GM Policies access via state legislatures and legislative referenda, relative to citizen initiatives, was likely more expedient, cheaper, and with less uncertainty. Nonetheless, it is possible that political learning™s influence flu ctuated over time. To further examine this, I plot in Figure 6.2 political learning™s effect on venue choice to pursue gay marriage bans at three different points in time: 1997, 2003, and 2009. 108 The plots represent the predicted 108 The repeated -events competing -risks multinomial logit model used to estimate and plot political learning™s effect over time includes an interaction term between the politic al learning and the time component variables. Although it may be tempting to calculate the average marginal effect for this interaction term, the value of the interaction term cannot change independently of constituent terms ™ values (Williams 2012). Instea d, after including the interaction term in the model, I plot political learning™s effect on picking either the legislature, legislative referendum, or citizen initiative re lative to picking no venue at three different points in time to pursue anti -gay marr iage policies. 196 probabilities of selecting the given venue in the specified year as political learning in all venues is taken into consideration. In particular, the Loess -smoothed lines for the three years trace the predicted probabilities of picking that venue in the given year over the range of p olitical learning values from all three venues. We see from the first plot displaying political learning™s effect on choosing the legislature that the impact was largest early in the 23 -year time period. For example, in 1997, as states™ success rate in pro hibiting same -sex marriage via the legislature increased, subsequent states™ likelihood of going that route increased to 30 percentage points. However, by 2003, with most states successful in any venue, subsequent states were less likely to pick the legisl ature. And by 2009 , a state™s propensity to pass statutory language against same -sex unions was near zero. This dynamic tracks with the qualitative narrative. Policy actors in California, Nebraska, Nevada , and Wisconsin all pursued action later in the cycl e via legislative referendum or citizen initiative without success in the legislature. As the fight over gay marriage evolved, opponents of marriage equality initially preferred the legislature, but then opted for more entrenched bans on same -sex unions regardless of whether action had been previously taken in the state legislature or not. In contrast to political learning™s early impact on picking the people™s branch, political learning™s effect on choosing legislative referenda or citizen initiative occurred later in the time period. Of course, political learning did not influence selectin g legislative referendum or plebiscite in 1997 because no state had gone that route. By 2003, with successful but limited action via legislative referenda in Hawaii and Alaska, political learning™s impact on picking legislative referenda was positive but m arginal. And by 2009, as other states™ success rate in venues was near perfect, states exhibited a likelihood of four percentage points to pick legislative referend a. Similarly, with multiple states™ early success via citizen initiative by 2003, other stat es™ risk of choosing a citizen -driven ballot 197 measure increased by two percentage points. Yet by 2009, political learning™s influence on picking plebiscitary action had decreased slightly .109 110 Beyond policy actors ™ learning about successful paths trod in oth er states, we also see from Table 6.3 and Figure 6.1 that policy actors look to emulate the successful routes taken by their institutional and political peers (H 2). A one -standard -deviation increase in similarity of legislative professionalism relative to those states that previously picked the legislature raises a state ™s risk of taking the same path by 0.7 percentage points in a given year. An equal shift in the similarity of legislative professionalism for legislative referenda, however, boosts a state ™s likelihood of going that route by 4.7 points in a given year. Meanwhile, states also follow others with similar hurdles in amending their constitutions. A one -standard -deviation increase in similarity in difficulty of amending a constitution increases a s tate ™s risk of picking legislative referendum by 0.6 points, and citizen initiative by four points in a given year. In the same manner, policy actors track states with similar electorates. An 18 -point shift in a state ™s citizen ideology (in the conservativ e direction) 109 This result may be because Arizona™s electorate rejected a constitutional ban on gay marriage via citizen initiative in 2006, but later approved it via legislative referendum in 2008. Religious right groups™ initial defeat in Arizona may have made subsequent states less likely to consider citizen initiative to circumscribe marriage equality. Indeed, Washington voters also rejected a ban via plebiscite in 2012. 110 Another way to assess political learning ™s effect on policy actors ™ choice of venue to pursue policy change is to plot political learning and time ™s interactive influence on forum selection. That is, how does the successful venue shopping in one arena by some states over time affect the propensity of subsequent states to pick th e same venue? Figure D.2 in the Appendix displays political learning and time ™s joint effect on policy actors ™ venue -shopping decision making to press for bans on gay marriage . Although related to Figure 6.2 which reveals political learning ™s effect on ven ue choice at three distinct points in time , Figure D.2 emphasizes political learning and time™s mutual influence on the dependent variable. We see from the plot that states had the highest propensity of picking the legislative arena to prohibit same -sex un ions early on. But as political learning and time increased, policy actors were more likely to choose citizen initiatives and legislative referenda to outlaw gay marriage. This is understandable as most states first pursued statutory language and then cons titutional bans to deny marriage equality. And once public attitudes shifted and federal courts started to rule on the issue, nearly all efforts to ban gay marriage via any venue ceased. Interestingly, although legislative referenda were slightly more successful than citizen initiatives in achieving anti -gay marriage policies, policy actors ™ probability of picking citizen initiative was marginally higher than selecting legislative referenda. This discrepancy may be because the religious right had greate r leeway to pursue a ban via citizen initiative if they could not convince legislators to put forward a legislative referendum, especially as public opinion shifted on the issue later in the cycle. While Figure 6. 2 shows that political learning ™s influence on venue shopping was variable at different points in time, Figure D.2 reveals that policy actors were more likely to pick the legislature to pass anti -gay marriage policies early on, but more likely to choose citizen initiatives or legislative referenda to achieve policy change as political learning and time jointly increased. Figure D.2 further reinforces the finding that external information from other states ™ successful venue shopping positively impacted subsequent states choice of forum over time . 198 relative to early movers ™ ideology, expands the state ™s propensity to pursue a ban via the legislature or referenda by two percentage points. Many of my expectations for the other factors influencing patterns of venue choice across U.S. states also hold. Although policy actors did not appear to copy their geographic neighbors ™ venue choices (H 3), policy actors did pay some attention to the cumul ative number of states opting to pick the legislature to successfully pass the policy (H 4). 111 Federal government activity on the issue area also had some influence on states ™ venue shopping strategy (H 5). Congress ™s passage of DOMA in 1996 had an average ma rginal effect of 8.7 points on states selecting legislative referenda, and 6.1 points on states opting for citizen initiative. Congress ™s acquiescence on the issue may have encouraged states to adopt even stricter prohibitions against gay marriage. Likewis e, the U.S. Supreme Court ™s 2003 Lawrence ruling declaring state sodomy bans unconstitutional resulted in a 7 -point uptick in the probability that policy actors would turn to state legislatures to pursue initial or repeated policy action on the issue . Als o lending some support to the national environment hypothesis (H 6), policy actors were , on average, one percentage point more likely to pursue citizen initiatives during presidential election years. But the gay rights movement™s policy successes had a limi ted effect on opponents™ venue choice (H 7), while the religious right™s prior policy activity (H 9) only made subsequent activity via citizen initiative less likely. 112 Despite considerable qualitative evidence for interest groups™ role in venue shopping ( H8a, H8b), neither the interest group (i.e., percentage of state population that is Christian Evangelical or member of Church of La tt er -day Saints) nor opposition interest group (i.e., percentage of state 111 An increase of 11 additional states picking the state legislature raised a state ™s probability of choosing the same venue by 5.3 percentage points in a given year. 112 For each additional state that allowed gay marriage, a state™s marginal likelihood of choos ing the legislature increased by 0.3 points. Adopting one ban via another venue decreased a state™s chances of pursuing an additional ban by way of citizen initiative by 0.5 points. 199 population that identifies as LGBT) variables had sta tistically reliable marginal effects. In fact, the only interest group variable significant in Table 6.3 above is a state™s Evangelical population™s effect on picking the citizen initiative route. An 11 percent increase in a state™s more conservative Chris tian population reduces a state™s probability of pursuing a ban via plebiscite by 0.5 points in a given year. This is because greater Evangelical populations raise the odds of picking state legislature and legislative referenda, relative to selecting citiz en initiative, by seven - and 17 -fold, respectively. States with larger Evangelical populations provide policy actors a political and institutional competitive advantage via the people™s branch, thus reducing the need to press for change via citizen initiat ive. Still, a perceptive reader may rightly wonder why the state interest group and opposition group strength variables are not more relevant in the current model. The fact that those variables are not statistically significant does not imply interest gro ups played no role in the venue shopping to pursue bans on gay marriage. 113 Rather, interest groups may exhibit null findings here because their actual influence is being captured via the political learning, institutional and political similarity variables (Lowery 2013). Interest groups are the ones engaging in these purposive se arches for information; that is their degree of influence in these dynamic venue shopping processes (Lowery 2013). Furthermore, national -level pressure groups may still play a role in this process. I do not account for national interest groups in the model because no one organization or measure can account for the heterogeneity of these movements. 114 113 And using numerous alternate surrogates for religious right and gay rights interest group strength (Button et al. 1997; CenterLink (2016); Conger and Djupe 2016; Equality Federation Institute and Movement Advancement Projects; Family Research Council; Tay lor et al. 2019) does not change the interest group variables™ effect on patterns of venue shopping across states. 114 The remaining control variables had mixed effects across the venues , although most comported with my expectations . A more professionalized state supreme court increased policy actors ™ marginal effect on picking the state legislature by 5.4 percentage points. Barclay and Fisher (2008) and Hume ™s (2011) research point to lawmakers trying to signal or circumvent a more professionalized judiciar y. Likewise, a more conservative state court of last resort increases the marginal effect of choosing the legislature by 1.5 points and decreases the marginal chances of picking citizen initiative by one percentage point. With a right -leaning court, change agents feel confident in pursuing bans on gay marriage via legislative venues. Finally, greater public opinion in favor of gay marriage made picking all the venues less likely, although only the citizen initiative coefficient was statistically significant at the 200 Taken as a whole, the political learning and similarity variables offer strong empirical evidence that policy actors seek out and consider external venue shoppi ng information . Policy actors learn which routes are successful (H 1) and they prioritize the venue shopping cues from institutionally and politically similar states (H 2). Given this, a state ™s choice of venue does depend, in part, on prior states ™ choice o f venue. At least in the pursuit of anti -gay marriage policies, venue diffusion does occur . This central finding holds even after controlling for states ™ institutional arrangements, internal political contexts, interest group strength, prior venue shopping , and other external factors. Policy actors do not only look inward but also look outward to assist in picking the most favorable avenue to press for policy change. Results for Pro -Gay Marriage Policies The anti -gay marriage model offered strong empirical support for the theory of venue diffusion. Opponents of gay marriage sought out and considered successful venue shopping strategies by actors in other states, especially similarly situated states. But did proponents of gay marriage behave in the same way? Table 6.4 exhibits the results for the repeated -events, competing risk multinomial logistic regression model of venue diffusion for pro -gay marriage policies. 115 The dependent variable here is a state™s likelihood of picking an institutional arena Šstate leg islature, state court, or federal court Što legalize same -sex unions, relative to picking no venue. Generally speaking, a state™s propensity to pick any venue to enact gay marriage is reduced, relative to their likelihood of selecting a forum to pursue a ba n on same -sex unions. Policy actors face a probability in support of same -sex unions decreased advocates™ chances of going the way of citizen initiatives by 0.4 points in a given year. 115 Looking to the model fit statistics, the model performs quite well. McFadden™s p seudo R2 value, a quasi -parallel to the amount of variance explained, is 0.523, while the area under the ROC curve, an indicator of model accuracy, is 0.958. As before, I test the Independence of Irrelevant Alternatives (IIA) assumption, despite theoretical reasons to treat and model each venue separately. The results point to independent alternatives, thus complying with the assumption for multinomial logistic models. Moreover, the likelihood ratio tests indicate that none of the choices should be combined. 201 Table 6.4: Venue Diffusion of Pro -Gay Marriage Policies using Mult . Logistic Regression Explanatory Variables Legislature State Court Federal Court Political Learning [+] 15.157* (4.540) -1.987* (0.679) -4.430* (1.888) Similarity in Legislative Professionalism [+] 0.045 (0.409) -0.274 (0.172) 0.019 (0.235) Similarity in Supreme Court Professionalism [+] -3.908 (5.051) 4.487* (1.812) 0.519 (3.395) Similarity in Citizen Ideology [+] 0.160* (0.048) 0.078* (0.030) 0.066* (0.020) Similarity in Supreme Court Ideology [+] 2.831ƒ (1.710) 1.330 (0.954) 2.045* (1.026) Similarity in District Court Ideology [+] 0.337 (1.152) 0.303 (0.459) 0.558 (0.643) Geographic Neighbor by Venue [+] -2.744ƒ (1.594) -2.187 (1.591) 1.220 (0.873) Policy Learn by Venue [+] 0.324* (0.161) 0.125ƒ (0.067) 0.369* (0.137) Lawrence v. Texas Sup. Ct. Decision [+] 27.665* (4.674) 1.318 (1.189) 16.696* (3.066) U.S. v. Windsor Sup. Ct. Decision [+] 1.535 (0.980) 1.972 (1.203) 4.473* (1.796) Presidential Election Year [ -] -2.504ƒ (1.351) 0.299 (0.574) 0.918 (0.977) Anti -Gay Marriage by State [+] -0.313 (0.547) 0.040 (0.477) 0.858* (0.332) Evangelical Population [ -] -0.071 (0.064) -0.040 (0.036) 0.037 (0.051) LGBT Population [+] 1.601ƒ (0.822) 0.616 (0.521) -0.142 (0.515) Prior Pro -GM Policy [ -] -1.800ƒ (1.056) -1.446* (0.731) -1.493* (0.469) Public Support for Gay Marriage [+] 0.108* (0.046) 0.029 (0.050) 0.102ƒ (0.054) State Population (Ln) [+] -0.007 (0.396) 0.026 (0.220) 0.027 (0.301) Constant -32.771* (5.894) -1.275 (2.404) -19.406* (7.127) N 3322 2 (54) : 8700.44 * AIC / aROC 519.84 / 0.958 Log Likelihood: -202.92 ƒ 5, two tailed. Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of picking a venue to pursue pro -gay marriage polic y. Dependent variable has four categories, baseline category is not picking a venue to pursue a pro -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. of choosing the state legislature at 0.7 points, th e state court at 0.6 points, and the federal court at 0.9 points in a given year. Again, these are the probabilities for any given state across the entire 23 - 202 year time period to pick a venue. Since the gay rights community moved at a slower pace than the religious right, the early years in the cycle saw much less activity lobbying for marriage equality. 116 The upshot from Table 6.4 is political learning ™s persistent effect on states ™ venue choice to pursue pro -gay marriage policies (H 1). Figure 6.3 displays the average marginal effects for the key variables in the model. We see right away that political learning™s marginal effect across most of the venues is substantively larger than all other key variables except for the U.S. Supreme Court™s Lawrence decisio n. A small increase in political learning in the legislature produces a 5.9 percentage point increase in subsequent policy actors™ probability of also going the route of the state legislature in a given year. Political learning appears to have the opposite effect on selecting the state and federal courts relative to the other available venues . A similar increase in political learning in the state and federal courts actually decreases the likelihood that other states will follow suit by 1.3 and 2.7 points , respectively . This latter result somewhat contrasts with my expectations of political learning™s positive effect. W hile opponents of gay marriage were quite successful via multiple routes, proponents™ success rate across venues was more mixed. Rec all from Table 6.2 that change agents were 88% successful via state legislatures, but only 53% successful before state supreme courts and 67% successful in federal courts. As policy actors became increasingly successful in state legislature, 116 Despite the excellent model fit, however, a few states™ observed values for pro -gay marriage policies deviated from their predicted values. The model an ticipated that Michigan, Ohio, and Georgia would pursue lawsuits via federal court in 2015. Of course, what the model did not know is that, indeed, litigants in Michigan and Ohio had appeals before the U.S. Supreme Court in 2015, following the negative rul ings by the U.S. 6th Circuit Court of Appeals in 2014. The predicted values also anticipated actions in Hawaii and Rhode Island via their state legislatures sooner than occurred. At the same time, California and Connecticut™s push for marriage equality via their state legislatures in 2005 was also unexpected, with predicted probabilities less than three percent. Likewise, the probability of New Hampshire™s civil unions in 2007, Nevada™s domestic partnerships in 2009, and New Jersey™s attempted gay marriage legislation in 2013 were all under five percent. Still more surprising was some states™ pursuit of same -sex unions via their state courts of last resort. Proponents™ lawsuits in Georgia in 2006, Connecticut in 2008, Montana in 2012, Kansas in 2014, and Lou isiana in 2015 all had predicted probabilities under one percent. Early federal court action by gay marriage proponents in Nebraska in 2006, California in 2012, and Utah in 2013 was also unanticipated, with the model™s probability for such suits under 0.3 percentage points. 203 Figure 6.3: Average Marginal Effects of Key Variables on Venue Diffusion for Pro -GM Model subsequent policy actors were less likely to pursue litigation in state or federal courts. Thus, it is possible that political learning™s influence fluctuated over time, even more so than in the pursuit of anti -gay marriage policies. See Appendix D (Figure D.3) for predicted probability plots of picking each venue to pursue pro -gay marriage policies as political learning increases across the venues. Figure 6.4 plots political learning™s effect on states™ choice of venue to pursue pro -gay marriage policies at three points in time: 2000, 2007, and 2014. 117 Importantly, these are the predicted probabilities of selecting the given venue at a snapshot in time as political learning in all venues is taken into consideration. Although positive, political learning™s effect on selecting the legislature to pursue marriage equality is attenuated over time. With state legislatures not pressing for same -sex unions until 2005, political learning played no role in 2000. But by 2007, successful venue shopping by early movers augmented succeeding states™ likelihood of picking the people™s branch by 117 Again, these plots are produced from a repeated -events competing -risks multinomial logit model that includes an interaction term between the political learning and time component variables. 204 four percentage points a year. By 2014, political learning still had a positive effect on selecting the legislature, although much smaller at a little over one percentage point in an y year. This decline in political learning™s influence is likely due to the U.S. Supreme Court™s U.S. v. Windsor ™s 2013 ruling; proponents turned to the federal courts in droves following the High Court™s decision. Examining the State Court plot in Figure 6.4, political learning™s impact on policy actors choosing state litigation is more variable over time. Early success via the state courts in Hawaii and Vermont increased subsequent states™ risk of also fighting for marriage equality in the courtroom by one percentage point. This parallels the qualitative evidence in the chapter. But ensuing failures in Georgia, Maryland, New York, Oregon, and Washington reduced other states™ Figure 6.4: Political Learning™s Effect on Venue Choice Over Time for Pro -GM Pol icies 205 probability of taking the same path. For this reason, states™ predicted probability of picking the state courts in 2007 is relatively flat. Subsequent states likely faced too much uncertainty by way of the courts given the number of failures in this arena, and thus turned to other arenas (namely state legislatures). Still, by 2014, political learning™s effect on pressing for c hange via state courts of last resort was curvilinear in nature as states were more successful via other institutions. Gleaning insights from the Federal Court plot in Figure 6.4, political learning ™s effect on picking the federal courts is also nonm onotonic. As political learning increased across the three venues in 2014, states ™ initial probability of pressing for policy change via the federal judiciary increased to nearly 30 percentage points . But as other routes (namely state legislatures) became more successful, states ™ risk of choosing the federal courts decreased to ten percentage points. It is also possible that as more states tried cases at the federal level, laggard states waited to see how the U.S. Supreme Court would adjudicate among the lo wer courts ™ myriad and conflicting rulings. 118 Despite fluid Šsometimes even negative Šeffects at distinct points in time , political learning impacted advocates ™ and interest groups™ venue shopping. Heightened success in one venue made subsequent policy actors more likely to pick the same venue, while more mixed records led successive policy actors to consider alternate venues. Political learning involves learning from the failures as much as the successes. Both the gay rights and religious rights movements lea rned from the wins and the losses, transferring those lessons to groups in other states. Indeed, Evan Wolfson, a pioneer in the same -sex marriage movement, liked to fitalk about ‚losing forward™fl (Cole 2016: 72). 118 Just as I did for venue choice to pursue a nti -gay marriage policies, I also plot political learning and time™s interactive effect on venue choice to press for marriage equality. Figure D.4 in the Appendix displays political learning and time™s joint influence on policy actor™s choice of legislatur e, state court, or federal court to try to obtain same -sex unions. Early on, gay rights groups were more likely to go the route of state courts, although the probability of going this route was admittedly low. But as political learning and time increased, policy actors had a higher propensity to press for gay marriage in the federal courts and state legislatures, with going the route of state courts less likely over time. While Figure 6.4 emphasizes political learning ™s influence on forum shopping at three different points in time, Figure D.4 highlights political learning and time ™s mutual effect on venue choice. The plot further underscores how policy actors ™ successful venue shopping in competing arenas influenced the venue choice of subsequent actors in r eal time. 206 But proponents of gay marriage d o not appea r to treat venue shopping information equally from all sources. Just like opponents of gay marriage, proponents prioritize the venue choices of their institutionally and politically similar peers (H 2). In considering the state legislature, policy actors fa ctor in their state™s citizen ideology relative to other state electorates that have previously picked the people™s branch. An 18 -point increase in ideological similarity in the liberal direction makes states 1.9 percentage points more likely to also pick the state legislature. At the same time, states with more proximate state supreme courts on an ideological dimension are 0.5 points more likely to select the legislature in a given year. These results reinforce the fact that change agents are aware of the shared institutional powers in their state, so they look to parallel states™ past experiences navigating these competing institutions. Policy actors considering the state courts weigh their similarity in supreme court professionalism and citizen ideol ogy with early mover states. A o ne-standard -deviation shift in a state™s similarity in supreme court professionalism or citizen ideology relative to other states going that route increases a state™s chances of picking the state court by 1.1 and 1.5 points in a given year, respectively. In selecting the federal court, it is a state™s proximity in supreme court and citizen ideology (rather than district court ideology) with first mover states that makes it more likely to choose that venue. A one -standard -deviation move in both similarities with previous venue shoppers makes a state 0.4 and 0.7 points, respectively, more likely to use the federal judiciary to advocate for marriage equality. Many of the other variables also influence the pattern of venue cho ice in the anticipated direction. States do not pay much attention to the venues chosen by their geographically contiguous neighbors (H 3), except when considering the state legislature as a potential avenue. States appear 0.1 percentage points less likely to pursue change in the state capitol as 14% more of a state™s neighbors go that route. Perhaps as states watch their neighbors permit gay marriage via the legislature, they 207 wait to see if there is any electoral fallout. Related, policy actors are less lik ely to pick the state legislature during presidential election years (H 6), again likely out of concern that incumbents would face a backlash. Nonetheless, policy actors do pick venues based upon the cumulative number of other states that have gone that rou te (H 4). The average marginal effect of policy learning on venue choice, however, is under 0.2 percentage points across all venues. Two of the variables with the largest effect on venue choice are those capturing the federal government ™s influence (H 5). The U.S. Supreme Court ™s 2003 Lawrence decision produced an average marginal effect of 9.7 points on states picking the state legislature, and marginal influence of 8.2 points on states choosing the federal courts. Not surprisingly, the Supreme Court ™s 2013 U.S. v. Windsor ruling had an average marginal effect of 2.3 percentage points on policy actors turning to the federal courts to press for marriage equality. The opposition™s countermobilization also influenced proponents™ venue choice (H 7). For every add itional ban adopted by a state, proponents of gay marriage in the state were 0.6 percentage points more likely to overturn the bans via the federal courts. And as I predicted (and as demonstrated by the Prior Pro -GM Policy coefficient), states that had pre viously pursued a pro -gay marriage policy in another venue were less likely to pick a new venue (H 9). 119 Returning to the interest group strength variables (H 8a, H8b), as we witnessed in the anti -gay marriage models, a state™s size of Christian Evangelical population and LGBT population appears to have little effect on venue choice. Although most coefficients are in the anticipated direction, the only coefficient that reached statistical significance near conventional levels was the LGBT population™s influen ce on picking the legislature, with an average marginal change of 0.6 points. Again, the insignificance of these variables is somewhat surprising, especially given gay rights groups™ 119 Public opinion also influenced venue choice, with greater support leading policy actors to choose the state legislature early on, and later the federal courts. The connection between venue shopping and public opinion should be explored furth er in subsequent research. 208 role in the qualitative evidence of venue diffusion. Using dozens of diff erent measures for interest group strength and opposition interest group strength did not alter the results. As before, pressure group influence may be captured in the political learning or similarity variables since these groups were responsible for drawi ng lessons from early movers and peer states. Or national groups may have played a role uncaptured by these models. Importantly, Lowery (2013) reminds us that null findings for interest groups does not imply zero influence; rather their conception of influ ence remains disguised. Here, I believe their influence is reflected in the main drivers of the venue choice to press for pro -gay marriage policies in the U.S. states. In sum, the results here suggest that proponents of gay marriage rely on external info rmation in selecting a venue. Policy actors look to the successful routes taken in other states, while simultaneously prioritizing information about venues selected in institutionally and politically similar states. Opponents of gay marriage considered the ir state ™s legislative professionalism and institutional hurdles relative to early movers, while proponents of gay marriage factored in their citizen ideology, state supreme court ideology, and state high court professionalism relative to states already ta king a given route. As boundedly rational actors, these change agents do not simply consume all information, but rather filter the relevant information (Meseguer 2005). And as Stone (1999) rightly points out, filesson -drawing is not politically neutral .fl These findings hold even after controlling for alternative external and internal considerations. Robustness Checks As a robustness check, I estimated states™ venue shopping patterns to pursue anti -gay marriage policies using binary logistic regression, complementary log -log regression, and ordered logistic regression models. The results are reflected in Table D.3 in the Appendix . Both the logit and complementary log -log models fall prey to the same atheoretical treatment of venue choice as past 209 research, but the principal results hold. Still, some researchers may argue that venue choice is not a nominal but rather an ordered choice, where policy actors begin by pursuing statutory bans in the legislature and then pursue constitutional bans via legislative referenda or citizen initiative. 120 The results , however, remain robust to the ordered logit model specification . As a further inquiry , I also estimate d a reduced form of the multinomial logistic regression model for the anti -gay marriage policies using a Cox proportional hazards model, stratified by venue (Table D.4 in the Appendix) . Cox proportional hazards models are useful because these models make fe w assumptions about the functional form of the hazard rate (Box -Steffensmeier and Jones 2004). The results are also broadly consistent with the findings from the multinomial logistic regression model. However, b ecause the Cox model is reduced form (since i t does not handle time -dependent variables as well as other approaches) and because it assumes that hazard functions in different strata are proportional over time (which may not be the case for competing venues), I stick with the multinomial logistic mode l specification (Box -Steffensmeier and Jones 2004). I follow the same robustness checks for the pro -gay marriage models as I did for the anti -gay marriage models (Table D.5 and Table D.6 in the Appendix). All the coefficients in these models tell roughly t he same story, except for the political learning coefficien ts. The political learning coefficients not only changed signs across the logit, complementary log -log, ordered logit, and Cox -proportional hazard models, but also did not reach conventional levels of statistical significance. On its face, such results appear to undermine my theory for political learning™s role in patterns of venue shopping across states. But when we consider political learning™s divergent effects across venues Ša positive impact on state legislature and a negative impact on state and federal courts Šeffectively canceling out the effects, these alternative estimations reinforce the need to model these discrete choices using 120 Although that certainly was the process for many states, it was not the case for all states. Nebraska, for example, never adopted any statutory language, but rather adopted a constitutional amendment via citizen initiative. M oreover, other states repeatedly picked the same venue to press for change, even after initial success. 210 a multinomial logistic specification. Considering the Cox mode l, it is essential to remember that political learning™s effect was variable within forums over time. Such variation in the hazard rate is a violation of the proportional hazards assumption of the Cox model. Hence, the Cox model is also not appropriate for the venue diffusion dynamics at play. Another legitimate concern is my operationalization of the independent variables. To ensure the results were not a product of any one measurement choice or the omission of a variable, I re -ran the multinomial lo gistic models for anti - and pro -gay marriage using alternate measures and additional variables. 121 Although some parameter estimates ebbed and flowed, none of these other variables or alternate measures diminished political learning ™s effect on venue choice. Conclusion In this chapter, I set out to test whether policy actors™ choice of venue, a central element of the agenda -setting process and to achieving policy change, was influenced by early movers™ venue shopping strategies. Both the qualitative and quantitative evid ence presented in this chapter support 121 I include d variables to capture a state ™s institutional context, including measures for the number of times a state ™s constitution has been amended (L utz 1994; Wall 2008), the state ™s amendment rate (Lutz 1994; Wall 2008), the length of its constitution (Lutz 1994; Wall 2008), how difficult it is to amend the state constitution (Lupia et al. 2010; Lutz 1994), whether state supreme court judges are elect ed (Hume 2011; Wall 2008), how insulated the legislature is (Bowler and Donovan 2004), whether states permit direct democracy (Bowler and Donovan 2004), how often states use direct democracy (Bowler and Donovan 2004; Lewis 2011, 2013), if states have term limits for legislators (Sarbaugh -Thompson 2010), and how much power the state Speaker of the House has (Mooney 2013). I also included different variables to account for political contexts that may matter to policy actors ™ venue choice, including party cont rol (Klarner 2013 a; Ranney 1976) and electoral competitiveness (Klarner 2013 a; Ranney 1976). In addition, I tested whether southern states behaved differently than states in other regions. Finally, I included the legislative professionalism, citizen ideolo gy, state supreme court professionalism, state supreme court ideology, and district court ideology variables in lieu of the ir corresponding similarity measures. Regarding my decision to measure the strength of the religious right using a state™s percentage of the population that identifies as Evangelical Christian or Mormon, and the strength of gay rights groups using the percentage of a state™s population that identifies as LGBT, I recognize that population size is not synonymous with interest gr oup strength. To be sure, these measures are only proxies for interest group capacity and resources. While there are alternative measures for interest group budgets, assets, income, staff size, and membership size (Conger and Djupe 2016; Haider -Markel 2001 a, 2001b; Kane 2003; Soule 2004; Taylor et al. 2019) for these respective epistemic communities, they either are not consistent across both communities or they are only available for a handful of years. Given this, I opt for the measures that is consistent for both communities and available across the time span of interest. Although crude, I believe these measures do tap into the size and strength of the religious right and LGBT interest group organizations. 211 the theory of venue diffusion. The narrative around the gay rights movement™s and the religious right™s countermovement™s fight over marriage equality provides a strong account of policy actors following policy pionee rs™ venue choice. Furthermore, the empirical results for the anti -gay marriage and pro -gay marriage models also bolster the existence of venue diffusion, even though the state interest group variables were not as prominent as I expected. As the models show , policy actors learn about and rely on the successful and failed venue shopping decisions in prior states when selecting their own avenue to press for change. And actors particularly look to peer states. To be sure, these results are probabilistic and not deterministic. Policy actors are not guaranteed to follow the lead of policy pioneers and early movers, but rather more likely to follow their lead. But this finding greatly expands our understanding of venue shopping. Policy actors™ venue shopping is bot h an internal, intra -jurisdictional process and an external process. Moreover, this chapter only presents evidence of venue diffusion in the context of a technically simple, highly salient morality policy. Although morality policies are different from ot her policy domains (Biggers 2014; Haider -Markel 1999; Mooney and Lee 1995; Mooney and Schuldt 2008 ), there are reasons we might expect political learning™s influence to be weakest in this policy case. Morality policies are typically marked by acute adoptio n rather than gradual learning. Therefore, the evidence of learning here suggests that more complex and less salient policies could yield even greater political learning. Future research should test the phenomenon of venue diffusion in other policy cases, different institutional venues, and other governmental jurisdictions. 212 CHAPTER 7: CONCLUSION The Takeaway In 1932, U.S. Supreme Court Justice Louis Brandeis penned in his opinion for New State Ice Co. v. Liebmann that fi a single courageo us State may, if its citizens choose, serve as a laboratory; and try novel social and economic experiments without ris k to the rest of the country...fl (Brandeis 1932 ). For Brandeis, one of the advantageous features of American federalism was the ability of multiple governmental jurisdictions to try different policy solutions to meet social, economic, and political problems Što be filaboratories of democracy.fl In the era of finew federalism,fl where the federal government has devolved greater policy implementati on responsibility to the states, opportunities are ripe for experimentation. But the reality is that states often act as emulators and not experimenters (Karch 2007 a; Parinandi 2014). As Karch (2007 a: 204) put it, fi[t]he characterization of the fifty state s as laboratories of democracy is an appealing image, but it is a standard that is rarely met in practice.fl Rather than fifty states carrying out trials and errors to determine the best solution for their problems, states learn from and copy the successful policies enacted in other states. This dissertation adds further evidence that policy actors emulate other states™ innovative ideas, even outside the legislative context. Relying on a random sample of ballot measures pursued across the U.S. states from 1 902 Œ 2016, I show that states learn about and copy the solutions proposed in other states. For every ten states that adopt a given ballot measure, subsequent states are one percentage point more likely to enact the same measure in any given year. Policy a ctors pay special attention to states with similar institutional arrangements and are more likely to pursue ballot measures in direct -democracy states and during presidential election years. Meanwhile states with greater interest group activity witness var ying degrees of ballot measure success. Similarly, using the policy case of gay marriage, I show that both anti - and pro -gay marriage policies diffused via multiple institutional venues across U.S. states. Policy actors learned about and 213 acted on what oth er states had achieved in some forums more than others (e.g., legislature, legislative referendum), even after controlling for prior policy activity, opposition policy successes, federal -government involvement, interest -group strength, institutional settin gs, and political contexts. Despite offer ing robust evidence that policy ideas do indeed spread in arenas beyond the legislative context, my research also suggests that policy scholars may be overstating policy diffusion™s existence . In reading the literature, it is easy to get the impression that all policies diffuse. Disciples of diffusion research are finding false gods in governments™ policy activity. Returning to the random sample of ballot measures pursued across U.S. states fr om 1902 Œ 2016, I find that nearly half of the measures were only pursued by one state. That is, these policy solutions did not appear to diffuse to others, at least not via plebiscitary action. And roughly three -quarters of the measures either did not dif fuse or were only pursued by fewer than a handful of states. Only six percent of the ballot measures were adopted by more than fifteen states. None of this is surprising because many policy solutions address provincial rather than universal problems. We sh ould not expect m ost policies to transfer to other jurisdictions. Likewise , when the ballot measures that have yet to diffuse or only diffuse narrowly are excluded from the models, the key mechanism™s Špolicy learning Šeffect is twice as large, potentially exaggerating its role in the process. Overall, I encourage policy scholars to consider these selection biases and rely on full policy sets (i.e., those innovations that do and do not diffuse) where possible to better estimate and explain the realities of di ffusion. Beyond simultaneously reaffirming and cautioning the existence of policy diffusion in various institutional arenas, this research has also contributed much more. Largely overlooked by prior research, I put forward political learning Šthe drawing of lessons about how best to maneuver within and manipulate the policy process to advance a new idea (Heclo 1974; May 1992) Šas a central mechanism in the diffusion process. By operationalizing and including it in the models, 214 political learning appears as i mportant as, if not more than, policy learning in driving diffusion. At least in the context of gay marriage, political learning™s marginal effect on policy adoption was larger than nearly all other external and internal factors. Ultimately, political lear ning should emerge as a mainstay in future research explaining policy transfer. Furthermore, I demonstrate the power of modeling the spread of innovations across multiple venues using multinomial logistic regression. Prior scholarship has either discounte d the variation in venue when choosing a modeling strategy or disregarded altogether policies that transfer via competing institutional paths. Multinomial logistic regression concurrently reveals the factors necessary for policy adoption in each venue and the inter -venue dynamics at play. As such, multinomial logistic regression should be added to policy scholars™ toolbox to uncover and explain policy contagion. Most importantly, however, my research integrates the policy diffusion and venue shopping literatures. In contrast to past venue shopping scholarship which identifies internal and intra -jurisdictional considerations as central to policy actors™ calculations in choosing a path to pursue policy change, I theorize that policy actors also weigh exter nal information. I charge that a state™s choice of venue to attempt a new policy is influenced by the venue shopping of other states previously pursuing the policy, a phenomenon I term venue diffusion. The choice of venue is no small matter. Venue choice a ffects the policy™s design, implementation, stakeholders, evaluation, and whether it survives. Venue choice is the product of prior decisions made in multiple ‚policy games™ and will affect future decisions ( Boehmke, Gailmard, Patty 2006; Karch 2009; Lubel l 2013) . I see policy actors in states not only as emulators of new ideas but also as emulators of political paths to enact those policies. Much like cross -country skiers follow the snow trails cut by previous skiers, policy actors are also more likely to follow the political route cut by policy entrepreneurs. This does not mea n that policy actors will never go off the trail. Some policy 215 advocates will follow their predisposition for an institutional arrangement or engage in an insular, independent process to determine the best venue to advance their cause. Most policy actors , however, will learn from the paths taken in other states, especially those jurisdictions with similar institutional arrangements and political contexts. Again, utilizing the policy case of gay marriage, I offer both qualitative and quantitative evide nce that the choice of venue in early mover states to press for an anti - or pro -same -sex union policy influences subsequent states™ venue selection . As policy actors learned about the successful paths picked by other states, they were more likely to take t he same route. The legislative branch was primarily susceptible to political learning, with legislators facing a higher propensity to pursue statutory action or refer policies to the state electorate as the other states™ success rate in those venues increa sed. Because legislators are electorally motivated (Mayhew 1974), they may be expressly attuned to success in other states. In addition, policy actors were prone to take venue shopping cues from jurisdictions similar along institutional or ideological dime nsions. And policy actors weighed signals from other states™ venue shopping processes in real time. As one route became less certain, policy actors in subsequent states would turn to a venue exhibiting a greater chance of success. Political learning™s vari able effect over time further reinforces the phenomenon of venue diffusion. In sum, policy actors consider their own capacities and their state™s internal characteristics, along with the tactical venue shopping done in other states when picking a path to p ress for change. Unanswered Questions Despite these contributions, a few questions remain. Some may rightly wonder whether policy actors purs uing policies via three competing venues really offers evidence of venue diffusion. Skeptics might question if policy actors truly learned from and copied others™ choice of venue, why then did states pick multiple venues, instead of one arena, to upend the status quo? They might 216 conclude that such behavior reflects typical venue shopping rather than venue diffusion. To be sure, I expect political learning™s effect on venue choice to be greatest when only one venue is utilized to alter pu blic policy. The overwhelming majority of diffusion research has focused on the legislative context because this is the venue where most policy activity occurs (although decreasing since the 1970s). Institutional hurdles tend to be lower relative to the po tential gains for policy success and entrenchment. In their purposive search for policy solutions to common problems, policy actors should also gain information about the successful tactics and paths taken by policy entrepreneurs and early movers. If all s tates follow the same route, this reinforces the process. But w e are unable to model venue diffusion without variation on the dependent variable. Furthermore, empirically parsing policy learning and political learning becomes more challenging in a one -venu e context. The inability to offer empirical evidence of political learning and venue diffusion, however, does not imply these processes are not occurring. On the contrary, any evidence of these phenomena occurring in multiple arenas should reinforce that t hey occur to a greater extent when only one arena is involved . Another fair question is whether alternative conceptualizations of political learning are appropriate. Beyond the standard operationalization of political learning, Chapter 5 did include institutional and political similarity variables. These measures do embody an aspect of the learning process, as policy actors are more likely to turn to and emulate their peers. Just like Graham, Shipan, and Volden (2013) challenged us to the come up with m ore direct measures of policy learning, other explicit measures of political learning may be needed. For instance, measures reflecting the degree of a policy win or loss in a given venue may serve as another proxy for political learning. Or a variable capt uring similarity of campaign tactics or messaging employed by policy actors across states may also imply political learning. Subsequent research should further explore and test diverse operationalizations of political learning. 217 Finally, given the narrative around state -level interest groups facilitating the venue shopping to press for anti - and pro -gay marriage policies, it is somewhat surprising the empirical models did not reveal greater support for the interest group strength variables. There are at leas t three plausible explanations for this. First, it is possible a different operationalization of interest group strength could better showcase state -level religious right and gay rights groups effect on policy and venue diffusion. This is unlikely as I car ried out robustness checks with more than a dozen measures of interest group strength and did not reach different conclusions. Second, pressure groups™ real influence may be captured via the political learning and similarity variables. Because organized gr oups are the main actors engaging in purposive searches for venue shopping information, their influence in the process may be accounted for in these variables. As Lowery (2013) reminds us, null findings for interest group variables do not imply no influenc e in the policy process. This is why mixed -methods approaches to understanding the diffusion process are still required (Starke 2013). The last reason that state -level interest groups™ effect may have been overshadowed is that, as the narrative also point ed out, national -level pressure groups were prominent players in communicating the successes and failures across subunits. National -level influence should not undermine the theory of venue diffusion or the main political learning mechanism, just as their presence does not undermine the theory of policy diffusion. Even if national organizations helped reduce information costs across states, they still had to work with state -level groups and via state -level venues. National groups did not simply supplant loca l groups™ goals. In fact, there is robust evidence that state - and local -interests disobeyed the national organizations™ recommended strategies, frequently innovating and surmising new tactics independent from national groups. I opt not to control for nati onal interest groups in the model because these groups are not monolithic. Variation in clout and strategy existed within the groups representing each movement, as the qualitative evidence pointed out. Treating them as uniform with one national -level measu re could be 218 disingenuous. Moreover , each side™s strength is likely correlated with time, with the religious right losing influence on the issue over time and gay rights groups gaining leverage in due course. Thus, the inclusion of national -level measures i s unlikely to produce divergent results. Moving Forward This research acknowledges that policy actors frequently, and increasingly so, turn to different venues outside of the legislative context to pursue policy change. I leverage America™s federated, multi -institutional system to understand better the interdependence among policy actors in emulating new policy solutions and the paths to enact such innovations. Overall, the findings imply a more systematic and greater level of connectedness among chang e agents in venue shopping and policy adoption than policy scholars have previously acknowledged. Still, the robust evidence presented here for venue diffusion and political learning™s role in the process stems from a s ingle policy issue: gay marriage. B ecause morality policies tend to be marked by rapid diffusion (due to competition over societal and cultural values) rather than gradual learning ( Boushey 2010; Mooney 2001; Mooney and Lee 1999 ), any evidence of political learning in the fight over gay marr iage could suggest a more sizeable effect in other policy domains. Of course, this is an empirical question. Future qualitative (Starke 2013) and quantitative assessments should test the phenomenon of venue diffusion in other policy domains. Just as differ ent policy types display different patterns of policy diffusion , diverse policy domains may also reveal divergent dynamics of venue diffusion . But careful attention should be paid to policy cases selected to test venue diffusion. Importantly, the selectio n of cases fishould allow for the possibility of at least some variation on the dependent variablefl (King, Keohane, and Verba 1994: 129). Exploring the spread of a policy via more than one venue allows us to make inferences about the diffusion of the innova tion, the role of 219 different venues in conditioning its transmission, and the diffusion of venue selection. 122 Furthermore, variation in policy type and attributes ( Makse and Volden 2001; Nicholson -Crotty 2009) should also be considered. For example, s alient and technically simple policies spread more quickly than policies off the public™s radar and requiring greater expertise to design and implement (Makse and Volden 2011 ; Mallinson 2016; Nicholson -Crotty 2009 ). While policy learning tends to be most e vident for high profile policies (even if not for morality policies), it is least apparent for complex policy innovations (Makse and Volden 2011). Venue diffusion and political learning™s role may also fluctuate along policy dimensions of salience and comp lexity . Two potential polic ies that exhibit variation in domain, salience, and complexity (relative to same -sex marriage, anyway) are the passage of tax and expenditure limits (TELs) across U.S. states and state policy restrictions on the use of eminent domain for economic development purposes. Since the 1970s, nearly two dozen states have adopted limitations on state revenue and expenditures via a variety of venues, including state legislatures, legislative referenda, and citizen initiatives. Likewise, following the U.S. Supreme Court™s 2005 Kelo v. City of New London decision, forty -two states passed restrictions on using lawful expropriation for economic development purposes. States passed these measures via state legislation, legislative referenda, an d ballot initiatives, with most of the policy activity occurring within three years of the Kelo ruling. Both t ax and expenditure limits and restrictions on the use of eminent domain represent highly salient yet technical policy area s. Since these policy ar eas garner ed considerable public attention, I anticipate heightened political learning in the venue shopping processes to pass TELs and post -Kelo reforms . Nevertheless, policies that are 122 A careful researcher may rightly ask whether or not I am selecting on the dependent variable by choosing policies that have been pursued via more than one institutional venue. Recall from my prior analysis of 95 policies compiled by Boehmke and Skinner (2012) that many policies are pursued in more than one arena. Venue shopping is not an infrequent occurrence. Moreover, in order to empirically demonstrate political learning between actors in states around the choice of venue, variation in venue must exist. It is entirely possible for political learning to occur when states pursue exactly the same institutional arena for change (say legislature to legislature), but without any variation in the dependent variable it is impossible to show empirically. As a result, the effect of political learning may be aug mented where the adoption of an innovation has been via a uniform venue across all adopters. 220 less salien t and more technical (e.g., licensing, city zoning , energy efficiency building codes ), or even less visible and technically simple , may experience different rates of venue diffusion. 123 Political learning may play a lesser role with venue diffusion less likely to occu r if the policy is not even on the public™ s radar. Outside of variation in policy type and corresponding characteristics, case selection should also emphasize variation in institutional forums, focusing especially on underexplored venues (e.g., gubernatorial executive orders, bureaucratic a gencies). The evidence presented here suggest s that electorally motivated policy actors (e.g., legislators, governors, supreme court judges) may be the most likely to learn from an innovation™s political feasibility in a given arena. Political learning , fo r example, may play a diminished role in bureaucratic agency decisions, although discretionary authority may be a conditioning factor (Parinandi 2013). The pace of venue diffusion may also be of interest, as some routes may transpire more quickly than othe rs. In addition, the theory of venue diffusion could apply beyond a U.S. state context. For example, policy actors working at the city level may look to others™ municipal venue shopping tactics (e.g., city council vote, ballot measure, mayoral executive d irective) previously applied by other local jurisdictions within or outside the state. Or policy actors may look cross -nationally to determine which institutional paths were most successful in achieving a policy innovation in other countries . Still, policy actors are most likely to look to units with analogous institutional and political settings. Given the variation in institutional and political arrangements across cities and countries, we might expect political learning™s effect to attenuate in the se contexts. And for modeling purposes, especially for multinomial logistic regression, units of interest (e.g., states, cities, nation -states) are assumed to have all discrete choices available during the time period. Too much variation in 123 Koski (2010) documents the diffusion of a low -salient and complex policy across 119 U.S. cities from 2000 to 2008: green building design standards. This may present an opportunity to test the theory of venue diffusion using a different policy area with distinct attributes and at the municipal level. 221 availability of venues across units would violate this assumption and could complicate inferences drawn from these diverse institutional and political environments. Still, it is possible that p olicy actors seek and process external venue shopping information from multipl e vertical or horizontal sources when deciding the most favorable avenue to press for change. Thinking beyond policy and venue diffusion, p olicy actors™ learning may not stop with new solutions or viable venues but may also extend to the other parts of t he policy cycle including policy design, policy winnowing, policy framing, agenda setting, implementation, evaluation, policy feedback, policy reform, among others . For example, actors may draw lessons from policy entrepreneurs ™ or early movers™ use of a s pecific frame or tactic to rout the opposition , as Gilardi, Shipan, and Wueest (2019) show . Or evaluators intent on retaining a policy may rely on similar metrics to paint the outcomes in the best light possible. We should not be so naïve to assume that such interdependence among policy actors stops with learning about policy solutions. Surely policy actors gain information and draw lessons throughout the lifespan of a policy. And it is likely that political lessons learned from one policy can also be appli ed to similar policies. 124 Indeed, five decades of policy diffusion research have reinforced the idea that policy actors are interconnected. Multiple jurisdictional layers of American government facilitate an environment where elected officials, int erest groups, and citizen activists can learn from one another, satisficing for policy solutions and the political paths to achieve such solutions. Boundless opportunities await scholars to tease out other aspects of the policy process dominated by this in terdependence. 124 In fact, subsequent research should explore whether the political lessons learned from venue shopping in the pursuit of anti - and pro -gay marriage policies are also transferrable to related policies that halt or hasten LGBTQ rights. Do policy actors and interest groups apply their venue shopping experiences from the fight over same -sex unions to press for or impe de LGBTQ -friendly adoption, housing, and employment policies, among others? Evidence of this would suggest an enduring process of political learning by actors and groups that is sustained across issue and time. 222 APPENDICES 223 APPENDIX A Figure A.1: Ballot Measures by Frequency of Policy Area by Type of Measure Note: Bar chart displays the percent frequency of ballot measures by policy area by type of measure (e.g., citizen initiatives, legislative referendum, popular referendum, others) attempted across the U.S. states from 1902{2016. Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. 224 Process for Matching Sample Ballot Measures Across Database From the full set of nearly 7,800 ballot measures (e.g., legislative referendums, citizen initiatives, popular referendums, other ballot measures) pursued across the U.S. from 1902 -2016 (Jordan and Grossmann 201 8; NC SL 2016), I randomly selected 50 ballot measures. Figure A.2 displays the policy areas represented by these 50 ballot measures. Although there are some policy topics unrepresented in the random sample, the distribution across policy areas parallels the dis tribution of the full set of ballot measures (see Figure 2.5 in the main text of the chapter for the distribution of topic area for the complete set). As further evidence of the random sample™s representation relative to the full set, Figure A.3 displays t he number of ballot measures pursued across the states by decade. The bimodal distribution of measures pursued over time in the random sample mirrors the bimodal distribution of measures over time in the full set, per Figure 2.1 in the chapter. After sam pling, I matched the 50 ballot measures by title, topic, and type to analogous measures pursued by the original or alternative U.S. states during the full time period. More specifically, each randomly selected ballot measure was assigned a unique policy id. In order to pair the selected ballot measure with similar measures pursued by other states, I searched for common key terms across the full set of ballot measures to identify and match parallel policies . For example, for a ballot measure adoptin g language to expand the production and sale of alcoholic beverages, I searched for the following terms: fialcohol,fl fialcoholic beverages,fl fibeer,fl filibations,fl filiquor,fl fispirits,fl and fiwine.fl Upon finding other ballot measures with these key terms, I woul d assess the intent of the ballot measure to see if matched the originally sampled measure. Returning to the alcohol example, a measure that allowed the sale of alcohol throughout the state or counties would be considered a fimatchfl for the policy id, while a measure that included one of the key words but 225 Figure A.2: Random Sample of 50 Ballot Measures by Topic Area Note: Bar chart displays the policy topic area for 50 randomly sampled ballot measures from the full set of nearly 7,800 ballot mea sures pursued across the U.S. states from 1902 -2016. Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measur es Database. that prohibited the regulation of or sale of alcoholic beverages would not be considered a fimatch.fl Although key words helped identify potentially comparable measures, each measure™s title and aim were also evaluated to ensure congruent ball ot measures linked by the ascribed policy id. Furthermore, I relied on the following websites as resources to further investigate intent of ballot 226 Figure A.3: Random Sample of 50 Ballot Measures Pursued by Decade Note: Bar chart displays the number of 50 randomly sampled ballot measures from the full set of nearly 7,800 ballot measures pursued across the U.S. states from 1902 -2016 by decade . Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. measure when unclear: Ballotpedia.com; UC Hastings California Ballot Measures; National Conference of State Legislatures; and respective secretary of state (or equivalent) offices. Not all ballot measures were matched, as many initiatives and r eferendums were only pursued within one state. Indeed, many problems or issues are unique to a state and require an individual solution. Nonetheless, this matching exercise produced a small sub -sample of 579 ballot measures (or 7.4% of the full set) , conta ining those ballot measures that did diffuse and those that did not diffuse. 227 Figure A.4: Random Sample of 50 Ballot Measures Pursued by Topic Area Note: Bar chart displays the number of 50 randomly sampled ballot measures from the full set of nearly 7,800 ballot measures pursued across the U.S. states from 1902 -2016 by topic area for those measures that appear to have not diffused (top chart) and those measures that may have diffused (bottom chart) . Source : National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database. 228 Table A.1 : Ballot Measures Model™s Var . Descriptions, Descriptive Statistics, and Sources Variable Name Description Mean Sd. Dev. Min Max Sources Policy Learning Cumulative number of states pursuing specific ballot measure by year 5.47 5.66 1 20 Author , relying on random sample and matching exercise Similarity in State Revenue per capita Euclidean distance between state™s revenue per capita and average revenue per capita for states in a given year. Reverse coded so that an increase indicates more similar. -0.52 1.09 -17.4 0 Author, using Klarner™s 2013 b measure Similarity in State Party Control Euclidean distance between state™s party control and averag e party control of states in a given year. Reverse coded so that an increase indicates more similar. -0.28 0.20 -0.77 0 Author, using Klarner 2013 a; Ranney 1976 measures Similarity in Citizen Ideology Euclidean distance between state™s citizen ideology and average ideology of states in a given year. Reverse coded so that an increase indicates more similar. -17.61 11.90 -50.2 0 Author, using Berry et al. ™s 1998, 2010 measure Similarity in Difficulty in Amending State Constitution Euclidean dis tance between state™s difficulty in amending constitution and average difficulty of states in a given year. Reverse coded so that an increase indicates more similar. -0.67 0.50 -1.96 0 Author, using Lupia et al. ™s 2010 measure Dir ect Democracy State Dummy=1 if state allows direct or indirect citizen initiatives 0.44 0.50 0 1 Ballotpedia 2016; Lupia et al. 2010; NCSL, Waters 2003 Statutory Legislative Referendum State Dummy=1 if state permits legislature to refer statutory language to voters 0.48 0.5 0 0 1 Ballotpedia 2016; Lupia et al. 2010; NCSL, Waters 2003 Popular Referendum State Dummy=1 if state permits citizens to repeal policies adopted by elected officials 0.52 0.50 0 1 Ballotpedia 2016; Lupia et al. 2010; NCSL, Waters 2003 Electoral Competitiveness Ranney measures of competitiveness, Four -Year Moving Average. Varies between .5 and 1, higher values representing higher 0.84 0.13 0.5 1 Klarner 2013 a; Ranney 1976 Pres. Election Year Dummy=1 if presidential election in that calendar year, 0=none 0.26 0.44 0 1 Author Government Reform Measures Dummy=1 if ballot measure deals with government reform of state, localities, judiciary, or related to federal governmental issues 0.31 0.46 0 1 NCSL 2016 Bond and Budget Measures Dummy=1 if ballot measure deals with bonds and budget issues 0.20 0.40 0 1 NCSL 2016 Tax and Revenue Measures Dummy=1 if ballot measure deals tax and revenue issues 0.14 0.35 0 1 NCSL 2016 Evangelical Population Percentage of population that identifies as Evangelical Christian or a member of the Church of Latter -day Saints 19.07 14.29 1.1 74 Sellers 2017 Union Membership Density Percentage of workforce that is unionized 16.58 8.26 2.3 44.8 Hirsch and Macpherson 2003 229 Table A.1 (cont™d) GINI Inequality Measure Measure of state™s variation in distribution of residents™ income and wealth where higher values indicate greater inequality 0.51 0.08 0.23 0.75 Frank 2009 California Dummy Dummy= 1 for California, 0 = all other U.S. states 0.02 0.14 0 1 Author Oregon Dummy Dummy= 1 for Oregon, 0 = all other U.S. states 0.02 0.14 0 1 Author Southern State Dummy= 1 if state is located in the South 0.30 0.46 0 1 U.S. Census Bureau State Population (Ln) Natural log of state population (in the thousands) 7.82 1.07 3.95 10.52 U.S. Census Bureau 230 APPENDIX B TABLE B.1: CHOICE OF VENUE AND DIFFUSION STATISTICS FOR SAMPLE OF 95 POLICIES Policy Policy Category Leg. Leg. Ref . Cit. Init. Morality Issue Diffus ion Start s Diffus ion End s Diffus ion No. of Years No. States Adopting Ratio : States / Years 1-parent Consent for Abortion by a Minor Abortion 1981 1999 18 15 0.83 1-parent Notification for Abortion by a Minor Abortion 1981 2000 19 17 0.89 Abortion pre -Roe Abortion 1966 1972 6 18 3.00 State Law Requiring Broad Community Notification of Sex Offenders Crime 1990 1997 7 18 2.57 Capital Punishment Crime 1972 1982 10 39 3.90 Child Abuse Reporting Legislation Crime 1963 1967 4 48 12.00 Civil Injunction Authority Crime 1998 200 1 3 15 5.00 Strategic Planning for Corrections Crime 1970 1991 21 18 0.86 Cyberstalking Definition and Penalty Crime 1998 200 1 3 21 7.00 Harassment Crime Crime 1998 200 1 3 11 3.67 State Hate Crime Laws Crime 1978 1994 16 33 2.06 ID Theft Protection Crime 1996 200 1 5 44 8.80 State Law Requiring Notification to Individuals/Organizati ons at Risk (Sex Offender Policy) Crime 1994 1997 3 14 4.67 Post -Conviction DNA Motions Crime 1997 2005 8 35 4.38 Access to Sex Offender Registries Crime 1991 1997 6 15 2.50 Stalking Definition and Penalty Crime 1998 200 1 3 24 8.00 Age Span Provisions for Statutory Rape Crime 1950 1998 48 43 0.90 Three Strikes Sentencing Requirement Crime 1993 1995 2 24 12.00 Victims' Compensation Crime 1965 1988 23 42 1.83 Victims' Rights Constitutional Amendment Crime 1982 1999 17 32 1.88 .08 per se penalty for DUI Drugs and Alcohol 1983 200 1 18 25 1.39 Beer Keg Registration Requirement Drugs and Alcohol 1978 1999 21 12 0.57 Symbolic Medical Marijuana Policy Drugs and Alcohol 1978 2008 30 31 1.03 Restrictions on OTC Medications with Methamphetamine Precursors Drugs and Alcohol 1996 2005 9 25 2.78 Minimum Legal Drinking Age 21 Drugs and Alcohol 1933 1988 55 50 0.9 1 Statewide Smoking Ban Drugs and Alcohol 1995 2009 14 25 1.79 231 Table B.1 (cont™d) Policy Policy Category Leg. Leg. Ref . Cit. Init. Morality Issue Diffus ion Start s Diffus ion End s Diffus ion No. of Years No. States Adopting Ratio : States / Years Zero Tolerance (<.02 BAC) for Underage Drinking Drugs and Alcohol 1983 1998 15 50 3.33 Planning Laws Requiring Loc/Reg Planners to Coordinate Growth Management Plan Developments Economic 1961 1998 37 10 0.27 Strategic Planning for Economic Development Economic 1981 1992 11 24 2. 18 Electricity Deregulation Economic 1996 1999 3 24 8.00 State Enterprise Zones Economic 198 1 1992 11 38 3.45 Charter Schools Education 1991 1996 5 25 5.00 Strategic Planning for Education Education 1970 199 1 21 14 0.67 High School Exit Exams Education 1976 1999 23 26 1.13 School Choice Education 1987 1992 5 16 3.20 Strategic Planning for Environmental Protection Environ . 1978 199 1 13 14 1.08 Strategic Planning for Natural Resources Environ . 1975 199 1 16 16 1.00 Interstate Pest Control Compact Environ . 1968 2009 41 36 0.88 State Renewable Portfolio Standards Environ . 199 1 2004 13 19 1.46 State allows Tribal Gaming Gambling 1990 1995 5 24 4.80 Lottery Gambling 1964 1993 29 36 1.24 Constitutional Amendment Banning Gay Marriage Gay Rights 1994 2008 14 33 2.36 Unrestricted Absentee Voting Government 1960 2003 43 26 0.60 In-Person Early Voting Government 1970 2002 32 15 0.47 Voter Registration by Mail Government 1972 1995 23 49 2. 13 Missouri Plan Government 1940 1976 36 20 0.56 Voter Registration with Driver's License Renewal Government 1976 1995 19 49 2.58 State Policy to Refuse to Comply with 2005 Federal Real ID Act Government 2007 2009 2 18 9.00 Public Campaign Funding Government 1973 1987 14 23 1.64 Protections Against Compelling Reporters to Disclose Sources in Court Government 1935 2009 74 34 0.46 Legislative Term Limits Government 1990 2000 10 15 1.50 Child Access to Guns Protection Law Gun Control 1989 2000 11 17 1.55 Strategic Planning for Aging Health 1974 1991 17 19 1.12 232 Table B.1 (cont™d) Policy Policy Category Leg. Leg. Ref . Cit. Init. Morality Issue Diffus ion Start s Diffus ion End s Diffus ion No. of Years No. States Adopting Ratio : States / Years Ban on Financial Incentives for Doctors to Perform Less Costly Procedures/Prescribe Less Costly Drugs Health 1996 200 1 5 29 5.80 Prohibits Agreements that Limits a Doctor's Ability to Inform Patients of All Treatment Options Health 1975 1999 24 46 1.92 Colorectal Cancer Screening Health 1991 2007 16 27 1.69 Insurers That Cover Prescription Drugs Cannot Exclude FDA -Approved Contraceptives Health 1996 2007 11 27 2.45 Strategic Planning for Health Services Health 1985 1991 6 23 3.83 Guaranteed Issue of Health Insurance Health 1990 1994 4 36 9.00 Guaranteed Renewal of Health Insurance Health 1990 1995 5 45 9.00 Health Insurance Portability Health 1990 1995 5 43 8.60 Health Insurance Preexisting Conditions Limits Health 1990 1994 4 39 9.75 Health Maintenance Organization Model Act (First) Health 1973 1988 15 23 1.53 Health Maintenance Organization Model Act (Second) Health 1989 1995 6 22 3.67 Newborn Hearing Screening Health 1990 2008 18 43 2.39 Mandated Coverage of Clinical Trials Health 1994 2008 14 23 1.64 Medical Savings Accounts Health 1993 1997 4 28 7.00 Prescription Drug Monitoring Health 1940 1999 59 14 0.24 Right to Die Health 1976 1988 12 15 1.25 Dependent Coverage Expansion Insurance for Young Adults Health 1994 2008 14 25 1.79 Senior Prescription Drugs Health 1975 200 1 26 27 1.04 Fair Employment Laws Labor Rights 1945 1964 19 25 1.32 Bottle Deposit Law Misc . 1971 2002 31 11 0.35 Restrictions on Displaying Credit Card Numbers on Sales Receipts Misc . 1999 2008 9 31 3.44 Limits Credit Agencies from Issuing a Credit Report without Consumer Consent Misc . 200 1 2006 5 25 5.00 Grandparents' Visitation Rights Misc . 1964 1987 23 50 2. 17 Living Wills Misc . 1976 1986 10 38 3.80 233 Table B.1 (cont™d) Policy Policy Category Leg. Leg. Ref . Cit. Init. Morality Issue Diffus ion Start s Diffus ion End s Diffus ion No. of Years No. States Adopting Ratio : States / Years Provisions by the States Maintaining Segregated Educational Systems for Out -Of-State Study by African -Americans Racial Issues 1927 1943 16 10 0.63 State Income Tax Tax 1916 1937 21 28 1.33 Lien Statutes Tax 1995 1999 4 27 6.75 Strategic Planning for Revenue Tax 198 1 1991 10 18 1.80 Tax and Expenditure Limits Tax 1976 1994 18 26 1.44 Child Seatbelt Requirement Transport . 1981 1984 3 49 16.33 State Graduated Driver's Licensing Program Transport . 1996 2009 13 49 3.77 Mandatory Bicycle Helmets for Minors Transport . 1992 2007 15 21 1.40 Lemon Laws Transport . 1982 1984 2 29 14.50 Motorcycle Helmet Requirement Transport . 1967 1985 18 50 2.78 Primary Seat Belt Laws Transport . 1984 2004 20 21 1.05 Strategic Planning for Transportation Transport . 1974 1991 17 20 1.18 Family Cap Exemptions Welfare 1992 1998 6 21 3.50 Individual Development Accounts Welfare 1993 200 1 8 35 4.38 Kinship Care Program Welfare 1998 2006 8 26 3.25 Time Limits on Welfare Benefits Welfare 1993 1996 3 18 6.00 Special Agent/Office for Women's Health Women's Rights 1993 2009 16 19 1.19 Allowance of Breastfeeding in Public Women's Rights 1993 2008 15 46 3.07 Note: A sample of 95 diverse policies (1916 Œ 2009), compiled by Boehmke and Skinner (2012), were assessed for the choice of institutional venue Šstate legislature, legislative referendum, citizen initiative or popular referendum Šwhere the policies were pursued by at least one state via those venues. The table also includes the years the first and last states adopted the policy, the number of states that have enacted the policy, and the r ate of adoption, as measured by the number of states passing the policy over the timeframe between the first and la test adoption. Leg. indicates Legislature, Leg. Ref. indicates Legislative Referendum, and Cit. Init. indicates Citizen Initiative. 234 APPENDIX C Table C.1: State by State Chronology of Anti - and Pro -Gay Marriage Policies, 1993 Œ2015 AL: o 1996 Anti Executive Order Pass o 1998 Anti Legislature Pass Statute o 2006 Anti Legislative Referendum Pass Amendment 774, Constitutional Amendment o 2015 Pro Circuit Court Pass Searcy v. Bentley , decided by Di strict Court, but Circuit and S. Courts refused to issue stay AK: o 1996 Anti Legislature Pass Statute, became law without gov.™s signature o 1998 Anti Legislative Referendum Pass Ballot Measure 2, Constitutional Amendment o 1998 Pro State High Court Fail Bess v. Ulmer , Superior Ct ruled for gay marr in ‚94, but Supreme Ct overturned bec Meas.2 o 2014 Pro Circuit Court Pass Hamby v. Parnell , denied stay by C. Court and Supreme Court AZ: 125 o 1975 Anti Legislature Pass Statute o 1996 Anti Legislature Pass Statute o 2006 Anti Citizen Initiative Fail Proposition 107, Constitutional Amendment o 2008 Anti Legislative Referendum Pass Proposition 102, Constitutional Amendment o 2014 Pro Circuit Court Pass Connolly v. Roche , Majors v. Horne , decided by District Ct, but C. Ct suspended proceedings AR: o 1997 Anti Legislature Pass Statute o 2004 Anti Citizen Initiative Pass Proposal 3, Constitutional Amendment CA: 126 o 1977 Anti Legislature Pass Statute o 1997 Anti Legislature Fail Statute, efforts put forward in ‚95, ‚96, ‚97, some of which did not make it out of committee o 2000 Anti Citizen Initiative Pass Proposition 22, statutory language reaffirming 1977 statute o 2005 Pro Legislature Pass Statute passed by legislature, but vetoed by governor o 2008 Pro State High Court Pass In re Marriage Cases o 2008 Anti Citizen Initiative Pass Proposition 8, Constitutional Amendment o 2009 Pro State High Court Fail Strauss v. Horton o 2012 Pro Circuit Court Pass Perry v. Brown 127 125 AZ: The Standhardt v. Superior Court cas e to push for gay marriage was heard by State Court of Appeals in 2003 but ruled against. 126 CA: The legislature did pass limited domestic partnership benefits to gay couples in 1999, but the statute did not encompass marriage. 127 CA: The US Supreme Court upheld ruling in Hollingsworth v. Perry in 2013. 235 Table C.1 (cont ™d) CO: 128 o 1975 Pro Circuit Court Fail Adams v. Howerton , sought marriage license in Boulder, for immigration purposes o 1996 Anti Legislature Pass Statute, vetoed by governor o 1997 Anti Legislature Pass Statute, vetoed by governor o 2000 Anti Legislature Pass Statute, signed into law by governor o 2006 Anti Citizen Initiative Pass Amendment 43, Constitutional Amendment o 2013 Pro Legislature Pass Statute, Civil Unions o 2014 Pro Circuit Court Pass Burns v. Hickenlooper , allowing Same -Sex Marriage; also Kitchen v. Herbert in Utah CT: o 2005 Pro Legislature Pass Statute, Civil Unions o 2008 Pro State High Court Pass Kerrigan v. Commissioner of Public Health DE: o 1996 Anti Legislature Pass Statute o 2011 Pro Legislature Pass Statute, Civil Unions o 2013 Pro Legislature Pass Statute, Same -Sex Marriage FL: o 1977 Anti Legislature Pass Statute o 1997 Anti Legislature Pass Statute, became law without governor™s signature o 2008 Anti Citizen Initiative Pass Amendment 2, Constitutional Amendment o 2015 Pro Circuit Court Pass Brenner v. Scott , decided by District Court, but Circuit would not continue stay GA: o 1996 Anti Legislature Pass Statute o 2004 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendmen t o 2006 Pro State High Court Fail Case name unknown, but Lambda Legal and ACLU v. State of Georgia HI: o 1993 Pro State High Court Pass Baehr v. Lewin , but the ruling remanded to trial court, not opening up same -sex marriage o 1994 Anti Legislature Pas s Statute o 1998 Anti Legislative Referendum Pass Amendment 2, Constitutional Amendment, legislature passed law following referendum o 1999 Pro State High Court Fail Baehr v. Miike , lost due to passage of legislative referendum outlawing gay marriage o 2010 Pro Legislature Pass Statute, Civil Unions, vetoed by the governor o 2011 Pro Legislature Pass Statute, Civil Unions, signed into law by the governor 128 CO: The legislature passed a statute in 2009 providing designated beneficiary agreements for gay couples, but this did not in clude the right to marry. 236 Table C.1 (cont™d) o 2013 Pro Legislature Pass Statute, Same -Sex Marriage, signed into law by the g overnor ID: o 1995 Anti Legislature Pass Statute, defining marriage between one man and one woman o 1996 Anti Legislature Pass Statute, mandating no recognition of same -sex marriages performed by other states o 2006 Anti Legislative Referendum Pass Amen dment 2, Constitutional Amend., state senate failed to put before citizens in ™04, ‚05 o 2014 Pro Circuit Court Pass Latta v. Otter , decided by District Court, but Circuit and Supreme Cts. let stay run out IL: o 1996 Anti Legislature Pass Statute o 2011 Pro Legislature Pass Statute, Civil Unions o 2013 Pro Legislature Pass Statute, Same -Sex Marriage IN: o 1986 Anti Legislature Pass Statute, defining marriage between one man and one woman o 1997 Anti Legislature Pass Statute, mandating no recog nition of same -sex marriages performed by other states 129 o 2014 Pro Circuit Court Pass Baskin v. Bogan , Dis. and Cir. Cts decided for gay marriage, Supreme Ct did not take up case IA: o 1998 Anti Legislature Pass Statute o 2009 Pro State High Court Pass Varnum v. Brien 130 KS: o 1996 Anti Legislature Pass Statute o 2005 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendment o 2014 Pro State High Court Pass Schmidt v. Moriarty , but only for the 10 th judicial district; 2015 before entire st ate o 2014 Pro Circuit Court Pass Marie v. Moser , Dis. Ct decided for gay marriage, Cir. and S. Cts would not issue stays KY: o 1973 Pro State High Court Fail Jones v. Hallahan , heard by State Court of Appeals, the state™s High Court at the time o 1998 Anti Legislature Pass Statute o 2004 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendment o 2014 Pro Circuit Court Fail Bourke v. Beshear 129 IN: Members of Indiana Legislature had made annua l attempts from 2004 Œ 2015 to put forward a legislative referendum for a constitutional amendment stating that only a marriage between one man and one woman would be valid and recognized in the state. All of these attempts never passed both chambers of th e state legislature to make it on the ballot . 130 IA: Members of the Iowa Legislature made several annual attempts following the Iowa Supreme Court™s ruling allowing same -sex marriage to put forward constitutional amendments, limiting marriage to one man an d one woman, without success. 237 Table C.1 (cont™d) LA: o 1988 Anti Legislature Pass Statute, defining marriage between one man and one woman o 1999 Anti Legislature Pass Statute, mandating no recognition of same -sex marriages performed by other states o 2004 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendment o 2015 Pro State High Court Pass Costanza v. C aldwell , trial Ct ruled for gay marriage, but St. SC did not rule until Fed. SC did o 2015 Pro Circuit Court Fail Robicheaux v. George , Dis. Ct. ruled against gay marriage, but C Ct. did not decide until SC ME: o 1997 Anti Legislature Pass Statute o 2009 Pro Legislature Pass Statute o 2009 Anti Popular Referendum Pass Question 1, Repeal of former Statute o 2012 Pro Citizen Initiative Pass Question 1, Statute allowing Same -Sex Marriage MD: o 1973 Anti Legislature Pass Statute 131 o 2007 Pro State High Court Fail Conaway v. Deane , ban on gay marriage ruled constitutional by Court of AP, MD™s High Court o 2012 Pro Legislative Referendum Pass Question 6, Statute MA: o 2004 Pro State High Court Pass Goodridge v. Department of Public Health , overtu rning historic marr. statute define 1man 1wman o 2004 Anti Legislature Fail Constitutional Amendment, via Constitutional Convention o 2005 Anti Legislature Fail Constitutional Amendment, via Constitutional Convention o 2006 Anti Legislature Fail Consti tutional Amendment, via Constitutional Convention o 2007 Anti Legislature Fail Constitutional Amendment, via Constitutional Convention MI: o 1995 Anti Legislature Pass Statute o 2004 Anti Citizen Initiative Pass Proposal 2, Constitutional Amendment o 2014 Pro Circuit Court Fail DeBoer v. Snyder MN: o 1971 Pro State High Court Fail Baker v. Nelson o 1997 Anti Legislature Pass Statute 131 MD: Attempts by members of the Maryland Legislature to adopt further anti -gay marriage policies in 1997 never made it out of committee. Also, attempts to bring forward a constitutional amendment via Legislative Referendum in 2004 were also unsuccessful, never making it out of the Maryland House. 238 Table C.1 (cont™d) o 2012 Anti Legislative Referendum Fail Amendment 1, Constitutional Amendment 132 o 2013 Pro Legislature Pass Statute MS: o 1996 Anti Executive Order Pass Banning same -sex marriage in the state o 1997 Anti Legislature Pass Statute o 2004 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendment o 2014 Pro Circuit Court Fail Cam paign for Southern Equality v. Bryant , Circuit Ct did not lift stay before Supreme Ct ruling MO: o 1997 Anti Legislature Pass Statute, overturned by State Supreme Court on procedural grounds o 2001 Anti Legislature Pass Statute o 2004 Anti Citizen Initiat ive Pass Amendment 2, Constitutional Amendment o 2013 Pro Executive Order Pass Recognizing same -sex marriages from other states o 2014 Pro Circuit Court Fail Lawson v. Kelly , District Ct ruled in favor of gay marriage but C. Ct. upheld stay MT: o 1997 Anti Legislature Pass Statute o 2004 Anti Citizen Initiative Pass Initiative 96, Constitutional Amendment o 2012 Pro State High Court Fail Donaldson v. State of Montana o 2014 Pro Circuit Court Pass Rolando v. Fox , District Court ruled in favor of gay marriage, C. Ct. suspend proceed in 2015 NE: o 2000 Anti Citizen Initiative Pass Initiative Measure 416, Constitutional Amendment o 2006 Pro Circuit Court Fail Citizens for Equal Protection v. Bruning o 2014 Pro Circuit Court Fail Waters v. Rick etts , District Court ruled for gay marriage, but C. Ct. stayed order NV: o 2000 Anti Citizen Initiative Pass Constitutional Amendment, State requires 2 votes passing to adopt amendment o 2002 Anti Citizen Initiative Pass Constitutional Amendment, State requires 2 votes passing to adopt amendment Œ Achieved o 2009 Pro Legislature Pass Statute, Domestic Partnerships equivalent to marriage, overrode governor™s veto o 2014 Pro Circuit Court Pass Sevcik v. Sandoval NH: o 1987 Anti Legislature Pass Statute 132 MN: Previous attempts to put a vote to the electorate restricting marriage equality were also made by the legislature in 2004, 20 06, 2007, and 2009, but were ultimately unsuccessfu l in making it out of the legislature. 239 Table C.1 (cont™d) o 2007 Pro Legislature Pass Statute, Civil Unions o 2009 Pro Legislature Pass Statute, Marriage NJ: o 1996 Anti Legislature Fail Statute o 2006 Pro State High Court Pass Lewis v. Harris , ruled that legis. had to address equa lity issue, legislature passed civil unions o 2013 Pro Legislature Pass Statute, Same -Sex Marriage, but vetoed by governor, not enough to override veto o 2013 Pro State High Court Pass Garden State Equality v. Dow , Superior Court ruled for gay marriage, S.S.C would not stay NM: o 2013 Pro State High Court Pass Griego v. Oliver 133 NY: o 2006 Pro State High Court Fail Hernandez v. Robles (among others), C. of Appeals, NY™s High Court, ruled against gay marr. o 2011 Pro Legislature Pass Statute 134 NC: o 1996 Anti Legislature Pass Statute o 2012 Anti Legislative Referendum Pass Amendment 1, Constitutional Amendment o 2014 Pro Circuit Court Pass General Synod of the United Church of Christ v. Cooper , D. Court ruled, C.C. no stay ND: o 1997 Anti Legislatu re Pass Statute o 2004 Anti Citizen Initiative Pass Measure 1, Constitutional Amendment OH: o 2004 Anti Legislature Pass Statute o 2004 Anti Citizen Initiative Pass State Issue 1, Constitutional Amendment o 2014 Pro Circuit Court Fail Obergefell v. Hodges OK: o 1975 Anti Legislature Pass Statute o 1996 Anti Legislature Pass Statute o 2004 Anti Legislative Referendum Pass Question 711, Constitutional Amendment o 2014 Pro Circuit Court Pass Bishop v. United States 133 NM: Multiple attempts by many members in legislature to restrict or expand gay marriage failed over the years. No evidence that N ew Mexico had a statutory ban on gay marriage. 134 NY: The New York Assembly passed pr o-gay -marriage policies in 2007, 2009, and 2011, but New York Senate did not pass these measures until 2011 . 240 Table C.1 (cont™d) OR: o 2004 Anti Citizen Initiative Pass Ballot Measure 36, Constitutional Amendment o 2005 Pro State High Court Fail Li and Kennedy v. State of Oregon o 2007 Pro Legislature Pass Statute, Domestic Partnership equivalent to Civil Unions o 2014 Pro Circuit Court Pass Geiger v. Kitzhaber , District Court ruled for gay marriage, C.C. refused stay PA: o 1996 Anti Legislature Pass Statute o 2014 Pro Circuit Court Pass Whitewood v. Wolf , District Court ruled for gay marriage, C.C. refused stay RI: o 2011 Pro Legislatu re Pass Statute, Civil Unions o 2012 Pro Executive Order Pass Recognizing out -of-state Same -Sex Marriages o 2013 Pro Legislature Pass Statute, Same -Sex Marriages SC: o 1996 Anti Legislature Pass Statute o 2006 Anti Legislative Referendum Pass Amendm ent 1, Constitutional Amendment o 2014 Pro Circuit Court Pass Bradacs v. Haley , Circuit Court ruling on another case, S.Court refused stay SD: o 1996 Anti Legislature Pass Statute o 2006 Anti Citizen Initiative Pass Amendment C, Constitutional Amendment o 2015 Pro Circuit Court Fail Rosenbrahn v. Daugaard , D. C. ruled for gay marriage, C.C. maintained stay until S.C. ruling TN: o 1996 Anti Legislature Pass Statute o 2006 Anti Legislative Referendum Pass Amendment 1, Constitutio nal Amendment TX: 135 o 1973 Anti Legislature Pass Statute, House Bill 103 amending Family Code to limit marriage to one -man -one-woman o 1997 Anti Legislature Pass Statute, not allowed to issue license to same -sex couples o 2003 Anti Legislature Pass Stat ute, void any Texas same -sex marriage or civil union o 2005 Anti Legislative Referendum Pass Texas Proposition 2, Constitutional Amendment o 2015 Pro Circuit Court Fail De Leon v. Perry , D.C ruled for gay marriage, CC. did not decide until S.C. ruling 135 TX: State Supreme Court ruled in 2015 via Texas v. Naylor that same -sex couple married in other state could get divorced. 241 Table C.1 (cont™d) UT: o 1977 Anti Legislature Pass Statute o 1995 Anti Legislature Pass Statute o 2004 Anti Legislature Pass Statute o 2004 Anti Legislative Referendum Pass Amendment 3, Constitutional Amendment o 2013 Pro Circuit Court Pass Kitchen v. Herbert , District Ct. and Circuit Ct. ruled for gay marriage, S Ct did not hear case VT: o 1999 Pro State High Court Pass Baker v. Vermont , following ruling legislature implemented Civil Unions in 2000 o 2009 Pro Legislature Pass Statute VA: o 1975 Anti Legislature Pass Statute, no same -sex marriage o 1997 Anti Legislature Pass Statute, will not recognize out -of-state same -sex marriages o 2004 Anti Legislature Pass Statute, against civil unions o 2006 Anti Legislative Referendum Pass Mars hall -Newman Amendment, Constitutional Amendment o 2014 Pro Circuit Court Pass Bostic v. Schaefer , D.C. and CC. ruled for gay marriage, SC. did not hear case WA: o 1974 Pro State High Court Fail Singer v. Hara , Court of Appeals ruled, State S.C. did not take case o 1997 Anti Legislature Pass Statute, governor vetoed bill o 1998 Anti Legislature Pass Statute, governor vetoed bill, legislators over road veto o 2006 Pro State High Court Fail Andersen v. King County o 2012 Pro Legislature Pass Statute o 2012 Anti Popular Referendum Fail Referendum 74 WV: o 2000 Anti Legislature Pass Statute o 2014 Pro Circuit Court Pass Bostic v. Schaefer , a VA case, but WV complied WI:136 o 1979 Anti Legislature Pass Statute, amending the Family Code (§765.0 01) limiting marriage to husband and wife o 2003 Anti Legislature Pass Statute, governor vetoed the legislation, not enough support to override 136 WI: The Wisconsin House attempted to pass anti -gay -marriage legislation in 1997, but the Senate did not take action. Also, the Wisconsin legislature did pass limited domestic partnership benefits to gay couples in 2009, but the statute did not encompass m arriage. 242 Table C.1 (cont™d) o 2006 Anti Legislative Referendum Pass Constitutional Amendment o 2014 Pro Circuit Court Pass Wolf v. Walker , District and Circuit Cts ruled for gay marriage, Supreme. Ct. did not take case WY: o 1977 Anti Legislature Pass Statute o 2003 Anti Legislature Pass Statute o 2014 Pro Circuit Court Pass Guzzo v. Mead , District Ct. ruling for gay m arriage, Circuit Ct. did not stay NOTES : The pursuit of anti - or pro -gay -marriage policies via any venue prior to 1993 are not included in the analyses (although these instances are listed above for informational purposes only), since the watershed moment for both the anti - and pro -gay -marriage m ovements was the Baehr v. Miike case in Hawaii in 1993. Only domestic partnerships or civil unions that extend marriage benefits to same -sex couples are considered equivalent to pro -same -sex marriage policies in the analyses. Court cases, legislation, or executive orders extending rights of divorce to same -sex couples are also not considered in the analyses since those policies did not affirm a right to a union for gay couples. Only those court cases appealed to and taken up by a state™s highest court are included in the analyses. Votes by the legislature for a constitutional convention, with the possible intent of being able to vote on gay marriage poli cies, are not considered in the analyses since constitutional conventions open the door for various amen dments. Most states only allow constitutional conventions once every ten years. Although several court cases and policies over the last two decades have dealt with myriad gay rights issues, the cases and l egislation listed above explicitly pertain to same -sex unions and marriage equality. SOURCES : Freedom to Marry. 2015. fiHistory and Timeline of the Freedom to Marry in the United States.fl June 26, 2015. Accessed January 25, 2016: http://www.freedomtomarry.org/pages/history -and -timeline -of-marriage ., National Gay and Lesbian Task Force. 2013. fiState Laws Prohibiting Recognition of Same -Sex Relationships.fl Accessed January 25, 2016: http://www.thetaskforce.org/static_html/downloads/reports/issue_maps/samesex_relationships_7_09.pdf . Stewart, Chuck, ed. 2015. Proud Heritage: People, Issues, and Documents of the LGBT Experience . Volume 3. Santa Barbara, CA: ABC -CLIO, LLC. Thompson, Mark, ed. 1994. Long Road to Freedom: The Advocate History of the Gay and Lesbian Movement . 1st ed. Stonewall Inn Editions. New York, NY: St. Martin™s Press. 243 Table C.2: Pursuit of Anti -Gay Marriage Policies by Venue, Year , and State, 1993 -2015 VENUE YEAR Œ STATE SUCCESS FAILURE Legislature 1994: HI 1995: ID, MI, UT 1996: AK, AZ, CO, DE, GA, ID, IL, KS, NJ, NC, OK, PA, SC, SD, TN 1997: AR, CA, CO, FL, IN, ME, MN, MS, MO, MT, ND, TX, VA, WA, 1998: AL, IA, KY, WA 1999: LA 2000: CO, WV 2001: MO 2003: TX, WI, WY 2004: MA, OH, UT, VA 2005: MA 2006: MA 2007: MA 1994: HI 1995: ID, MI, UT 1996: AK, AZ, DE, GA, ID, IL, KS, NC, OK, PA, SC, SD, TN 1997: AR, FL, IN, ME, MN, MS, MT, ND, TX, VA 1998: AL, IA, KY, WA 1999: LA 2000: CO, WV 2001: MO 2003: TX, WY 2004: OH, UT, VA 1996: CO, NJ 1997: CA, CO, MO, WA 2003: WI 2004: MA 2005: MA 2006: MA 2007: MA State High Court --- --- --- Legislative Referendum 1998: AK, HI 2004: GA, KY, LA, MS, OK, UT 2005: KS, TX 2006: AL, ID, SC, TN, VA, WI 2008: AZ 2012: MN, NC 1998: AK, HI 2004: GA, KY, LA, MS, OK, UT 2005: KS, TX 2006: AL, ID, SC, TN, VA, WI 2008: AZ 2012: NC 2012: MN Citizen Initiative 2000: CA, NE, NV 2002: NV 2004: AR, MI, MO, MT, ND, OH, OR 2006: AZ, CO, SD 2008: CA, FL 2009: ME 2012: WA 2000: CA, NE, NV 2002: NV 2004: AR, MI, MO, MT, ND, OH, OR 2006: CO, SD 2008: CA, FL 2009: ME 2006: AZ 2012: WA Executive Order 1996: AL, MS 1996: AL, MS Federal Courts --- --- --- *Success defined as successful enactment (not just pass) of anti -gay -marriage policy. Failure defined as failure to adopt anti -gay -marriage policy (e.g., legislature passes law, but governor vetoes legislation). 244 Table C.3: Pursuit of Pro -Gay Marriage Policies by Venue , Year , and State, 1993 -2015 VENUE YEAR Œ STATE SUCCESS FAILURE Legislature 2005: CA, CT 2007: NH, OR 2009: ME, NV, NH, VT 2010: HI 2011: DE, HI, IL, NY, RI 2012: WA 2013: CO, DE, HI, IL, MN, NJ, RI 2005: CT 2007: NH, OR 2009: ME, NV, NH, VT 2011: DE, HI, IL, NY, RI 2012: WA 2013: CO, DE, HI, IL, MN, RI 2005: CA 2010: HI 2013: NJ State High Court 1993: HI 1998: AK 1999: HI, VT 2004: MA 2005: OR 2006: GA, NJ, NY, WA 2007: MD 2008: CA, CT 2009: CA, IA 2012: MT 2013: NJ, NM 2014: KS 2015: LA 1993: HI 1999: VT 2004: MA 2006: NJ 2008: CA, CT 2009: IA 2013: NJ, NM 2014: KS 1998: AK 1999: HI 2005: OR 2006: GA, NY, WA 2007: MD 2009: CA 2012: MT 2015: LA Legislative Referendum 2012: MD 2012: MD Citizen Initiative 2012: ME 2012: ME Executive Order 2012: RI 2013: MO 2012: RI 2013: MO Federal Courts 2006: NE 2012: CA 2013: UT 2014: AK, AZ, CO, ID, IN, KS, KY, MI, MS, MO, MT, NE, NV, NC, OH, OR, PA, SC, VA, WV, WI, WY 2015: AL, FL, LA, SD, TX 2012: CA 2013: UT 2014: AK, AZ, CO, ID, IN, KS, MT, NV, NC, OR, PA, SC, VA, WV, WI, WY 2015: AL, FL 2006: NE 2014: KY, MI, MS, MO, NE, OH 2015: LA, SD, TX *Success defined as successful enactment (not just pass) of pro -gay -marriage policy. The one exception is for the 1993 Baehr v. Lewin case in Hawaii; although that case did not result in the successful enactment of same -sex marriage, the partial success le d to precipitation of policy activity across the states. Failure defined as failure to adopt pro -gay -marriage policy (e.g., District Court rules in favor of gay -marriage, but Circuit Court issues stay that never allows for implementation before Supreme Cou rt rules). 245 Figure C.1: Probability of Adopting Anti -Gay Marriage Policy by Key Explanatory Variables 246 Figure C.2: Prob. of Adopting Anti -GM Policy by Venue as Political Learning Increases 247 Figure C.3: Prob. of Adopting Pro -GM Policy by Venue as Political Learning Increases 248 Table C. 4: Anti -Gay Marriage Models™ Var iable Descriptions, Descriptive Statistics, and Sources Variable Name Description Mean Sd. Dev. Min Max Sources Political Learning Proportion of states successful in their pursuit of anti -gay marriage policies via institutional venue 0.74 0.36 0 1 Author Policy Learning Cumulative number of states adopt ing anti -gay marriage policy by year 32.35 13.33 0 41 Author Policy Learning from Leg Cumulative number of states that adopted anti -gay marriage policy via legislature by start of year 29.39 11.65 0 36 Author Policy Learning from Leg Ref Cumulative number of states that adopted anti -gay marriage policy via legislative referendum by year 8.78 7.74 0 18 Author Policy Learning from Cit Init Cumulative number of states that adopted anti -gay marriage policy via citizen initiative by start of year 7.17 6.16 0 14 Author Geographic Neighbor Proportion of geographically contiguous neighbors that adopted anti -same-sex marriage policy by start of year 0.64 0.40 0 1 Author, figeographically contiguousfl defined by Berry & Berry 1990 Federal Gov . DOMA Dummy =1 for every year post passage of 1996 D efense of Marriage Act 0.83 0.38 0 1 Author Lawrence v. Texas Sup Ct. Decision Dummy =1 for every year post U.S. Supreme Court™s 2003 Lawrence v. Texas decision that invalidated states™ sodomy laws 0.57 0.50 0 1 Author NYT Issue Salience Cumulative number of times New York Times ran a story on gay marriage during the year, as an indicator of national salience 35.35 34.63 0 153 New York Times Index Pres. Election Year Dummy=1 if presidential election in that calendar year, 0=none 0.22 0.41 0 1 Author Pro -GM Counter Cumulat . number pro -gay marriage policies enacted regardless of venue 9.09 13.32 0 46 Author Legislative Professionalism 1st dimension of state legislative professionalism, from multidimensional scaling of legislators™ salaries, legislative expenditures, and session lengths. Higher values indicate a more professionalized legislature 0.12 1.60 -1.85 8.58 Bowen & Greene 2014 State Supreme Court Professionalism Professionalism score s based on judicial salaries, number of staff, and degree of docket control ; higher scores indicate greater capacity 0.58 0.15 0.25 1.00 Squire 2008 Difficulty Amending Constitution Degree of difficulty in amending state constitution, where 1=only signature requir ements for ballot measure ; 2=simple leg. majority; 3=passage during multiple legislative sessions or voter supermajority; and 4=both multiple sessions and voter supermajority 2.04 0.82 1 4 Lupia et al. 2010 State Gov. Party Control Party control of state government; 0=unified Republican control, 0.5=bipartisan control, 1=unified Democratic control 0.47 0.35 0 1 Klarner 2013a 249 Table C.4 (cont™d) State Supreme Court Ideology Aggregate state -year measure based upon individual state supreme court justice scores ; more positive scores indicate mo re conservative -0.09 0.48 -1.18 1.04 Bonica & Woodruff 2015 Public Support for Gay Marriage Public support for same -sex marriage, estimated from MRP analysis relying on state and national polls 34.90 10.47 10.29 67.21 Lewis & Jacobsmeier 2017 Evangelical Pop . Pct. of population that is Evangelical Christian or Latter -day Saints 28.92 11.76 10 62 Taylor et al. 2019 LGBT Population Percentage of population that identifies LGBT 2.32 0.95 0.675 6.44 Taylor et al. 2019 Prior Anti -GM Policy Running tally of gay marriage bans passed by state in other venues 0.73 0.83 0 3 Author Sodomy Ban State has adopted sodomy ban prohibiting gay sex 0.32 0.47 0 1 Caughey & Warshaw 2015 LGBT Hate Crime Law State has adopted hate crime law increasing penalties for crimes committed on the basis of LGBT identity 0.39 0.49 0 1 Movement Advancement Project. 2019 Rac/Eth Minority Pop. Percentage of state residents that identify as racial/ethnic minority 24.43 14.96 2 78 Kelly & Witko 2014 Citizen Education Percent of state™s population 25 and older with bachelor™s degree 25.90 5.17 12.2 41.4 U.S. Census Bureau State Population (Ln) Natural log of state population (in the thousands) 8.19 1.01 6.16 10.57 U.S. Census Bureau Dir. Democracy State Dummy=1 if state allows direct or indirect citizen initiatives 0.48 0.50 0 1 NCSL 250 Table C. 5: Pro -Gay Marriage Models™ Var . Descriptions, Descriptive Statistics, and Sources Variable Name Description Mean Sd. Dev. Min Max Sources Political Learning Proportion of states successful in their pursuit of pro -gay marriage policies via institutional venue 0.36 0.36 0 1 Author Policy Learning Cumulative number of states adopt ing pro -gay marriage policy by year 9.09 13.32 0 46 Author Geographic Neighbor Proportion of geographically contiguous neighbors that adopted pro -same-sex marriage policy by start of year 0.15 0.29 0 1 Author, figeographically contiguousfl defined by Berry & Berry 1990 U.S. v. Windsor Sup Ct. Decision Dummy =1 for every year post U.S. Supreme Court™s 20 13 U.S. v. Windsor decision that invalidated federal DOMA 0.13 0.34 0 1 Author NYT Issue Salience Cumulative number of times New York Times ran a story on gay marriage during the year, as an indicator of national salience 35.35 34.63 0 153 New York Times Index Legislative Professionalism 1st dimension of state legislative professionalism, from multidimensional scaling of legislators™ salaries, legislative expenditures, and session lengths. Higher values indicate a more professionalized legislature 0.12 1.60 -1.85 8.58 Bowen & Greene 2014 State Supreme Court Professionalism Professionalism score s based on judicial salaries, number of staff, and degree of docket control ; higher scores indicate great er capacity 0.58 0.15 0.25 1.00 Squire 2008 State Gov. Party Control Party control of state government; 0=unified Republican control, 0.5=bipartisan control, 1=unified Democratic control 0.47 0.35 0 1 Klarner 2013a State Supreme Court Ideology Aggregate state -year measure based upon individual state supreme court justice scores ; more positive scores indicate mo re conservative -0.09 0.48 -1.18 1.04 Bonica & Woodruff 2015 District Court Ideology Aggregate state -year measure based upon individual federal district court judges' ideology at the state level ; more positive scores indicate more conservative judges. -0.20 0.64 -1.49 0.67 Bonica et al. 2017 Public Support for Gay Marriage Public support for same -sex marriage, estimated from MRP analysis relying on state and national polls 34.90 10.47 10.29 67.21 Lewis & Jacobsmeier 2017 Evangelical Population Percentage of population that identifies as Evangelical Christian or member of Latter -day Saints 28.92 11.76 10 62 Taylor et al. 2019 LGBT Population Percentage of population that identifies LGBT 2.32 0.95 0.675 6.44 Taylor et al. 2019 Prior Anti -GM Policy Running tally of number of gay marriage bans passed by state by year 0.98 0.87 0 3 Author Prior Pro -GM Policy Running tally of number of pro -gay marriage policies passed in state in other venues by year 0.12 0.39 0 2 Author Sodomy Ban State has adopted sodomy ban prohibiting gay sex 0.32 0.47 0 1 Caughey & Warshaw 2015 LGBT Hate Crime Law State has adopted hate crime law increasing penalties for crimes committed on the basis of LGBT identity 0.39 0.49 0 1 Movement Advancement Project. 2019 251 Table C.5 (cont™d) Rac/Eth Minority Pop. Percentage of state residents that identify as racial/ethnic minority 24.43 14.96 2 78 Kelly & Witko 2014 Citizen Education Percent of state™s population 25 and older with bachelor™s degree 25.90 5.17 12.2 41.4 U.S. Census Bureau State Population (Ln) Natural log of state population (in the thousands) 8.19 1.01 6.16 10.57 U.S. Census Bureau 252 Table C. 6: Robustness Check: Policy Diffusion of Anti -Gay Marriage Policies Using CLogLog Explanatory Variables Anti -GM: Cloglog of Model 3 Political Learning [+] 3.508* (0.350) Policy Learning [+] --- Policy Learn from Leg [+] 0.214* (0.078) Policy Learn from Leg Ref [+] 0.271ƒ (0.141) Policy Learn from Cit Init [+] 0.075 (0.270) Geographic Neighbor [+] 0.802 (0.560) Federal Government DOMA [ -/+] -0.874 (1.169) Lawrence v. Texas Sup. Ct. Decision [+] 3.642ƒ (1.962) U.S. v. Windsor Sup. Ct. Decision [+] --- NYT Issue Salience [+] -0.003 (0.010) Presidential Election Year [+] 0.964 (0.664) Pro -Gay Marriage Counter [+] 0.198* (0.083) Legislative Professionalism [+] 0.020 (0.110) State Supreme Court Professionalism [+] 1.071 (1.239) Difficulty Amending Constitution [ -] -0.167 (0.161) State Gov. Party Control [ -] -0.917* (0.384) State Supreme Court Ideology [+] -0.023 (0.045) Public Support for Gay Marriage [ -] -0.023 (0.045) Evangelical Population [+] 0.024 (0.019) LGBT Population [ -] 0.145 (0.373) Sodomy Ban [+] 0.205 (0.324) LGBT Hate Crime Law [ -] 0.367 (0.385) Racial/Ethnic Minority Population [+] 0.013 (0.012) Population with College Degree [ -] -0.019 (0.041) State Population (Ln) [ -] -0.209 (0.226) Constant -1.965 (2.210) N 2451 2 (22) / Log Likelihood 365.19 * / -246.27 AIC 542.54 . variable is likelihood of adopting anti -gay marriage policy (irrespective of venue). Statistically significant complementary log -log coefficients are in bold face. Robust standard errors, clustered by state -year, are in parentheses. Model also include s a time variable to account for temporal dependence; coefficient omitted for space considerations. The hypothesized direction of the indep endent variable is in brackets. AIC=Akaike information criterion 253 Table C. 7: Robustness Check: Policy Diffusion of Pro -Gay Marriage Policies Using CLogLog Explanatory Variables Pro -GM: Cloglog of Multinomial Logistic Model Political Learning [+] 2.956* (0.503) Policy Learn [+] 0.073ƒ (0.042) Geographic Neighbor [+] 0.797 (0.711) U.S. v. Windsor Sup. Ct. Decision [+] 0.582 (0.774) NYT Issue Salience [+] 0.006 (0.010) Prior Anti -GM Policy [ -] 0.210 (0.272) Legislative Professionalism [+] 0.052 (0.262) State Supreme Court Professionalism [+] -0.202 (2.082) State Gov. Party Control [+] 1.011 (0.632) State Supreme Court Ideology [ -] -1.171* (0.499) District Court Ideology [ -] 0.689 ƒ (0.398) Public Support for Gay Marriage [+] 0.068* (0.034) Evangelical Population [ -] 0.006 (0.030) LGBT Population [+] 0.129 (0.293) Prior Pro -GM Policy [ -] -2.545* (0.520) Sodomy Ban [ -] 0.239 (0.581) LGBT Hate Crime Law [+] 0.925* (0.468) Racial/Ethnic Minority Population [ -] -0.002 (0.012) Population with College Degree [+] 0.059 (0.055) State Population (Ln) [+] -0.233 (0.296) Constant -10.053* (2.546) N 3253 2 (21) / Log Likelihood 201.46* / -158.05 AIC / aROC 360.11 pro -gay marriage policy (irrespective of venue). Statistically significant complementary log -log regression coefficients are in bold face. Robust standard errors, clustered by s tate -year, are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesi zed direction of the independent variable effect is in brackets. AIC = Ak aike information criterion. 254 Table C. 8: Robustness Check: Anti -GM Policies using Cox -Proportional -Hazards Model Explanatory Variables Cox -Proportional -Hazard Ratios Political Learning [+] 2.67 x 10 14* (2.30 x 10 15) Geographic Neighbor [+] 0.403 ƒ (0.215) Presidential Election Year [+] 1.987* (0.554) State Supreme Court Professionalism [+] 11.737* (9.926) Difficulty Amending Constitution [ -] 0.859 (0.167) Direct Democracy [ -/+] 2.051* (0.719) State Gov. Party Control [ -] 0.738 (0.283) Public Support for Gay Marriage [ -] 0.873* (0.026) Evangelical Population [+] 0.980 (0.015) LGBT Population [ -] 0.372* (0.122) Prior Anti -GM Policy [ -] 0.388* (0.104) N 3,234 2 (11) : 242.40 * ƒ 5, two tailed. Model is Cox-proportional hazards model with venues (state legislature, legislative referendum, citizen initiative) as the strata, with dependent variable as the hazard ratio for adopting anti -gay marriage policy. Statistically significant el are in bold face. The hypothesized direction of the independent variable effect is in brackets. Table C. 9: Robustness Check: Pro -GM Policies using Cox -Proportional -Hazards Model Explanatory Variables Cox -Proportional -Hazard Ratios Political Learning [+] 2724.707* (5831.775) Policy Learn [+] 0.781 * (0.0 41) Geographic Neighbor [+] 0.912 (0.480) Legislative Professionalism [+] 1.089 (0.08 3) State Supreme Court Ideology [ -] 0.648 (0.1 86) Evangelical Population [+] 0.986 (0.015) LGBT Population [ -] 1.082 (0.238) N 3,164 2 (7) : 73.39 * ƒ 5, two tailed. Model is Cox -proportional hazards model with venues (state legislature, state court, federal court) as the strata, with dependent variable as the hazard ratio for adopting pro -gay marriage policy. Statistically significant are in bold fa ce. The hypothesized direction of the independent variable effect is in brackets. 255 Table C. 10: Policy Diffusion of Anti -GM Policies using Mult. Log. Reg . Clustered by State Explanatory Variables Legislature Leg. Referendum Citizen Initiative Political Learning [+] 3.717* (0.373) 52.688* (13.172) -1.894 (5.089) Policy Learn [+] 0.191 (0.173) -1.982* (0.797) 2.848* (1.126) Geographic Neighbor [+] 0.598 (0.773) 2.830ƒ (1.478) -13.046* (3.205) Federal Government DOMA [ -/+] 0.487 (2.559) 59.476* (17.360) -15.601 (15.841) Lawrence v. Texas Sup. Ct. Decision [+] 2.409 (3.381) -1.765 (2.708) -1.922 (2.800) NYT Issue Salience [+] 0.023 (0.023) 0.020 (0.018) -0.002 (0.020) Presidential Election Year [+] 0.343 (1.637) 0.383 (1.357) 2.777ƒ (1.555) Pro -Gay Marriage Counter [+] -1.307 (1.443) -1.049 (0.744) 0.080 (0.120) Legislative Professionalism [+] -0.163 (0.202) 0.247 (0.342) -0.408 (0.574) State Supreme Court Professionalism [+] 4.123* (1.892) 3.943 (4.008) -13.074* (5.441) Difficulty Amending Constitution [ -] 0.171 (0.229) 0.086 (0.477) -26.223* (5.383) Direct Democracy [ -/+] 0.810ƒ (0.492) -0.563 (0.874) 17.407* (3.525) State Gov. Party Control [ -] -1.858* (0.731) -0.343 (0.950) 4.963ƒ (2.646) State Supreme Court Ideology [+] 0.037 (0.475) 0.844 (0.808) -6.826* (1.759) Public Support for Gay Marriage [ -] -0.152ƒ (0.080) -0.226ƒ (0.121) -0.125 (0.085) Evangelical Population [+] 0.001 (0.025) 0.050 (0.057) -0.575* (0.130) LGBT Population [ -] -0.507 (0.734) 0.179 (0.646) -5.649* (2.167) Prior Anti -GM Policy [ -] -2.518* (1.067) -0.902ƒ (0.542) -7.614* (2.005) Sodomy Ban [+] 0.573 (0.478) 0.422 (0.670) 1.361 (1.585) LGBT Hate Crime Law [ -] 0.370 (0.737) 0.835 (0.733) -0.519 (1.106) Racial/Ethnic Minority Population [+] 0.012 (0.026) 0.072* (0.033) -0.096 (0.081) Population with College Degree [ -] 0.011 (0.053) 0.158 (0.104) -0.410* (0.146) State Population (Ln) [ -] -0.073 (0.331) -1.172ƒ (0.619) 2.017ƒ (1.189) Constant -3.236 (2.877) -72.357* (14.965) -33.383* (16.419) N 2451 2 (48) : 421.14 * AIC 521.14 Log Likelihood: -210.57 Repeated -events competing risks model estimated using multinomial logit model. Dependent variable is likelihood of adopting anti -gay marriage policy by venue. Dependent variable has four categories, baseline category is not adopting an anti -gay marriage po licy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state (rather than state -year) , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion. 256 Table C.1 1: Policy Diffusion of Pro -GM Policies using Mult. Log. Reg. Clustered by State Explanatory Variables Legislature State Court Federal Court Political Learning [+] 14.643* (4.912) 1.215 (0.854) 0.882 ƒ (0.466) Policy Learn [+] -0.192* (0.076) -0.029 (0.091) 0.337* (0.103) Geographic Neighbor [+] -1.781 (2.001) -3.735* (1.743) 2.966ƒ (1.521) U.S. v. Windsor Sup. Ct. Decision [+] 3.040* (1.127) 2.812 (2.020) -1.502 (2.389) NYT Issue Salience [+] 0.004 (0.008) 0.006 (0.010) 0.037* (0.012) Prior Anti -GM Policy [ -] -0.028 (0.628) -0.510 (1.478) 0.642 (0.663) Legislative Professionalism [+] 0.015 (0.519) 0.367 (0.383) 0.416 (0.572) State Supreme Court Professionalism [+] 3.473 (5.522) 0.659 (5.280) -5.768 (4.226) State Gov. Party Control [+] 0.512 (1.593) -0.287 (1.804) 1.289 (1.542) State Supreme Court Ideology [ -] -2.217 (1.750) -0.091 (1.044) -1.935 (1.194) District Court Ideology [ -] 0.795 (1.095) -1.327 (0.813) 1.391* (0.690) Public Support for Gay Marriage [+] 0.070 (0.073) 0.000 (0.060) 0.220* (0.057) Evangelical Population [ -] -0.078 (0.070) -0.155 (0.140) 0.128* (0.052) LGBT Population [+] 1.193 (0.744) -0.748 (0.697) -0.533 (0.670) Prior Pro -GM Policy [ -] -3.259 (0.774) -1.503 (1.183) -3.696* (1.065) Sodomy Ban [ -] -14.102* (1.113) 0.689 (1.335) 1.105 (0.978) LGBT Hate Crime Law [+] 16.539* (1.201) 2.018 (1.345) 0.445 (1.115) Racial/Ethnic Minority Population [ -] -0.039 (0.042) -0.009 (0.042) -0.000 (0.030) Population with College Degree [+] 0.086 (0.151) 0.074 (0.151) 0.010 (0.098) State Population (Ln) [+] -0.448 (0.793) -0.718 (0.666) 0.584 (0.582) Constant -38.673* (6.839) -3.215 (5.160) -21.486* (5.510) N 3253 2 (48) : 278.12 * AIC 380.12 Log Likelihood : -139.06 . Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of adopting pro -gay marriage polic y by venue. Dependent variable has four categories, baseline category is not adopting a pro -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state (rather than state -year) , ar e in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike informat ion criterion . 257 APPENDIX D Table D.1 : Anti -Gay Marriage Models™ Var . Descriptions, Descriptive Statistics, and Sources Variable Name Description Mean Sd. Dev. Min Max Sources Political Learning Prop ortion of states successful in their pursuit of anti -gay marriage policies via institutional venue 0.74 0.36 0 1 Author Similarity in Legislative Professionalism Euclidean distance between state™s legislative professionalism score and average score of those states that pursued anti -gay marriage policy via given venue. Reverse coded so that increase indicates more similar. -1.17 1.21 -8.60 0 Author, using Bowen and Greene™s 2014 measure Similarity in Citizen Ideology Euclidean distance between state™s citizen ideology and average ideology of those states that pursued anti -gay marriage policy via given venue. Reverse coded so that increase indicates more similar. -20.54 16.59 -71.41 0 Author, using Berry et al. ™s 2010 measure Similarity in Difficulty in Amending Constitution Euclidean distance between state™s difficulty in amending constitution and avg. difficulty of those states that pursued anti -gay marriage policy via given venue. Reverse coded so that increase indicates more similar. -1.03 0.82 -4 0 Author, using Lupia et al. ™s 2010 measure Geographic Neighbor by Venue Proportion of geographically contiguous neighbors that adopted anti -same -sex marriage policy by venue by start of year 0.29 0.37 0 1 Author, figeographically contiguousfl defined by Berry & Berry 1990 Policy Learning by Venue Cumulative number of states successfully picking venue to pursue anti -gay marriage policy by year 15.12 13.42 0 36 Author Federal Gov . DOMA Dummy =1 for every year post passage of 1996 Defense of Marriage Act 0.83 0.38 0 1 Author Lawrence v. Texas Sup Ct. Decision Dummy =1 for every year post U.S. Supreme Court™s 2003 Lawrence v. Texas decision that invalidated states™ sodomy laws 0.57 0.50 0 1 Author Pres. Election Year Dummy=1 if presidential election in that calendar year, 0=none 0.22 0.41 0 1 Author Pro -Gay Marriage Counter Cumulative number pro -gay marriage policies enacted regardless of venue 9.09 13.32 0 46 Author 258 Table D.1 (cont™d) Evangelical Population Pct. of population that identifies as Evangelical Christian or member of Latter -day Saints 28.92 11.76 10 62 Taylor et al. 2019 LGBT Population Percentage of population that identifies LGBT 2.32 0.95 0.675 6.44 Taylor et al. 2019 Prior Anti -GM Policy Success Running tally of gay marriage bans passed by the state in other venues 0.76 0.84 0 3 Author State Supreme Court Professionalism Professionalism score s based on judicial salaries, number of staff, and docket control ; higher scores indicate greater capacity 0.58 0.15 0.25 1.00 Squire 2008 State Supreme Court Ideology Aggregate state -year measure based upon individual state supreme court justice scores ; more posi tive scores indicate more conservative -0.09 0.48 -1.18 1.04 Bonica & Woodruff 2015 Dir. Democracy State Dummy=1 if state allows direct or indirect citizen initiatives 0.48 0.50 0 1 NCSL Public Support for Gay Marriage Public support for same -sex marriage, estimated from MRP analysis relying on state and national polls 34.90 10.47 10.29 67.21 Lewis & Jacobsmeier 2017 State Population (Ln) Natural log of state population (in thousands) 8.19 1.01 6.16 10.57 U.S. Census Bureau 259 Table D.2 : Pro -Gay Marriage Models™ Var . Descriptions, Descriptive Statistics, and Sources Variable Name Description Mean Sd. Dev. Min Max Sources Political Learning Proportion of states successful in their pursuit of pro -gay marriage policies via institutional venue 0.36 0.36 0 1 Author Similarity in Legislative Professionalism Euclidean distance between state™s legislative professionalism score and average score of those states that picked same venue to pursue pro -gay marriage policy. Reverse coded so that increase indicates more similar. -1.35 1.30 -8.58 0 Author, using Bowen and Greene™s 2014 measure Similarity in Supreme Court Professionalism Euclidean distance between state™s high court professionalism score and average score of those states that picked same venue to pursue pro -gay marriage policy. Reverse coded so that inc rease indicates more similar. -0.29 0.25 -1.00 0 Author, using Squire™s 2008 measure Similarity in Citizen Ideology Euclidean distance between state™s citizen ideology and average ideology of those states that picked same venue to pursue pro -gay marriage policy. Reverse coded so that increase indicates more similar. -28.48 19.01 -71.84 0 Author, using Berry et al. ™s 20 10 measure Similarity in Supreme Court Ideology Euclidean distance between state™s high court ideology and average ideology of those states that picked same venue to pursue pro -gay marriage policy. Reverse coded so that increase indicates more similar. -0.51 0.38 -1.86 0 Author, using Bonica and Woodruff™s 2015 measure Similarity in District Court Ideology Euclidean distance between state™s district court ideology and average ideology of those states that picked same venue to pursue pro -gay marriage policy. Reverse coded so that increase indicates more similar. -0.59 0.43 -1.67 0 Author, using Bonica et al. ™s 2017 measure Geographic Neighbor by Venue Proportion of geographically contiguous neighbors that picked venue by start of year to adopted pro -same -sex marriage policy 0.05 0.16 0 1 Author, figeographically contiguousfl defined by Berry & Berry 1990 Policy Learning by Venue Cumulative number of states successfully picking venue by year to pursue pro -gay marriage policy 2.84 4.73 0 20 Author 260 Table D.2 (cont™d) Lawrence v. Texas Sup Ct. Decision Dummy =1 for every year post U.S. Supreme Court™s 2003 Lawrence v. Texas decision that invalidated states™ sodomy laws 0.57 0.50 0 1 Author U.S. v. Windsor Sup Ct. Decision Dummy =1 for every year post U.S. Supreme Court™s 2013 U.S. v. Windsor decision that invalidated federal DOMA 0.13 0.34 0 1 Author Pres. Election Year Dummy=1 if presidential election in that calendar year, 0=none 0.22 0.41 0 1 Author Prior Anti -Gay Marriage Policy Running tally of number of gay marriage bans passed by state by year 0.98 0.87 0 3 Author Evangelical Population Percentage of population that identifies as Evangelical Christian or member of Latter -day Saints 28.92 11.76 10 62 Taylor et al. 2019 LGBT Population Percentage of population that identifies LGBT 2.32 0.95 0.675 6.44 Taylor et al. 2019 Prior Pro -GM Policy Success Running tally of pro -gay marriage policies passed by the state in other venues 0.12 0.39 0 2 Author Public Support for Gay Marriage Public support for same -sex marriage, estimated from MRP analysis relying on state and national polls 34.90 10.47 10.29 67.21 Lewis & Jacobsmeier 2017 State Population (Ln) Natural log of state population (in the thousands) 8.19 1.01 6.16 10.57 U.S. Census Bureau 261 Figure D.1: Pred. Prob. of Picking Venue for Anti -GM Policies as Political Learning Increases 262 Figure D.2: Pol . Learning and Time™s Interactive Effect on Venue Choice for Anti -GM Policies 263 Table D.3 : Venue Diffusion of Anti -GM Policies using Logit, Comp. Log -Log, and Ord . Logit Explanatory Variables Logit Comp. Log -Log Ordered Logit Political Learning [+] 1.693* (0. 680) 1.768* (0.641) 1.752* (0.736) Similarity in Legislative Professionalism [+] 0.267* (0. 114) 0.236* (0.102) 0.257* (0.118) Similarity in Difficulty Amending Constitution [+] 1.294* (0.267) 1.145* (0.243) 1.365* (0.262) Similarity in Citizen Ideology [+] 0.06 3* (0.0 17) 0.055* (0.016) 0.062* (0.016) Geographic Neighbor [+] 0.953ƒ (0.572) 0.786 (0.500) 0.943ƒ (0.559) Policy Learn by Venue [+] 0.048 * (0.0 16) 0.046* (0.015) 0.038* (0.016) Federal Government DOMA [+] 0.487 (0.604) 0.519 (0.545) 0.598 (0.575) Lawrence v. Texas Sup. Ct. Decision [+] 1.774 * (0.590 ) 1.704* (0.544) 1.707* (0.581) Presidential Election Year [+] 1.019* (0.274) 0.887* (0.240) 1.085* (0.281) Pro -Gay Marriage Counter [+] 0.027 (0.033) 0.028 (0.032) 0.020 (0.033) Evangelical Population [+] 0.006 (0.019) 0.005 (0.017) 0.002 (0.019) LGBT Population [ -] 0.767* (0.351) 0.659* (0.312) 0.780* (0.340) Prior Anti -GM Policy [ -] -0.732* (0.310) -0.666* (0.230) -0.837* (0.304) State Supreme Court Professionalism [+] 3.194 * (1. 252) 3.014* (1.123) 3.166* (1.211) State Supreme Court Ideology [+] 0.703 * (0. 317) 0.554* (0.277) 0.594ƒ (0.307) Direct Democracy [ -/+] 1.096* (0.362) 1.026* (0.335) 1.166* (0.350) Public Support for Gay Marriage [ -] -0.074* (0.036) -0.071* (0.033) -0.086* (0.037) State Population (Ln) [ -] -0.174 (0.155) -0.169 (0.136) -0.157 (0.144) Constant , Constant, Cuts -1.931 (1.972) -2.093 (1.762) 1.937, 2.98, 3.77 N 2505 2 (19) : 267.91 *, 322.16*, 240.14 * AIC 525.59, 527.88, 702.65 Log Likelihood: -242.79, -243.94 , -329.33 ƒ 5, two tailed. First model is logistic regression model, with dependent variable as the likelihood of picking any venue to pursue anti -gay marriage policy. Second model is a complementary log -log model, with dependent variable as the likelihood of picking any venue to p ursue anti -gay marriage policy. Third model is ordered logistic regression model where dependent variable has four ordered categories: no venue, state legislature, legislative referendum, citizen initiative. Statistically significant coefficients level are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized directi on of the independent variable effect is in brackets. 264 Table D.4 : Venue Diffusion of Anti -GM Policies using Cox -Proportional -Hazards Model Explanatory Variables Cox -Proportional -Hazard Ratios Political Learning [+] 2.86 x 10 24* (2.04 x 10 25) Similarity in Legislative Professionalism [+] 1.042* (0.106) Similarity in Difficulty Amending Constitution [+] 2.630* (0.577) Similarity in Citizen Ideology [+] 1.053* (0.016) Presidential Election Year [+] 2.115* (0.511) Evangelical Population [+] 1.005* (0.010) State Supreme Court Ideology [+] 1.156* (0.268) Direct Democracy [ -/+] 1.762 * (0. 404) N 3,300 2 (8) : 226.39* ƒ 5, two tailed. Model is Cox-proportional hazards model with venues (state legislature, legislative referendum, citizen initiative) as the strata, with dependent variable as the hazard ratio for picking venue to pursue anti -gay marriage policy. Statistically significant hazard rati os are in bold face. The hypothesized direction of the independent variable effect is in brackets. 265 Figure D. 3: Pred. Probability of Picking Venue for Pro -GM Policies as Political Learning Increases 266 Figure D. 4: Pol . Learning and Time™s Interactive Effect on Venue Choice for Pro -GM Policies 267 Table D.5 : Venue Diffusion of Pro -GM Policies using Logit, Comp. Log -Log, and Ord . Logit Explanatory Variables Logit Comp. Log -Log Ordered Logit Political Learning [+] -0.042 (0.779) 0.227 (0.704) -0.750 (0.78 3) Similarity in Legislative Professionalism [+] -0.090 (.146) -0.089 (0.14 6) -0.11 6 (0.14 3) Similarity in Supreme Court Professionalism [+] 0.972 (1.467) 1.623 (1.418 ) 1.48 7 (1.46 8) Similarity in Citizen Ideology [+] 0.081* (0.015) 0.073* (0.013 ) 0.077* (0.015) Similarity in Supreme Court Ideology [+] 1.837* (0.561) 1.640* (0.507) 1.775* (0.571) Similarity in District Court Ideology [+] 0.663 (0.408) 0.52 1 (0.364 ) 0.542 (0.420 ) Geographic Neighbor by Venue [+] 0.308 (0.248) 0.193 (0.478 ) 0.820 (0.570) Policy Learn by Venue [+] 0.235* (0.056) 0.206* (0.048) 0.280* (0.065) Lawrence v. Texas Sup. Ct. Decision [+] 2.170ƒ (1.130) 2.294* (1.098) 2.09 5ƒ (1.13 7) U.S. v. Windsor Sup. Ct. Decision [+] 1.395* (0.558) 1.277* (0.512) 1.258* (0.509) Presidential Election Year [ -] -0.656 (0.526) -0.710 (0.495 ) -0.545 (0.490) Anti -Gay Marriage by State [+] 0.308 (0.549) 0.26 4 (0.22 7) 0.422 ƒ (0.252 ) Evangelical Population [ -] -0.019 (0.027) -0.014 (0.023 ) -0.021 (0.02 8) LGBT Population [+] 0.450 (0.294) 0.301 (0.250) 0.31 3 (0.29 5) Prior Pro -GM Policy [ -] -1.752* (0.372) -1.487* (0.329) -1.611* (0.370) Public Support for Gay Marriage [+] 0.054ƒ (0.030) 0.055* (0.026) 0.04 1 (0.031 ) State Population (Ln) [+] -0.060 (0.157) -0.050 (0.143 ) -0.00 9 (0.15 5) Constant , Constant, Cuts -2.782 (2.208) -3.13 6 (1.998 ) 2.055 , 2.542, 3.206 N 3,322 2 (18) : 235.07 *, 278.80*, 292.14* AIC Log Likelihood: -195.54, -195.53, -271.92 ƒ 5, two tailed. First model is logistic regression model, with dependent variable as the likelihood of picking any venue to pursue pro -gay marriage policy. Second model is a complementary log -log model, with dependent variable as the likelihood of picking any venue to pu rsue pro -gay marriage policy. Third model is ordered logistic regression model where dependent variable has four ordered categories: no venue, state legislature, state court, federal court. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state -year , are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the ind ependent variable effect is in brackets. 268 Table D.6 : Venue Diffusion of Pro -GM Policies using Cox -Proportional -Hazards Model Explanatory Variables Cox -Proportional -Hazard Ratios Political Learning [+] 0.106 (0.275) Similarity in Supreme Court Ideology [+] 8.779* (4.546) Similarity in Citizen Ideology [+] 1.064* (0.016) Prior Pro -GM Policy [ -] 0.213* (0.070) Evangelical Population [ -] 0.949 * (0.01 2) LGBT Population [+] 0.802 (0.162) N 3,173 2 (6) : 118.54 * ƒ 5, two tailed. Model is Cox -proportional hazards model with venues (state legislature, state court, federal court) as the strata, with dependent variable as the hazard ratio for picking venue to pursue pro -gay marriage policy. Statistically significant hazard ratios are in bold face. The hypothesized direction of the independent variable effect is in brackets. 269 Table D.7 : Venue Diffusion of Anti -GM Policies using Mult. Logistic Reg . Clustered by State Explanatory Variables Legislature Leg. Referendum Citizen Initiative Political Learning [+] 2.03 8ƒ (1.19 5) 46.327ƒ (24.620) -3.170ƒ (1.885) Similarity in Legislative Professionalism [+] 0.364* (0.148) 2.306* (0.840) 0.267 (0.202) Similarity in Difficulty Amending Constitution [+] 0.66 9ƒ (0.37 5) 1.156ƒ (0.608) 4.347* (1.433) Similarity in Citizen Ideology [+] 0.064* (0.030) 0.101* (0.040) 0.033 (0.033) Geographic Neighbor by Venue [+] .239 (0.813 ) 1.071 (2.267) 2.102ƒ (1.173) Policy Learn by Venue [+] 0.227* (0.039) 0.162 (0124) -0.237ƒ (0.142) Federal Government DOMA [+] -0.35 3 (1.309 ) 14.950* (2.976) 12.926* (0.759) Lawrence v. Texas Sup. Ct. Decision [+] 5.000* (1.365) 0.508 (0.901) -1.044 (1.261) Presidential Election Year [+] .73 9 (.63 2) 0.832 (0.601) 2.226* (0.699) Pro -Gay Marriage Counter [+] 0.198* (0.098) 0.007 (0.104) -0.252* (0.102) Evangelical Population [+] 0.01 3 (0.026 ) 0.086 (0.057) -0.154* (0.058) LGBT Population [ -] .817 (1.06 3) 0.758 (0.617) -1.310* (0.648) Prior Anti -GM Policy [ -] -0.72 4 (0.86 4) 0.136 (0.769) -1.650* (0.626) State Supreme Court Professionalism [+] 3.854* (1.673) 2.130 (3.447) 1.187 (3.661) State Supreme Court Ideology [+] 1.045* (0.522) 1.186 (0.763) 0.033 (0.033) Direct Democracy [ -/+] 1.202* (.564) -1.262 (0.861) 19.415* (1.420) Public Support for Gay Marriage [ -] -0.06 4 (0.073 ) -0.018 (0.082) -0.110* (0.039) State Population (Ln) [ -] -0.21 3 (0.25 4) -0.403 (0.648) -0.298 (0.493) Constant -1.90 4 (2.870 ) -61.773* (24.493) -27.959* (2.899) N 2505 2 (46) : 437.04 * AIC 535.04 Log Likelihood: -218.52 ƒ 5, two tailed. Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of picking a venue to pursue anti -gay marriage polic y. Dependent variable has four categories, baseline category is not picking a venue to pursu e an anti -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state, are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitted from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. 270 Table D.8 : Venue Diffusion of Pro -GM Policies using Mult. Logistic Reg . Clustered by State Explanatory Variables Legislature State Court Federal Court Political Learning [+] 15.157* (4.498) -1.987* (0.649) -4.430* (1.782) Similarity in Legislative Professionalism [+] 0.045 (0.378) -0.274ƒ (0.156) 0.019 (0.191) Similarity in Supreme Court Professionalism [+] -3.908 (4.611) 4.487* (1.850) 0.519 (3.140) Similarity in Citizen Ideology [+] 0.160* (0.046) 0.078* (0.030) 0.066* (0.022) Similarity in Supreme Court Ideology [+] 2.831ƒ (1.571) 1.330 (0.911) 2.045* (0.941) Similarity in District Court Ideology [+] 0.337 (1.044) 0.303 (0.390) 0.558 (0584) Geographic Neighbor by Venue [+] -2.744* (1.199) -2.187 (1.358) 1.220 (0.917) Policy Learn by Venue [+] 0.324ƒ (0.167) 0.125ƒ (0.073) 0.369* (0.139) Lawrence v. Texas Sup. Ct. Decision [+] 27.666* (4.524) 1.318 (1.108) 16.696* (3.240) U.S. v. Windsor Sup. Ct. Decision [+] 1.535ƒ (0.873) 1.972 (1.165) 4.473* (1.743) Presidential Election Year [ -] -2.504ƒ (1.354) 0.299 (0.530) 0.918 (0.963) Anti -Gay Marriage by State [+] -0.313 (0.520) 0.040 (0.425) 0.858* (0.314) Evangelical Population [ -] -0.071 (0.055) -0.040 (0.033) 0.037 (0.050) LGBT Population [+] 1.601* (0.638) 0.616 (0.452) -0.142 (0.541) Prior Pro -GM Policy [ -] -1.797* (0.906) -1.446* (0.737) -1.493* (0.469) Public Support for Gay Marriage [+] 0.108* (0.045) 0.029 (0.045) 0.102* (0.050) State Population (Ln) [+] -0.007 (0.353) 0.023 (0.220) 0.027 (0.267) Constant -32.771* (5.383) -1.275 (2.537) -19.406* (6.785) N 3322 2 (49) : 405.84 * AIC 509.84 Log Likelihood: -202.92 ƒ 5, two tailed. Repeated -events competing -risks model estimated using multinomial logit model. Dependent variable is likelihood of picking a venue to pursue pro -gay marriage polic y. Dependent variable has four categories, baseline category is not picking a venue to pursue a pro -gay marriage policy. Statistically significant coefficients are in bold face. Robust standard errors, clustered by state, are in parentheses. Models also include a time variable to account for temporal dependence; coefficient is omitt ed from the table due to space considerations. The hypothesized direction of the independent variable effect is in brackets. AIC = Akaike information criterion and aROC = Area under the ROC curve. 271 REFERENCES 272 REFERENCES Allen, M . D., C . Pettus, and D . P. Haider -Markel. 2004. fiMaking the National Local: Specifying the Conditions for National Government Influence on State Policymaking.fl State Politics & Policy Quarterly 4 (3): 318 Œ44. Allison, Paul. 2012. fiHow Relevant Is the Independence of Irrelevant Alternatives?fl Statistical Horizons. October 8, 2012. Accessed on February 26, 2019 at: https://statisticalhorizons.com/iia . Andersen, Ellen Ann. 2005. Out of the Close ts & into the Courts: Legal Opportunity Structure and Gay Rights Litigation . Ann Arbor: University of Michigan Press. Austen -Smith, David and John R. Wright. 1994. fiCounteractive Lobbying.fl American Journal of Political Science 38 (1): 25 -44. Bailey, M. A., and M. C. Rom. 2004. fiA Wider Race? Interstate Competition across Health and Welfare Programs.fl Journal of Politic s 66 (2) : 326 -47. Balla, Steven J. 2001. fiInterstate Professional Associations and the Diffusion of Policy Innovations.fl American Politics Research 29 (3): 221 Œ45. Ballotpedia. 2010. fiCalifornia Proposition 19, the Marijuana Legalization Initiative (201 0).fl Accessed on July 15, 2016 at: https://ballotpedia.org/California_Proposition_19,_the_Marijuana_Legalization_Initiative_( 2010) . Ballotpedia . 2016. fiForms of Direct Democracy in the American States.fl Accessed at August 8, 2016 at https://ballotpedia.org/Forms_of_direct_democracy_in_the_American_states . Banducci, S., 1998, ‚Direct Legislation: When is it used and when does it pass?™ in Bowler, S., Donovan, T. and Tolbert, C. (Eds), Citizens as Legislators: Direct Democracy in the United States . Columbus, OH: Ohio State University Press . Banfield , Edward C. 1961. Political Influence: A New Theory of Urban Politics . New York , NY : The Free Press . Barclay, Scott. 2010. fiIn Search of Judicial Activism in the Same -Sex Marriage Cases: Sorting the Evidence from Courts, Legislatures, Initiatives and Amendm ents.fl Perspectives on Politics 8 (1): 111Œ26. Barclay, Scott and Shauna Fisher. 2003. fiThe States and the Differing Impetus for Divergent Paths on Same -Sex Marriage, 1990 -2001.fl Policy Studies Journal 31 (3): 331 -352. 273 Barclay, Scott and Shauna Fisher. 2008. fiSaid and Unsaid: State Legislative Signaling to State Courts over Same Sex Marriage 1990 -2014.fl Law & Policy 30 (2): 255 -275. Barcott, Bruce. 2015. Weed the People: The Future of Legal Marijuana in America . New York, NY: Time Books. Barrilleaux , Charles, Thomas Holbrook, and Laura Langer. 2002. fiElectoral Competition, Legislative Balance, and American State Welfare Policy.fl American Journal of Political Science 46 (2): 415 Œ27. Baumgartner, Frank R., Jeffrey M. Berry, Marie Hojnacki, Beth L. Lee ch, and David C. Kimball. 2009. Lobbying and Policy Change: Who Wins, Who Loses, and Why . Chicago, IL: University of Chicago Press. Baumgartner, Frank R., and Bryan D. Jones. 1991. fiAgenda Dynamics and Policy Subsystems.fl The Journal of Politics 53 (4): 1044 Œ74. Baumgartner, Frank R., and Bryan D. Jones. 1993. Agendas and Instability in American Politics, First Edition . Chicago, IL: University of Chicago Press. Baybeck, B., W. D. Berry, and D. A. Siegel. 2011. fiA Strategic Theory of Policy Diff usion via Intergovernmental Competition.fl The Journal of Politics . 73 (1) : 232 -247. Beck, N . J. N. Katz, and R . Tucker. 1998. fiTaking Time Seriously: Time -Series -Cross -Section Analysis with a Binary Dependent Variable.fl American Journal of Political Science 42 (4): 1260 Œ88. Bennett, Colin J. 1991. fiWhat Is Policy Convergence and What Causes It?fl British Journal of Political Science 21 (2): 215 -233. Berry, F rances Stokes . 1994. fiInnovation of Public Management ŠThe Adoption of Strategic -Planning.fl Public Administration Review 54 (4): 322Œ330. Berry, William D., and Brady Baybeck. 2005. fiUsing Geographic Information Systems to Study Interstate Competition.fl The American Political Science Review 99 (4): 505 Œ19. Berry, Frances Stokes, and William D. Berry. 1990. fiState Lottery Adoptions as Policy Innovations: An Event History Analysis.fl The American Political Science Review 84 (2): 395 -415. Berry, Frances Stokes, and William D. Berry. 1992. fiTax Innovation in the States: Capitalizing on Political Opportunity.fl American Journal of Political Scienc e 36 (3): 715 Œ42. Berry, F rancis Stokes , and William D. Berry. 2014. fiInnovation and Diffusion Models in Policy Research. fl In Theories of the Policy Process , P. A. Sabatier & C. M. Weible ( eds.), Boulder, CO. Westview. 307Œ360. 274 Berry, William D., Evan J. Ringquist, Richard C. Fording and Russell L. Hanson. 1998. fiMeasuring Citizen and Government Ideology in the American States, 1960 -93.fl American Journal of Political Science 42 (1): 327 -348. Berry, William D., Richard C. Fording , Evan J. Ringquist, Russell L. Hanson, and Carl Klarner. 2010. fiMeasuring Citizen and Government Ideology in the American States: A Re -appraisal.fl State Politics & Policy Quarterly 10 (2): 117 -35. Biggers, Daniel R. 2014. Morality at the Ballot: Direct D emocracy and Political Engagement in the United States . New York, NY: Cambridge University Press. Blossfeld, H ans -Peter , Katrin Golsch, and Götz Rohwer. 2007. Event History Analysis with Stata . Mahwah, NJ: Lawrence Erlbaum Associates. Boehmke, Frederick J. 2002. fiThe Effect of Direct Democracy on the Size and Diversity of State Interest Group Populations.fl The Journal of Politics 64 (3): 827 Œ44. Boehmke, Frederick J. 2005. The Indirect Effect of Direct Legislation: How Institutions Shape Interest Group Systems . Columbus , OH : Ohio State University Press. Boehmke, Frederick J. 2008. fiThe Initiative Process and the Dynamics of State Interest Group Populations.fl State Politics & Policy Quarterly 8 (4 ): 362 Œ83. Boehmke, Frederick J. 2009a. fiApproaches to Modeling the Adoption and Diffusion of Policies with Multiple Components.fl State Politics & Policy Quarterly 9 (2): 229 Œ53. Boehmke, Frederick J. 2009 b. fiPolicy Emulation or Policy Convergence? Poten tial Ambiguities in the Dyadic Event History Approach to State Policy Emulation.fl The Journal of Politics 71 (3): 1125 -1140. Boehmke, Frederick J., Sean Gailmard, and John Wiggs Patty. 2006. fiWhose Ear to Bend? Information Sources and Venue Choice in Poli cy -Making.fl Quarterly Journal of Political Science 1 (2): 139+. Boehmke, Frederick J., Sean Gailmard, and John W iggs Patty. 2013. fiBusiness as Usual: Interest Group Access and Representation across Policy -Making Venues.fl Journal of Public Policy 33 (1): 3 Œ33. Boehmke, Frederick J., Tracy L. Osborn, and Emily U. Schilling. 2015. fiPivotal Politics and Initiative Use in the American States.fl Political Research Quarterly 68 (4): 665 -677. Boehmke, Frederick J., and John W iggs Patty. 2007. fiThe Selection of Policies for Ballot Initiatives: What Voters Can Learn from Legislative Inaction.fl Economics & Politics 19 (1): 97 -121. 275 Boehmke, Frederick J., and Charles R. Shipan. 2015. fiOversight Capabilities in the States Are Professionalized Legislatures Better at Getting What They Want?fl State Politics & Policy Quarterly 15 (3): 366-386 . Boehmke, Frederick J., and Paul Skinner. 2012. fiState Policy Innovativeness Revisited.fl State Politics and Policy Quarterly 12 (3): 303-329. Bonica, Adam, Adam Chilton, Jacob Goldin, Kyle Rozema, and Maya Sen. 2017. fiMeasuring Judicial Ideology Using Law Clerk Hiring.fl American Law and Economics Review 19 (1): 129 -161. Bonica, Adam, and Michael J. Woodruff. 2015. fiA Common -Space Measure of State Supreme Court Ideology.fl The Journal of Law, Economics, and Organization 31 (3): 472 Œ98. Boushey, Graeme. 2010. Policy Diffusion Dynamics in America . New York, NY: Cambridge University Press. Boushey, Graeme T. 2012. fiPunctuated Equilibrium Theory and the Diffusion of Innovations.fl Policy Studies Journal 40 (1): 127 Œ46. Boushey, Graeme. 2016. fiTargeted for Diffusion? How the Use and Acceptance of Stereotypes Shape the Diffusion of Criminal Justice Policy Innovations in the American States.fl American Political Science Review 110 (01): 198 Œ214. Boushey, Graeme T., and Mis ti Lynn Knight -Finley. 2016. fiPolicy Winnowing and the Diffusion of Innovations.fl Presented at 2016 Midwest Political Science Association, April 2016. Working Paper. Bowen, Daniel C., and Zachary Greene. 2014. fiShould We Measure Professionalism with an In dex? A Note on Theory and Practice in State Legislative Professionalism Research.fl State Politics & Policy Quarterly 14 (3): 277 Œ96. Bowler, Shaun and Todd Donovan. 2004. fiMeasuring the Effect of Direct Democracy on State Policy: Not All Initiatives are C reated Equal.fl State Politics & Policy Quarterly 4 (3): 345 -363. Bowler, Shaun and Todd Donovan. 2006. fiDirect Democracy and Political Parties in America.fl Party Politics 12 (5): 649 -669. Bowler, S haun and Amihai Glazer. 2008. Direct Democracy™s Impact o n American Political Institutions . New York, NY: Palgrave Macmillan . Bowman, Ann O™M., Neal D. Woods, and Milton R. Stark. 2010. fiGovernors Turn Pro: Separation of Powers and the Institutionalization of the American Governorship.fl Political Research Quart erly 63 (2): 304 Œ15. 276 Box -Steffensmeier, Janet M., and Bradford S. Jones. 2004. Event History Modeling: A Guide for Social Scientists . New York, NY: Cambridge University Press. Box -Steffensmeier, Janet M., and Christopher Zorn. 2002. fiDuration Models for Repeated Events.fl The Journal of Politics 64 (4): 1069 -94. Brace, P aul , and M elinda Gann Hall. 2001. fi‚Haves™ Versus ‚Have Nots™ in State Supreme Courts: Allocating Docket Space and Wins in Power Asymmetric Cases.fl Law & Society Review 35 (2): 393Œ417. Brandeis, L. D. 1932. Dissenting Opinion. New State Ice Co. v. Liebmann , 285 U.S. 262, 311. Braun, Dietmar, and Fabrizio Gilardi. 2006. fiTaking ‚Galton™s Problem™ Seriously: Towards a Theory of Policy Diffusion.fl Journal of Theoretical Politics 18 (3): 298 Œ322. Buckley, Jack, and Chad Westerland. 2004. fiDuration Dependence, Functional Form, and Corrected Standard E rrors: Improving EHA Models of State Policy Diffusion.fl State Politics & Policy Quarterly 4 (1): 94 -113. Buffardi, Anne L., Robert J. Pekkanen, and Steven Rathgeb Smith. 2015. fiShopping or Specialization? Venue Targeting among Nonprofits Engaged in Advoca cy.fl Policy Studies Journal 43 (2): 188 Œ206. Butler, Daniel M., and Miguel M. Pereira. 2018. fiTRENDS: How Does Partisanship Influence Policy Diffusion?fl Political Research Quarterly 71 (4): 801 -812. Butler, Daniel M., Craig Volden, Adam M. Dynes, and Boris Shor. 2015. fiIdeology, Learning, and Policy Diffusion: Experimental Evidence.fl American Journal of Political Science 61 (1): 37-49. Button, James W., Barbara A. Rienzo, and Kenneth D. Wald. 1997. Private Lives, Public Conflicts: Battles over Gay Rights in American Communities . Washington, DC: Congressional Quarterly Inc. Caldeira, Gregory A. 1985. fiThe Transmission of Legal Precedent: A Study of State Supreme Courts.fl American Political Science Rev iew 79 (1): 178 -193. Calvert, Randall L., Mathew D. McCubbins, and Barry R. Weingast. 1989. fiA Theory of Political Control and Agency Discretion.fl American Journal of Political Science 33 (3): 588 Œ611. Cameron, A Colin and Douglas L. Miller. 2015. fiA Pra ctitioner's Guide to Cluster -Robust Inference.fl The Journal of Human Resources 50 (2): 317 -372 Camp, Bayliss J. 2008. fiMobilizing the Base and Embarrassing the Opposition: Defense of Marriage Referenda and Cross -Cutting Electoral Cleavages.fl Sociological Perspectives 51 (4): 713 -733. Campbell, Andrea Louise. 2012. fiPolicy Makes Mass Politics.fl Ann ual Review of Political Science 15 (1): 333Œ51. 277 Cann, Damon, and Teena Wilhelm. 2011. fiPolicy Venues and Policy Change: The Case of Education Finance Reform.fl Social Science Quarterly 92 (4): 1074 Œ95. Canon, Bradley C., and Lawrence Baum. 1981. fiPattern s of Adoption of Tort Law Innovations: An Application of Diffusion Theory to Judicial Doctrines.fl The American Political Science Review 75 (4): 975 Œ87. Caughey, Devin, and Christopher Warshaw. 2015. fiThe Dynamics of State Policy Liberalism, 1936 Œ2014.fl Am erican Journal of Political Science 60 (4): 899 -913. CenterLink. 2016. fiCenterLink LGBT Community Center Member Directory. fl Accessed on March 10, 2019: http://www.lgbtcenters.org:80/Centers/find -a-center.aspx . Cheng, Simon and J. Scott Long. 2007. fiTesting for IIA in the Multinomial Logit Model.fl Sociological Methods & Research 35 (4 ): 583 -600. Chubb, John. 1983. Interest Groups and the Bureaucracy . Stanford, CA: Stanford University Press. Clark, Jill. 1985. fiPolicy Diffusion and Program Scope: Research Directions.fl Publius 15 (4): 61 Œ70. CNN. 2004. fiTranscript of Bush Statement.fl February 24, 2004. Accessed on January 17, 2019 at: http://www.cnn.com/2004/ALLPOLITICS/02/24/elec04.prez.bush.transcript/ . Cobb, Roger W., and Charles D. Elder. 1983 . Participation in American Politics: The Dynamics of Agenda -Building . Baltimore, MD: Johns Hopkins University Press. Cohen -Vogel, L. and W. K. Ingle. 2007. fiWhen Neighbors Matter Most: Innovation, Diffusion and State Adoption in Tertiary Education.fl Journal of Education Policy 22 (3): 241 -262. Cole, David. 2016. Engines of Liberty: The Power of Citizen Activists to Make Constitutional Law . New York: NY. Basic Books. Colvin, Roddrick A. 2004. fiAgenda Setting, Innovation, and State Gay Rights Policy: An Event History Analysis.fl American Review of Politics 25 (Fall): 241 -263. Conger, Kimberly H. 2009. The Christian Right in Republican State Politics . New York, NY: Palgrave Macmillan. Conger, Kimberly H. and Paul A. Djupe. 2016. fiCulture War Counter -Mobilization: Gay Rights and Religious Right Groups in the States.fl Interest Groups & Advocacy 5 (3): 278 -300. Constantelos, John. 2010. fiPlaying the Field: Federalism and the Politics of Venue Shopping in the United States and Canada.fl Publius: The Journal of Federalism 40 (3): 460 Œ83. 278 Damore, David F., Shaun Bowler, and Stephen P. Nicholson. 2012. fiAgenda S etting by Direct Democracy: Comparing the Initiative and the Referendum.fl State Politics & Policy Quarterly 12 (4): 367 -393. Dear, Jake, and Edward W. Jessen. 2007. fiMeasuring the Comparative Influence of State Supreme Courts: Comments on Our ‚Followed Ra tes™ Essay.fl Accessed June 2, 2016. http://lawreview.law.ucdavis.edu/issues/41/4/addendum/41 -4_Dear -Jessen.pdf . Desmarais, Bruce A., Jeffrey J. Harden, and Frederic k J. Boehmke. 2015. fiPersistent Policy Pathways: Inferring Diffusion Networks in the American States.fl American Political Science Review 109 (2): 392 Œ406. Dilger, Robert Jay, George A. Krause, and Randolph R. Moffett. 1995. fiState Legislative Professional ism and Gubernatorial Effectiveness, 1978 -1991.fl Legislative Studies Quarterly 20 (4): 553 Œ71. Dinan, John. 2018. State Constitutional Politics: Governing by Amendment in the American States . Chicago, IL: University of Chicago Press. Dinan, John, and Dale Krane. 2005. fiThe State of American Federalism.fl Publius: The Journal of Federalism 36 (3): 327 -374. Dolowitz, David P., and David Marsh. 2000. fiLearning from Abroad: The Role of Policy Transfer in Contemporary Policy -Making.fl Govern ance 13 (1): 5 Œ23. Donovan, Todd and Shaun Bowler. 1998. fiDirect Democracy and Minority Rights: An Extension.fl American Journal of Political Science 42 (3): 1020 -1024. Donovan, Todd, Shaun Bowler, and David McCuan. 2001. fiPolitical Consultants and the Initiative Industrial Complex.fl In Dangerous Democracy? The Battle Over Ballot Initiatives in America by Larry J. Sabato, Howard R. Ernst, and Bruce A. Larson (Eds.), 101 -134. Lanham, MD: Rowman & Littlefield. Donovan, Todd, Jim Wenzel, and Shaun Bowler. 2000. fiDirect Democracy and Gay Rights Initiatives After Romer fl in The Politics of Gay Rights by Rimmerman, Craig A., Kenneth D. Wald, and Clyde Wilcox, eds. The Chicago Series on Sexuality, History, and Society. Chicago, IL: University of Chicago Press. Dorf, Michael C., and Sidney Tarrow. 2014. fiStrange Bedfellows: How an Anticipatory Countermovement Brought Same -Sex Marriage into the Public Arena.fl Law & Social Inquiry 39 (2): 449 Œ73. Dow, Jay K., and James W. Endersby. 2004. fiMultinomial Probit and M ultinomial Logit: A Comparison of Choice Models for Voting Research.fl Electoral Studies 23: 107 -122. 279 Dumas, Marion. 2017. fiTaking the Law to Court: Citizen Suits and the Legislative Process.fl American Journal of Political Science 61 (4): 944 -957. Dye, Thomas. 1966. Politics, Economics, and the Public: Policy Outcomes in the American States . Chicago, IL: Rand -McNally. Earl, Jennifer and Sarah A. Soule. 2001. fiThe Differential Protection of Minority Groups: The Inclusion of Sexual Orientation, Gender and Disability in State Hate Crime Laws, 1976 Œ1995.fl Research in Political Sociology: The Politics of Social Inequality 9:3Œ34. Elder, Charles D., and Roger W. Cobb. 1984. fiAgenda -Building and the Politics of Aging.fl Policy Studies Journal 13 (1): 115 Œ29. Elkins, Zachary, and Beth Simmons. 2005. fiOn Waves, Clusters, and Diffusion: A Conceptual Framework.fl Annals of the American Academy of Political and Social Science 598 (March): 33 Œ51. Engel, Stephen M. 2007. fiOrganizational Identity as a Constraint on St rategic Action: A Comparative Analysis of Gay and Lesbian Interest Groups.fl American Political Development 21 (2): 66 -91. Engel, Stephen M. 2016. Fragmented Citizens: The Changing Landscape of Gay and Lesbian Lives . New York, NY: NYU Press. Enns, Pete r K., and Julianna Koch. 2013. fiPublic Opinion in the U.S. States 1956 to 2010.fl State Politics & Policy Quarterly 13 (3): 349 Œ72. Equality Federation Institute and Movement Advancement Project . 2006 . fiState of the States Report: A Survey Assessing the Ca pacity of State -Based Lesbian, Gay, Bisexual, and Transgender Advocacy Organizations. fl Equality Federation Institute and Movement Advancement Project . 2007. fiState of the States Report: A Survey Assessing the Capacity of State -Based Lesbian, Gay, Bisexual, and Transgender Advocacy Organizations. fl Equality Federation Institute and Movement Advancement Project . 2008. fiState of the States Report: A Survey Assessing the Capacity of State -Based Lesbian, Gay, Bisexual, and Transgender Advocacy Organizat ions. fl Equality Federation Institute and Movement Advancement Project . 2009. fiState of the States Report: A Survey Assessing the Capacity of State -Based Lesbian, Gay, Bisexual, and Transgender Advocacy Organizations. fl Equality Federation Institute and Mo vement Advancement Project . 2011. fiState of the States Report: A Survey Assessing the Capacity of State -Based Lesbian, Gay, Bisexual, and Transgender Advocacy Organizations. fl 280 Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1993. Statehouse Democ racy: Public Opinion, and Policy in the American States . Cambridge University Press. Eshbaugh -Soha, Matthew. 2006. fiThe Conditioning Effects of Policy Salience and Complexity on American Political Institutions.fl The Policy Studies Journal 34 (2): 223 Œ43. Eyestone, Robert. 1977. fiConfusion, Diffusion, and Innovation.fl The American Political Science Review 71 (2): 441 Œ447. . Family Research Council . 2002. State Policy Organizations. Accessed archives on March 10, 2019 : https://www.frc.org/state -policy -organizations . Family Research Council . 2018. State Policy Organizations. Accessed on March 10, 2019 : https://www.frc.org/state -policy -organizations . Fay, Daniel L., and Jeffrey B. Wenger. 2015. fiThe Political Structure of Policy Diffusion.fl Policy Studies Journal 44 (3):349 -365. Ferraiolo, Kat hleen. 2008. fiState Policy Innovation and the Federalism Implications of Direct Democracy.fl Publius: The Journal of Federalism 38 (3): 488 -514. Fetner, Tina. 2001. fiWorking Anita Bryant: The Impact of Christian Anti -Gay Activism on Lesbian and Gay Movemen t Claims.fl Social Problems 48 (3): 411 Œ28. Fetner, Tina. 2008. How the Religious Right Shaped Lesbian and Gay Activism . Social Movements, Protest, and Contention, v. 31. Minneapolis, MN: University of Minnesota Press. Finnemore, Martha, and Kathryn Sikki nk. 1998. fiInternational Norm Dynamics and Political Change.fl International Organization 52 (4): 887 Œ917. Fiorino, Daniel J. 1976. fiThe Diffusion of Judicial Innovations: Court Decisions as a Source of Change.fl (Paper presented at the annual convention of the International Communication Association, Portland, OR.) Fleischmann, Arnold and Laura Moyer. 2009. fiCompeting Social Movements and Local Political Culture: Voting on Ballot Propositions to Ban Same -Sex Marriage in the U.S. States.fl Social Science Quarterly 90 (1): 134 -149. Foster, John L. 1978. fiRegionalism a nd Innovation in the American States.fl The Journal of Politics 40 (1): 179 Œ87. Frank, Mark. W. 2009. fiInequality and Growth in the United States: Evidence from a New State -Level Panel of Income Inequality Measure.fl Economic Inquiry 47 (1): 55Œ68. 281 Franze se, Robert J., Jr., and Jude C. Hays. 2006. fiStrategic Interaction among EU Governments in Active Labor Market Policy -Making: Subsidiarity and Policy Coordination under the European Employment Strategy.fl European Union Politics 7:167 Œ189. Frederickson, H. George, Gary Alan Johnson, and Curtis Wood. 2004. fiThe Changing Structure of American Cities: A Study of the Diffusion of Innovation.fl Public Administration Review 64 (3): 320Œ30. Freedom to Marry. 2015. fiHistory and Timeline of the Freed om to Marry in the United States.fl June 26, 2015. Accessed January 25, 2016: http://www.freedomtomarry.org/pages/history -and -timeline -of-marriage . Freeman, Richard. 2008. fiLearning in Public Policy.fl In The Oxford Handbook of Public Policy , Robert E. Goodin, Michael Moran, and Martin Rein (eds.). Oxford: Oxford University Press. Füglister, Katharina. 2012. fiWhere Does Learning Take Place? The Role o f Intergovernmental Cooperation in Policy Diffusion.fl European Journal of Political Research 51 (3): 316 Œ49. Gallagher, John, and Chris Bull. 2001. Perfect Enemies: The Battle between the Religious Right and the Gay Movement . Updated edition, First Madiso n Books edition. Lanham, MD: Madison Books. Garrett, Kristin N., and Joshua M. Jansa. 2015. fiInterest Group Influence in Policy Diffusion Networks.fl State Politics & Policy Quarterly 15 (3): 387 Œ417. Gerber, Elisabeth R. 1996. fiLegislative Response to th e Threat of Popular Initiatives.fl American Journal of Political Science 40 (1): 99 -128. Gilardi, Fabrizio. 2010. fiWho Learns from What in Policy Diffusion Processes?fl American Journal of Political Science 54 (3): 650 Œ66. Gilardi, Fabrizio. 2016. fiFour Ways We Can Improve Policy Diffusion Research.fl State Politics & Policy Quarterly 16 (1): 8 -21. Gilardi, F abrizio , and K atharina Füglister . 2008. fiEmpirical Modeling of Policy Diffusion in Federal States: The Dyadic Approach .fl Swiss Political Science Review 14 (3): 413 Œ50. Gilardi Fabrizio, Charles R. Shipan and Bruno Wueest. 2019. fiPolicy Diffusion: The Issue -Definition Stage.fl Working Paper. Accessed on May 27, 2019: https://www.fabriziogilardi.org/resources/papers/policy -diffusion -issue -definition.pdf . Gilardi, Fabrizio, and Fabio Wasserfallen. 2019. fiThe Politics of Policy Diffusion.fl European Journal of Political Rese arch . Online version. Print version forthcoming. Givan, Rebecca Kolins, Kenneth M. Roberts, and Sarah Anne Soule, eds. 2010. The Diffusion of Social Movements: Actors, Mechanisms, and Political Effects . New York, NY: Cambridge University Press. 282 Givel, Michael. 2006. fiPunctuated Equilibrium in Limbo: The Tobacco Lobby and U.S. State Policymaking from 1990 to 2003.fl Policy Studies Journal 34 (3): 405 -418. Glick, Henry R., and Scott P. Hays. 1991. fiInnovation and Reinvention in State Policymaking: Theory and the Evolution of Living Will Laws.fl The Journal of Politics 53 (3): 835 Œ50. Glick, Henry R. 1992. fiJudicial Innovation and Policy Re -Invention: State Supreme Courts and the Right to Die.fl The Western Political Quarterly 45 (1): 71 Œ92. https://doi.org/10.2307/448763 . Goggin, M. L., Bowman, A. O. M., Lester, J. P., and O™Toole, J. L. J. 1990 . Implementation Theory and Practice: Toward a Third Generation . Glenview, IL: Scott Foresman/Little, Brown and Company. Gormley, William T. 1986. fiRegulatory Issue Networks in a Federal System.fl Polity 18 (4): 595 Œ620. Grace, Fannie. 2002. fiArnold™s Marijuana Moment.fl CBS News . November 15, 2002. Accessed on July 15, 2016 at: http://www.cbsnews.com/news/arnolds -marijuana -moment/ . Graham, Erin R., Charles R. Shipan, and Craig Volden. 2013. fiThe Diffusion of Policy Diffusion Research in Political Science.fl British Jou rnal of Political Science 43 (3): 673 -701. Gray, Virginia. 1973. fiInnovation in the States: A Diffusion Study.fl The American Political Science Review 67 (4): 1174 -1185. Gray, Virginia, and David Lowery. 1988. fiInterest Group Politics and Economic Growth in the U.S. States.fl The American Political Science Review 82 (1): 109 Œ31. Gray, Virginia, and David Lowery. 1996. The Population Ecology of Interest Representation: Lobbying Communities in the American States . Ann Arbor , MI : The University of Michigan Press. Gray, V., D. Lowery, M. Fellowes, and A. McAtee. 2004. fiPublic Opinion, Public Policy, and Organized Interests in the American States.fl Political Research Quarterly 57(3): 411 -. Green, John C. 2000. fiAntigay: Varieties of Opposition to Gay Rightsfl in The Politics of Gay Rights by Rimmerman, Craig A., Kenneth D. Wald, and Clyde Wilcox, eds. The Chicago Series on Sexuality, History, and Society. Chicago, IL: University of Chicago Pr ess. Haider -Markel, Don ald P. 1998. fiThe Politics of Social Regulatory Policy: State and Federal Hate Crime Policy and Implementation Effort.fl Political Research Quarterly 51 (1): 69 Œ88. Haider -Markel, Donald P. 2000. fiLesbian and Gay Politics in the Sta tes: Interest Groups, Electoral Politics, and Policyfl in The Politics of Gay Rights by Rimmerman, Craig A., Kenneth D. Wald, and Clyde Wilcox, eds. The Chicago Series on Sexuality, History, and Society. Chicago, IL: University of Chicago Press. 283 Haider -Mar kel, Donald P. 2001a. fiPolicy Diffusion as a Geographic Expansion of the Scope of Political Conflict: Same -Sex Marriage Bans in the 1990s.fl State Politics & Policy Quarterly 1 (Spring): 5 Œ 26. Haider -Markel, Donald P. 2001b. fiShopping for Favorable Venue s in the States: Institutional Influences on Legislative Outcomes of Same -Sex Marriage Bills.fl American Review of Politics 22 (Spring): 27 -53. Haider -Markel, Donald P. 2007. fiRepresentation and Backlash: The Positive and Negative Influence of Descriptive Representation.fl Legislative Studies Quarterly 32 (1): 107 -133. Haider -Markel, Donald P. and Kenneth J. Meier. 1996. fiThe Politics of Gay and Lesbian Rights: Expanding the Scope of the Conflict.fl The Journal of Politics 58 (2): 332 -349. Haider -Markel, Donald P. and Kenneth J. Meier. 2003. fiLegislative Victory, Electoral Uncertainty: Explaining Outcomes in the Battles over Lesbian and Gay Civil Rights.fl Review of Policy Research 20 (4): 671 -690. Hannah, A. Lee, and Daniel J. Mallinson. 2018. fiDefiant I nnovation: The Adoption of Medical Marijuana Laws in the American States.fl Policy Studies Journal 46 (2): 402 Œ23. Harrison, Brian F., and Melissa R. Michelson. 2017. Listen, We Need to Talk: How to Change Attitudes about LGBT Rights . New York, NY: Oxford University Press. Hays, Scott P. 1996. fiPatterns of Reinvention: The Nature of Evolution During Innovation.fl Policy Studies Journal 24 (4): 551 Œ66. Hecht, Peter. 2014. Weed Land: Inside America's Marijuana Epicenter and How Pot Went Legit . Berkeley, CA: University of California Press. Heclo, Hugh. 1974. Modern Social Politics in Britain and Sweden: From Relief to Income Maintenance . New Haven, CT: Yale University Press. Hero, Rodney E., and Caroline J. Tolbert. 1996. fiA Racial/Ethnic Diver sity Interpretation of Politics and Policy in the States of the U.S.fl American Journal of Political Science 40 (3): 851 Œ71. Hetherington, Marc J., and Thomas J. Rudolph. 2015. Why Washington Won™t Work: Polarization, Political Trust, and the Governing Crisis . Chicago, IL: University of Chicago Press. Hicks, William D. 2013. fiInitiatives within Representative Government: Political Competition and Initiative Use in the American States.fl State Politics & Policy Quarterly 13 (4): 471 -494. Higdon , James. 2019. fiWhy 2019 Could Be Marijuana™s Biggest Year Yet: A green tide in Congress raises hopes that pot could be legal under federal law by year™s end.fl Politico Magazine . January 21, 2019. Accessed on January 21, 2019 at: 284 https://www.politico.com/magazine/story/2019/01/21/marijuana -legalization -congress - . Hinchliffe, Kelsey L., and Frances E. Lee. 2016. fiParty Competition and Conflict in State Legislatures.fl State Politics & Policy Quarterly 16 (2): 172 Œ97. Hinkle, Rachael K. 2015. fiInto the Words: Using Statutory Text to Explore the Impact of Federal Courts on State Policy Diffusion.fl American Journal of Poli tical Science 59 (4): 1002 -1021. Hinkle, Rachael K., and Michael J. Nelson. 2016. fiThe Transmission of Legal Precedent among State Supreme Courts in the Twenty -First Century.fl State Politics & Policy Quarterly 16 (4): 391Œ410. Hirsch , Barry T. and David A. Macpherson . 2003. fiUnion Membership and Coverage Database from the Current Population Survey: Note .fl Industrial and Labor Relations Review 56 (2): 349 -354 Hirshman, Linda. 2012. Victory: The Tr iumphant Gay Revolution . New York, NY: Harper Collins. Hoeffel, John. 2010. fiHolder vows fight over prop. 19.fl L.A. Times . October 16, 2010. Accessed on July 15, 2016 at: http://articles.latimes.com/2010/oct/16/local/la -me-marijuana -holder -20101016 . Hofferbert, Richard I. 1974. The Study of Public Policy . Indianapolis, IN: Bobbs -Merrill. Holbrook, Thomas M., and Emily Van Dunk. 1993. fiElectoral Competition in the American States.fl The American Political Science Review 87 (4): 955 Œ62. Holyoke, Thomas T. 2003. fiChoosing Battlegrounds: Interest Group Lobbying across Multiple Venues.fl Political Research Quarterly 56 (3): 325 Œ36. Holyoke, Thomas T., Heath Brown, and Jeffrey R. Henig. 2012. fiShopping in the Political Arena Strategic State a nd Local Venue Selection by Advocates.fl State and Local Government Review 44 (1): 9 Œ20. Hollander, Robyn and Haig Patapan. 2016. fiMorality Policy and Federalism: Innovation, Diffusion and Limits.fl Publius: The Journal of Federalism 47 (1): 1 -26. Howlett , Michael, and Jeremy Rayner. 2008. fiThird Generation Policy Diffusion Studies and the Analysis of Policy Mixes: Two Steps Forward and One Step Back?fl Journal of Comparative Policy Analysis: Research and Practice 10 (4): 385 Œ402. Hume, Robert J. 2011. fiCo mparing Institutional and Policy Explanations for the Adoption of State Constitutional Amendments: The Case of Same -Sex Marriage.fl American Politics Research 39 (6): 1097 -1126. 285 Jacoby, William G., and Saundra K. Schneider. 2014. fiState Policy and Democratic Representation.fl In t he Oxford Handbook of State and Local Government by Donald P. Haider -Markel (eds.). Oxford: Oxford University Press. Jenkins -Smith, Hank C., Daniel Nohr stedt, Christopher M. Weible, and Paul A. Sabatier. 2014. fiThe Advocacy Coalition Framework: Foundations, Evolution, and Ongoing Research.fl in Paul A. Sabatier and Christopher M. Weible (eds.) Theories of the Policy Process . Boulder, CO: Westview Press. John, Peter. 1999. fiIdeas and Interests; Agendas and Implementation: An Evolutionary Explanation of Policy Change in British Local Government Finance.fl The British Journal of Politics & International Relations 1 (1): 39 Œ62. Jones, Bradford S., and Regina P . Branton. 2005. fiBeyond Logit and Probit: Cox Duration Models of Single, Repeating, and Competing Events for State Policy Adoption.fl State Politics & Policy Quarterly 5 (4): 420 Œ43. Jordan, Marty P., and Matt Grossmann. 2018. The Correlates of State Poli cy Project v.2.1 . East Lansing, MI: Institute for Public Policy and Social Research (IPPSR). Accessed at: http://ippsr.msu.edu/public -policy/correlates -state -policy . Jourdain, Charlotte, Simon Hug, and Frédéric Varone. 2017. fiLobbying Across Venues: An Issue -Tracing Approach.fl State Politics & Policy Quarterly 17 (2): 127 Œ153. Kamin, Sam. 2015. fiThe Battle of the Bulge: The Surprising Last Stand Against State Marijuana Legalization.fl Publius: The Journal of Federalism 45 (3): 427 Œ51. Kane, Melinda D. 2010. fiYou™ve Won, Now What? The Influence of Legal Change on Gay and Lesbian Mobilization, 1974 -1999.fl The Sociological Quarterly 51 (2): 255 -277. Karch, A ndrew. 2006. fiNational Intervention and the Diffusion of Policy Innovations.fl Ame rican Politics Research , 34(4): 403 Œ26. Karch, Andrew. 2007 a. Democratic Laboratories: Policy Diffusion Among the American States . Ann Arbor, MI: The University of Michigan Press. Karch, Andrew. 2007b. fiEmerging Issues and Future Directions in State Poli cy Diffusion Research.fl State Politics & Policy Quarterly , 7(1): 54 Œ80. Karch, Andrew . 2009. fiVenue Shopping, Policy Feedback, and American Preschool Education.fl Journal of Policy History , 21(1): 38 Œ60. Karch, Andrew . 2012. fiVertical Diffusion and the Policy -Making Process The Politics of Embryonic Stem Cell Research.fl Political Research Quarterly 65(1): 48 Œ61. 286 Karch, A ., and M . Cravens. 2014. fiRapid Diffusion and Policy Reform The Adoption and Modification of Three Strikes Laws.fl State Politics & Policy Quarterly 14(4): 461 Œ91. Karch, Andrew, Sean C. Nicholson -Crotty, Neal D. Woods, and Ann O™M Bowman. 2016. fiPolicy Diffusion and the Pro -Innovation Bias.fl Political Research Quarterly 69 (1): 83 Œ95. Karch, Andrew, a nd Aaron Rosenthal. 2015. fiVertical Diffusion and the Shifting Politics of Electronic Commerce.fl State Politics & Policy Quarterly 16 (1), 22 -43. Keck, Thomas M. 2009. fiBeyond Backlash: Assessing the Impact of Judicial Decisions on LGBT Rights.fl Law & Soc iety Review 43 (1): 151 -185. Kelly, Nathan J., and Christopher Witko. 2014. fiGovernment Ideology and Unemployment in the U.S. States.fl State Politics & Policy Quarterly 14 (4): 389 -413. Key, V. 0. 1949. Southern Politics in State and Nation . New York , NY : Alfred A. Knopf. Kingdon, John W. 1984. Agendas, Alternatives, and Public Policies . New York, NY: HarperCollins College Publishers. King, Gary, Robert O. Keohane, and Sidney Verba. 1994. Designing Social Inquiry: Scientific Inference in Qualitative Research . Princeton, NJ: Princeton University Press. Klarner, Carl, 2013 a. fiOther Scholars Competitiveness Measures.fl Accessed at: http://hdl.handle. net/1902.1/22519,HarvardDataverse,V1. Klarner, Carl, 2013 b. fiState Economic Data.fl Accessed at: http://hdl.handle.net/1902.1/20404, HarvardDataverse,V1. Koski, Chris. 2010. fiGreening America™s Skylines: The Diffusion of Low -Salience Policies.fl Policy Studies Journal 38 (1): 93 Œ117. Kropko, Jonathan. 2008. fiChoosing Between Multinomial Logit and Multinomial Probit Models for Analysis of Unordered Choice Data.fl Masters Thesis. Accessed on May 20, 2019 at: https://c dr.lib.unc.edu/concern/dissertations/d217qp57h . Lang, C. and S. Pearson -Merkowitz . 2014. fiPartisan Sorting in the United States, 1972 -2012: New Evidence from a Dynamic Analysis. fl Political Geography 48: 119-129. Lawrence, Eric D., Todd Donovan, and Sha un Bowler. 2009. fiAdopting Direct Democracy: Tests of Competing Explanations of Institutional Change.fl American Politics Research 37 (6): 1024 -1047. Lax, Jeffrey R. and Justin H. Phillips. 2009. fiGay Rights in the States: Public Opinion and Policy Respons iveness.fl American Political Science Review 103 (3): 367 -386. 287 Leighley, Jan E., and Jonathan Nagler. 2013. Who Votes Now?: Demographics, Issues, Inequality, and Turnout in the United States . Princeton, NJ: Princeton University Press. Lewis, Daniel C. 2011. fiDirect Democracy and Minority Rights: Same -Sex Marriage Bans in the U.S. States.fl Social Science Quarterly 92 (2): 364 -383. Lewis, Daniel. 2013. Direct Democracy and Minority Rights: A Critical Assessment of the Tyranny of the Maj ority in the American States . New York, NY: Routledge. Lewis, Daniel C., and Matthew L. Jacobsmeier. 2017. fiEvaluating Policy Representation with Dynamic MRP Estimates: Direct Democracy and Same -Sex Relationship Policies in the United States.fl State Polit ics & Policy Quarterly 17 (4): 441 Œ64. Lewis, Gregory B., and Charles W. Gossett. 2008. fiChanging Public Opinion on Same -Sex Marriage: The Case of California.fl Politics and Policy 36 (1): 4 -30. Lewis, Daniel C., Saundra K. Schneider, and William G. Jacoby. 2015. fiInstitutional Characteristics and State Policy Priorities the Impact of Legislatures and Governors.fl State Politics & Policy Quarterly 15 (4): 447-475 . Ley, Aaron J. 2014. fiThe Costs an d Benefits of American Policy -Making Venues.fl Law & Society Review 48 (1): 91 -126. Ley, Aaron J., and Edward P. Weber. 2015. fiThe Adaptive Venue Shopping Framework: How Emergent Groups Choose Environmental Policymaking Venues.fl Environmental Politics 24 (5): 703 Œ22. Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables . Thousand Oaks, CA: SAGE Publications . Lowery, David. 2013. fiLobbying Influence: Meaning, Measurement and Missing.fl Interest Group & Advocacy 2 (1): 1 Œ 26. Lowery, David, Virginia Gray, and Frank R. Baumgartner. 2011. fiPolicy Attention in State and Nation: Is Anyone Listening to the Laborator ies of Democracy?fl Publius: The Journal of Federalism 41 (2): 286 Œ310. Lowry, William R. 2005. fiPolicy Reversal and Changing Politics: State Governments and Dam Removals.fl State Politics & Policy Quarterly 5 (4): 394 Œ419. Lubell, Mark. 2013. fiGoverning Institutional Complexity: The Ecology of Games Framework.fl Policy Studies Journal 41 (3): 537 Œ59. Lubell, Mark, Adam Douglas Henry, and Mike McCoy. 2010. fiCollaborative Institutions in an Ecology of Games.fl American Journal of Political Science 54 (2): 28 7Œ300. 288 Lupia, Arthur, Yanna Krupnikov, Adam Seth Levine, Spencer Piston, and Alexander Von Hagen -Jamar. 2010. fiWhy State Constitutions Differ in Their Treatment of Same -Sex Marriage.fl The Journal of Politics 72 (4): 1222 Œ35. Lutz, Donald S. 1994. fiToward a Theory of Constitutional Amendment.fl The American Political Science Review 88 (2): 355 Œ70. Magleby, David B. 1988. fiTaking the Initiative: Direct Legislation and Direct Democracy in the 1980s.fl PS: Political Science and Politics 21 (3): 600 -611. Makse, Todd, and Craig Volden. 2011. fiThe Role of Policy Attributes in the Diffusion of Innovations. fl The Journal of Politics 73 (1): 108 -124. Mallinson, Daniel J. 2016. fiBuilding a Better Speed Trap Measuring Policy Adoption Speed in the American States.fl State Politics & Policy Quarterly 16 (1): 98 Œ120. Maltzman, Forrest, and Charles R. Shipan. 2008. fiChange, Continuity, and the Evolution of the Law.fl American Journal of Political Science 52 (2): 252 Œ67. Marijuana Policy Project (MPP). 2016. State Policy . Accessed on July 15, 2016 at: https://www.mpp.org/states/ . Martin, Alyson and Nushin Rashidian. 2014. A New Leaf: The End of Cannabis Prohibition . New York, NY: The New Press. Martinez, Edecio. 2010. fiProp 19 Results: California Marijuana Legalization Measur e Loses.fl CBS News . November 3, 2010. Accessed on July 15, 2016 at: http://www.cbsnews.com/news/prop -19-results -california -marijuana -legalization -measure -loses/ . Matsusaka, John G. and Nolan M. McCarty. 2001. fiPolitical Resource Allocation: Benefits and Costs of Voter Initiatives.fl Journal of Law, Economics, & Organization 17 (2): 413 -448. May, P . J. 1992. fiPolicy Learning and Failure.fl Journal of Public Policy 12 (4): 331 Œ54. Mayhew, David R. 1974. Congress: The Electoral Connection . New Haven, CT: Yale University Press. McCann, Pamela J. Clouser, Charles R. Shipan, and Craig Volden. 2015. fiTop -Down Federalism: State Policy Responses to National Government Discussions.fl Publius: The Journal of Federalism 45 (4): 495 Œ525. McConnell, Allan. 2010 . fiPolicy Success, Policy Failure and Grey Areas In -Between.fl Journal of Public Policy 30 (3): 345 Œ362. McGrath, Robert J. 2011. fiElectoral Competition and the Frequency of Initiative Use in the U.S. States.fl American Politics Research 39 (3): 611 -638. 289 Meseguer, Covadonga. 2005. fiPolicy Learning, Policy Diffusion, and the Making of a New Order.fl The ANNALS of the American Academy of Political and Social Science 598 (1): 67 Œ82 . Meyer, David S., and Suzanne Staggenborg. 1996. fiMovements, Countermovements, and the Structure of Political Opportunity.fl American Journal of Sociology 101 (6): 1628 Œ60. Mezey, Susan Gluck. 2017. Beyond Marriage: Continuing Battles for LGBT Rights . Lanham, MD: Rowman & Littlefield. Miller, Kenneth P. 2009. Direct Democracy and the Courts . Cambridge: Cambridge University Press. Miller, Susan E., Jill Nicholson -Crotty, and Sean Nicholson -Crotty. 2018. fiThe Consequences of Legislative Term Limits for Policy Diffusion.fl Political Resear ch Quarterly 71 (3): 573 -585. Miller, Susan M., Eve M. Ringsmuth, and Joshua M. Little. 2015. fiPushing Constitutional Limits in the U.S. States Legislative Professionalism and Judicial Review of State Laws by the U.S. Supreme Court.fl State Politics & Pol icy Quarterly 15 (4): 476 Œ91. Mintrom, M. 1997. fiPolicy Entrepreneurs and the Diffusion of Innovation.fl American Journal of Political Science 41(3): 738 -770. Mintrom, M. and S. Vergari. 1998. fiPolicy Networks and Innovation Diffusion: The Case of State Education Reforms.fl The Journal of Politics 60 (1) : 126 -148. Mooney, Christopher Z. 2001. fiModeling Regional Effects on State Policy Diffusion.fl Political Research Quarterly 54(1): 103 Œ24. Mooney, Christopher Z. 2013. fiMeasuring State House Speakers™ Formal Powers, 1981 Œ2010.fl State Politics & Policy Quarterly 13 (2): 262 Œ73. Mooney, Christopher Z. and Mei -Hsien Lee. 1995. fiLegislative Morality in the American States: The Case of Pre -Roe Ab ortion Regulation Reform.fl American Journal of Political Science 39 (3): 599 -627. Mooney, C. Z. and M. H. Lee. 1999. fiThe Temporal Diffusion of Morality Policy: The Case of Death Penalty Legislation in the American States.fl Policy Studies Journal 27 (4): 766-780. Mooney, Christopher Z. and Richard G. Schuldt. 2008. fiDoes Morality Policy Exist? Testing a Basic Assumption.fl Policy Studies Journal 36 (2): 199 Œ218. Morehouse, Sarah McCally. 1981. State Politics, Parties and Policy . New York, NY: CBS Colleg e Publishing. Movement Advancement Project. 2019. fiHate Crime Laws. fl Accessed on January 8, 2019: http://www.lgbtmap.org/equality -maps/hate_crime_laws . 290 Mucciaroni, Gary. 2008. Same Se x, Different Politics: Success and Failure in the Struggles for Gay Rights . Chicago, IL: University of Chicago Press. Mucciaroni, Gary. 2011. fiThe Study of LGBT Politics and Its Contributions to Political Science.fl PS: Political Science & Politics 44 (1): 17Œ21. National Conference of State Legislatures (NCSL). 2016. Ballot Measures Database . Accessed at: http://www.ncsl.org/research/elections -and -campaigns/ballot -measures -database.aspx . National Conference of State Legislatures (NCSL). 2018. fiMarijuana Overvie w.fl December 14, 2018. Accessed on January 10, 2019 at: http://www.ncsl.org/research/civil -and -criminal -justice/marijuana -overview.aspx . National Gay and Lesbian Task Force. 2013. fiState Laws Prohibiting Recognition of Same -Sex Relationships.fl Accessed January 25, 2016: http://www.thetaskforce.org/static_html/downloads/reports/issue_maps/samesex_relatio nships_7_09.pdf . New York Times Index. New York , NY : New York Times Co . Nicholson -Crotty, Sean. 2006. fiReassessing Madison™s Diversity Hypothesis: The Case of Same -Sex Marriage.fl The Journal of Politics 68 (4): 922 Œ30. Nicholson -Crotty, Sean. 2009. fiThe Politics of Diffusion: Public Po licy in the American States.fl Journal of Politics 71 (1): 192 Œ205. Nicholson -Crotty, Sean, and Sanya Carley. 2015. fiEffectiveness, Implementation, and Policy Diffusion Or ‚Can We Make That Work for Us?™fl State Politics & Policy Quarterly 16 (1): 78-97. Nownes, Anthony J. and Daniel Lipinski. 2005. fiThe Population Ecology of Interest Group Death: Gay and Lesbian Rights Interest Groups in the United States, 1945 -98.fl British Journal of Political Science 35 (2): 303 -319. Oakley, M. R. 2009. fiAgenda Settin g and State Policy Diffusion: The Effects of Media Attention, State Court Decisions, and Policy Learning on Fetal Killing Policy.fl Social Science Quarterly 90 (1): 164 Œ78. O™Dell , Rob and Nick Penzenstadler . 2019. fiYou Elected Them to Write New Laws. They ™re Letting Corporations Do It Instead.fl USA Today. April 4, 2019. Accessed April 22, 2019: https://www.usatoday.com/in -depth/news/investigations/2019/04/03/abortion -gun -laws -stand -your -ground -model -bills -conservatives -liberal -corporate -influence -lobbyists/3162173002/ . 291 Pacheco, Julianna. 2011. fiMeasuring State Public Opinion over Time Using National Surveys: A Guideline for Scholars.fl State Politics and Policy Quarterly 11 (4):415 -39. Pacheco, Julianna. 2012. fiThe Social Contagion Model: Exploring the Role of Public Opinion on the Diffusion of Antismoking Legislation across the American States.fl The Journal of Politics 74 (1): 187 Œ202. Pacheco, Julianna. 2014. fiMeasuring and Evaluating Changes in State Opinion Across Eight Issues.fl American Politics Research 42 (6): 986 Œ1009. Pacheco, Julianna. 2017. fiFree -Riders or Competitive Races? Strategic Interaction Across the American States on Tobacco Policy Making.fl State Politics & Policy Quarterly 17(3): 1 -20. Parent, Jonathan. 2010. fi‚Horizontal™ and ‚Vertical™ Venue Selection: LGBT Rights and Abortion Policy in Canada and the United States.fl SSRN Scholarly Paper ID 1580530. Rochester, NY: Social Science Research Network. . Parinandi, Srinivas. 2013. fiConditional Bureaucratic Discretion and State Welfare Diffusion under AFDC.fl State Politics & Policy Quarterly 13 (2): 244 Œ61. Parinandi, Srinivas C. 2014. fiSolving Brandeis™s Mystery: Professionalism, Reelection, and Legislative Policy Innovation in the American States.fl Working Paper. Accessed July 29, 2016 at: . Patashnik, Eric M. 2014. Reforms at Risk: What Happens After Major Policy Changes Are Enacted . Princeton, NJ: Princeton University Press. Pernick, Sam. fiLessons from 2018 Michigan Ballot Initiative Campaigns.fl Roundtable. Michigan chapter of Scholars Strategy Network. East Lansi ng, MI. January 10, 2019. Peterson, Paul E. 1995. The Price of Federalism . Washington, DC: The Brookings Institution. Pickerill, J. Mitchell, and Paul Chen. 2008. fiMedical Marijuana Policy and the Virtues of Federalism.fl Publius: The Journal of Federal ism 38 (1): 22 Œ55. Pierce, Patrick A., and Donald E. Miller. 1999. fiVariations in the Diffusion of State Lottery Adoptions: How Revenue Dedication Changes Morality Politics.fl Policy Studies Journal 27 (4): 696Œ706. Pierceson, Jason. 2013. Same -Sex Marria ge in the United States: The Road to the Supreme Court . Lanham, MD: Rowman & Littlefield Publishers. 292 Pierceson, Jason, Adriana Piatti -Crocker, and Shawn Schulenberg. 2010. Same -sex Marriage in the Americas: Policy Innovation for Same -Sex Relationships . Plymouth, United Kingdom: Lexington Books. Pinello, Dan. 2015. fiEdited American Appellate Cases, Adjudicating Lesbian and Gay Rights Claims , Arranged by Court. fl Accessed December 29, 2015: http://www.danpinello.com/CasesByCourt.htm . Piott, Steven A. 2003. Giving Voters a Voice: The Origins of the Initiative and Referendum in America . Columbia, MO: University of Missouri Press. Polsb y, Nelson W. 1985. Political Innovation in America: The Politics of Policy Initiation . New Haven, CT: Yale University Press. Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political -Economic History of Roll Call Voting . New York, NY: Oxford Unive rsity Press. Poole, Keith T. and Howard Rosenthal. 2007. Ideology and Congress . New Brunswick, NJ: Transaction Publishers. Popelier, Patricia. 2015. fiThe Role of Courts in Legislative Policy Diffusion and Divergence.fl The Theory and Practice of Legislati on 3 (3): 315 Œ31. Pralle, Sarah B. 2003. fiVenue Shopping, Political Strategy, and Policy Change: The Internationalization of Canadian Forest Advocacy.fl Journal of Public Policy 23(3): 233 Œ60. Primo, David M., Matthew L. Jacobsmeier, and Jeffrey Milyo. 2 007. fiEstimating the Impact of State Policies and Institutions with Mixed -Level Data.fl State Politics & Policy Quarterly 7 (4): 446 Œ59. Quilon, Amelia and Nancy Wang. fiLessons from 2018 Michigan Ballot Initiative Campaigns.fl Roundtable. Michigan chapter o f Scholars Strategy Network. East Lansing, MI. January 10, 2019. Ranney, Austin. 1976. Parties in State Politics. In Politics in the American States , 3rd ed., edited by Herbert Jacob and Kenneth Vines. Boston, MA: Little, Brown & Co. Rayside, David. 2010 . fiThe Distinctive Centrality of Marriage in the United States.fl In Same -sex Marriage in the Americas: Policy Innovation for Same -Sex Relationships by Jason Pierceson, Adriana Piatti -Crocker, and Shawn Schulenberg (Eds). Plymouth, United Kingdom: Lexington Books. Reilly, Shauna. 2010. Design, Meaning and Choice in Direct Democracy: The Influence of Petitioners and Voters . Surrey, U.K.: Ashgate Publ ishing Limited. 293 Rogers, Everett M. 1962. Diffusion of Innovations . New York, NY: Free Press of Glencoe. Rogers, Everett M. 1983. Diffusion of Innovations . 2nd ed. New York , NY : The Free Press. Rose, Richard. 1991. fiWhat Is Lesson -Drawing?fl Journal of Public Policy 11 (1): 3 Œ30. Sabatier, Paul A. 1988. fiAn Advocacy Coalition Framework of Policy Change and the Role of Policy -Oriented Learning Therein.fl Policy Sciences 21 (2/3): 129 Œ68. Sabatier, Paul A. 1998. fiThe Advocacy Coalition Framework: Revisions and Relevance for Europe.fl Journal of European Public Policy 5 (1): 98 Œ130. Sabatier, Paul A. and Hank C. Jenkins -Smith. 1993. Policy Change and Learning: An Advocacy Coalition Approach . Boulder, CO: Westview Press. Sabatier, Paul A., and Hank C. Jenkins -Smith. 1999. fiThe Advocacy Coalition Framework: An Assessment.fl In Theories of the Policy Process, Paul A. Sabatier (ed.), 117 -166. Boulder, CO: Westview Press. Salokar, Rebecca Mae. 199 7. fiBeyond Gay Rights Litigation: Using a Systematic Strategy to Effect Political Change in the United States.fl GLQ: A Journal of Lesbian and Gay Studies 3: 385 -415. Sarbaugh -Thompson, Marjorie. 2010. fiMeasuring ‚Term Limitedness™ in U.S. Multi -State Res earch.fl State Politics & Policy Quarterly 10 (2): 199 Œ217. Savage, Robert. 1985. fiWhen a Policy™s Time Has Come: Cases of Rapid Policy Diffusion, 1983 -1984.fl Publius: The Journal of Federalism 15 (3): 111 Œ26. Schattschneider, Elmer Eric. 1935. Politics, Pressure and the Tariff . New York, NY: Prentice -Hall. Seljan, E. C. and N. Weller. 2011. fiDiffusion in Direct Democracy: The Effect of Political Information on Proposals for Tax and Expenditure Limits in the U.S. States.fl State Politics & Policy Quarterly 11 (3) : 348 -368. Sellers, Mitchell D. 2017. fiGubernatorial Use of Executive Orders: Unilateral Action and Policy Adoption.fl Journal of Public Policy 37 (3): 1 -25 Sharman, Jason Campbell. 2008. fiPower and Discourse in Policy Diffusion: Anti -Money Laundering in Developing States.fl International Studies Quarterly 52 (3): 635 Œ56. Shipan, Charles R., and Craig Volden. 2006. fiBottom -up Federalism: The Diffusion of Anti - Smoking Policies from U.S. Cities to States.fl American Journal of Political Science 50 (4): 825 -843. Shipan, Charles. R., and Craig Volden. 2008. fiThe Mechanisms of Policy Diffusion.fl American Journal of Political Science 52 (4): 840 -857. 294 Shipan, Charl es R., and Craig Volden. 2012. fiPolicy Diffusion: Seven Lessons for Scholars and Practitioners.fl Public Administration Review 72 (6): 788 Œ96. Shipan, Charles R., and Craig Volden. 2014. fiWhen the Smoke Clears: Expertise, Learning and Policy Diffusion.fl Jo urnal of Public Policy 34 (3): 357-387 . Shor, Boris and Nolan McCarty. 2011. fiThe Ideological Mapping of American Legislatures.fl The American Political Science Review 105 (3): 530 Œ 551. Simmons, Beth A., Frank Dobbin, and Geoffrey Garrett. 2006. fiIntro duction: The International Diffusion of Liberalism.fl International Organization 60 (4): 781 Œ810. Simmons, B eth A. and Z achary Elkins. 2004. fiThe Globalization of Liberalization: Policy Diffusion in the International Political Economy.fl American Political Science Review 98 (1): 171 -189. Simon, Herbert. 1972. fiTheories of Bounded Rationality.fl in C. B. McGuire and R. Radner (eds.), Decision and Organization , Amsterdam: North -Holland Publishing Company. Simon, Herbert A. 1985. fiHuman Nature in Politics: The Dialogue of Psychology with Political Science.fl The American Political Science Review 79 (2): 293 Œ304. Smith, Daniel A., Matthew DeSantis, and Jason Kassel. 2006. fiSame -Sex Marriage Ballot Measures and the 2004 Presidential Election.fl State & Local Government Review 38 (2): 78 Œ91. Smith, Daniel A., and Dustin Fridkin. 2008. fiDelegating Direct Democracy: Interparty Legislative Competition and the Adoption of the Initiative in the American States.fl American Political Science Review 102 (3): 333 -350. Smith, Daniel A. and Caroline J. T olbert. 2004. Educated by Initiative: The Effects of Direct Democracy on Citizens and Political Organizations in the American States . Ann Arbor, MI: University of Michigan Press. Smith, Daniel A., and Caroline Tolbert. 2007. fiThe Instrumental and Educativ e Effects of Ballot Measures: Research on Direct Democracy in the American States.fl State Politics & Policy Quarterly 7 (4): 416 Œ45. Smith, Miriam. 2008. Political Institutions and Lesbian and Gay Rights in the United States and Canada . New York, NY: Rout ledge. Smith, Raymond A., and Donald Haider -Markel. 2002. Gay and Lesbian Americans and Political Participation: A Reference Handbook . Santa Barbara, CA: ABC -CLIO. Solomon, Marc. 2014. Winning Marriage: The Inside Story of How Same -Sex Couples Took on the Politicians and Pundits Š and Won . Lebanon, NH: ForeEdge, University Press of New England. 295 Soule, Sarah A. 2004. fiGoing to the Chapel? Same -Sex Marriage Bans in the United States, 1973 -2000.fl Social Problems 51 (4): 453 -477. Squire, P. 2007. fiMeas uring Legislative Professionalism: The Squire Index Revisited.fl State Politics & Policy Quarterly 7 (2) : 211 -227. Squire, P. 2008. fiMeasuring the Professionalization of State Courts of Last Resort.fl State Politics & Policy Quarterly 8 (3) : 223 -238. Sta rke, Peter. 2013. fiQualitative Methods for the Study of Policy Diffusion: Challenges and Available Solutions.fl Policy Studies Journal 41 (4): 561 -582. Starrs, Paul and Peter Goin. 2010. Field Guide to California Agriculture . Berkeley, CA: University of California Press. Stewart, Chuck, ed. 2015. Proud Heritage: People, Issues, and Documents of the LGBT Experience . Volume 3. Santa Barbara, CA: ABC -CLIO, LLC. Stone, Amy L. 2012. Gay Rights at the Ballot Box . Minneapolis, MN: University of Minnesota Press. Stone, Amy L. 2016. fiThe Impact of Anti -Gay Politics on the LGBTQ Movement.fl Sociology Compass 10 (6): 459 Œ67. Stone, Diane. 1999. fiLearning Lessons and Transferring Policy across Time, Space and Disciplines.fl Politics 19 (1): 51 Œ59. Strang, David. 1991. fiAdding Social Structure to Diffusion Models: An Event History Framework.fl Sociological Methods & Research 19 (3): 324 Œ53. Stream, Christopher. 1999. fiHealth Reform in the States: A Model of State Small Group Health Insurance Market Reforms.fl Political Research Quarterly 52 (3): 499 -525 . Tarrow, Sidney. 2005. The New Transnational Activism . Cambridge, UK: Cambridge University Press. Taylor, Jami K., Donald P. Haider -Markel, and Benjamin Rogers. 2019. fiToward a New Measure of State -Level LGBT Interest Group Strength.fl State Politics & Policy Quarterly 19 (3): 334-350 . Taylor, Jami K., Dan iel C. Lewis, Matthew L. Jacobsmeier, and Brian DiSarro. 2012. fiContent and Complexity in Policy Reinvention and Diffusion: Gay and Transgender -Inclusive Laws against Discrimination.fl State Politics & Policy Quarterly 12 (1): 75 Œ98. Teodoro, Manuel P. 200 9. fiBureaucratic Job Mobility and the Diffusion of Innovations.fl American Journal of Political Science 53 (1): 175 Œ89. Thomas, Clive S., and Ronald J. Hrebenar. 2008. fiUnderstanding Interest Groups, Lobbying and Lobbyists in Developing Democracies.fl Journ al of Public Affairs 8 (1 Œ2): 1 Œ14. 296 Thompson, Mark, ed. 1994. Long Road to Freedom: The Advocate History of the Gay and Lesbian Movement . 1st ed. Stonewall Inn Editions. New York, NY: St. Martin™s Press. True, Jacqui, and Michael Mintrom. 2001. fiTransnational Networks and Policy Diffusion: The Case of Gender Mainstreaming.fl International Studies Quarterly 45 (1): 27 Œ57. Tversky, Amos, and Daniel Kahneman. 1974. fiJudgment under Uncertainty: Heuristics and Biases.fl Science 185 (4157): 1124 Œ31. US Census Bureau. Intercensal Estimates of the Resident Population by Single Year of Age and Sex for States and the United States: April 1, 2000 to July 1, 2010. Accessed at: https://www2.census.gov/programs -surveys/popest/datasets/2000 -2010/intercensal/state/st -est00int -agesex.csv . US Census Bureau. National, State, and Puerto Rico Commonwealt h Totals Datasets: Population, population change, and estimated components of population change: April 1, 2010 to July 1, 2016. Accessed at: https://www2.census.gov/programs -surveys/popest/datasets/2010 -2016/national/totals/nst -est2016 -alldata.csv . US Census Bureau. State and County Intercensal Tables: 1990 -2000. Accessed at: https://www2.census.gov/programs -surveys/popest/tables/1990 -2000/intercensal/st -co/coest2001 -12 -00.pdf . Valelly, Richard M. 2012. fiLGBT Politics and American Political Development.fl Annual Rev iew of Political Science 15 (1): 313 Œ32. Verba, Sidney, Kay Lehman Schlozman, and Henry E. Brady. 1995. Voice and Equality: Civic Voluntarism in American Politics . Cambridge, MA: Harvard University Press. Volden, Craig. 2002. fiThe Politics of Competitive Federalism: A Race to the Bottom in Welfare Benefits?fl American Journal of Political Science 46 (2): 352 Œ63. Volden, Craig. 2006. fiStates as Policy Laboratories: Emulating Success in the Children™s Health Insurance Program.fl American Journal of Political Science 50 (2): 294 -312. Volden, Craig. 2015. fiFailures Diffusion, Learning, and Policy Abandonment.fl State Politics & Policy Quarterly 16 (1): 44 - 77. Volden, Craig, Michael M. Ting, and Daniel P. Carp enter. 2008. fiA Formal Model of Learning and Policy Diffusion.fl The American Political Science Review 102 (3): 319 -332. Wald, Kenneth D., James W. Button, and Barbara A. Rienzo. 1996. fiThe Politics of Gay Rights in American Communities: Explaining Antidis crimination Ordinances and Policies.fl American Journal of Political Science 40 (4): 1152 -1178. 297 Walker, Jack L. 1969. fiThe Diffusion of Innovations among the American States.fl American Political Science Review 63 (3): 880 -899. Wall, Audrey S. ed. 2008. fiGeneral Information on State Constitutions.fl The Book of the States . Lexington, KY: Council of State Governments. Waters, M. Dane. 2003. Initiative and Referendum Almanac . Durham, NC: Carolina Academic Press. Welch, Susan, and Kay Thompson. 1980. fiThe Impact of Federal Incentives on State Policy Innovation.fl American Journal of Political Science 24 (4): 715 Œ29. Werum, Regina, and Bill Winders. 2001. fiWho™s ‚In™ and Who™s ‚Out™: State Fragmentation and the Struggle over Gay Rig hts, 1974 Œ1999.fl Social Problems 48 (3): 386 Œ410. Weyland, Kurt. 2005. fiTheories of Policy Diffusion: Lessons from Latin American Pension Reform.fl World Politics 57 (2): 262 Œ95. Weyland, Kurt. 2006. Bounded Rationality and Policy Diffusion: Social Sector Reform in Latin America . Princeton, NJ: Princeton University Press. Wilkerson, John, David Smith, and Nicholas Stramp. 2015. fiTracing the Flow of Policy Ideas in Legislatures: A Text Reuse Approach.fl American Journal of Political Science 59 (4): 94 3Œ56. Williams, Richard. 2012. fiUsing the Margins Command to Estimate and Interpret Adjusted Predictions and Marginal Effects.fl The Stata Journal 12 (2): 308-331. Windett, Jason H., Jeffrey J. Harden, and Matthew E. K. Hall. 2015. fiEstimating Dynamic Ide al Points for State Supreme Courts.fl Political Analysis 23 (3): 461 Œ69. Wolak, Jennifer, Adam J. Newmark, Todd McNoldy, David Lowery, and Virginia Gray. 2002. fiMuch of Politics is Still Local: Multi -State Lobbying in State Interest Communities.fl Legislati ve Studies Quarterly 27 (4): 527 -555. Workman, Samuel, Bryan D. Jones, and Ashley E. Jochim. 2009. fiInformation Processing and Policy Dynamics.fl Policy Studies Journal 37 (1): 75 Œ92. Wright, Gerald C., Jr., Robert S. Erikson, and John P. McIver. 1987. fiPublic Opinion and Policy Liberalism in the American States.fl American Journal of Political Science 31 (4): 980 Œ1001. Yates, J., H. Tankersley, and P. Brace. 2010. fiAssessing the Impact of State Judicial Structures on Citizen Litigiousness.fl Political Research Quarterly 63 (4): 796 Œ810. Zelizer, Adam. 2019. fiIs Position -Taking Contagious? Evidence of Cue -Taking from Two Field Experiments in a State Legislature.fl American Political Scien ce Review 1Œ13.