Policy Diffusion Dynamics in America

Policy Diffusion Dynamics in America Policy Diffusion Dynamics in America integrates research from agenda setting and ...

Author: Graeme Boushey

9 downloads 1024 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Policy Diffusion Dynamics in America

Policy Diffusion Dynamics in America integrates research from agenda setting and epidemiology to model factors that shape the speed and scope of public policy diffusion. Drawing on a data set of more than 130 policy innovations, the research demonstrates that the “laboratories of democracy” metaphor for incremental policy evaluation and emulation is insufficient to capture the dynamic process of policy diffusion in America. A significant subset of innovations triggers outbreaks – the extremely rapid adoption of innovation across states. The book demonstrates how variation in the characteristics of policies, the political and institutional traits of states, and differences among interestgroup carriers interact to produce distinct patterns of policy diffusion. Graeme Boushey is Robert Wood Johnson Scholar in Health Policy Research at the University of Michigan, on leave from his post as Assistant Professor of Political Science at San Francisco State University. His teaching and research are organized around practical and theoretical questions of state and federal policy making. He recently coauthored a review of individual and organizational decision making for the Handbook of Public Policy, and he also has coauthored an article in the Journal of Comparative Policy Analysis on immigration policy in federations.


GRAEME BOUSHEY San Francisco State University

cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao ˜ Paulo, Delhi, Dubai, Tokyo, Mexico City Cambridge University Press 32 Avenue of the Americas, New York, NY 10013-2473, USA www.cambridge.org Information on this title: www.cambridge.org/9780521762816 © Graeme Boushey 2010 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2010 Printed in the United States of America A catalog record for this publication is available from the British Library. Library of Congress Cataloging in Publication data Boushey, Graeme. Policy diffusion dynamics in America / Graeme T. Boushey. p. cm. Includes bibliographical references and index. ISBN 978-0-521-76281-6 (hardback) 1. Policy sciences. 2. Diffusion of innovations – Political aspects – United States. JK468.P64B73 2010 320.60973–dc22 2010005435

I. Title.

ISBN 978-0-521-76281-6 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is, or will remain, accurate or appropriate.

To my parents, Homer and Virginia Boushey

Contents

List of Figures List of Tables Acknowledgments 1

Contagion in the Laboratories of Democracy

2

Incrementalism and Policy Outbreaks in the American States Policy Agents: Innovation Attributes and Diffusion Dynamics Innovation Hosts: State Characteristics and Diffusion Dynamics

3 4 5 6

Policy Vectors: Interest Groups and Diffusion Dynamics Conclusion

page ix xi xiii 1 22 62 92 139 169

Appendix A: List of Innovations Collected Appendix B: Policies Collected by Historical Era

187 193

Appendix C: Innovations Collected by Policy Type and Target Appendix D: State Receptivity to Innovation Ranked by Policy Type References Index

197 201 205 215

vii

Figures

1.1. 1.2. 2.1.

The epidemiologic triad of disease. page 11 The epidemiologic framework of innovation diffusion. 14 Distinct patterns of policy diffusion: The death penalty and state lotteries. 27 2.2. Distinct patterns of policy diffusion: Charter schools and the Amber Alert. 27 2.3. Episodic patterns of policy diffusion: Gubernatorial term limits and English Only legislation. 28 2.4. Newell’s bands of rationality. 35 2.5. S-shaped adoption curve. 39 2.6. S-shaped adoption curves representing three different rates of innovation diffusion. 43 2.7. R-shaped exponential adoption curve. 44 2.8. Simulated theoretical diffusion curves. 48 2.9. Cumulative distribution of adoption times: All policies. 56 2.10. Cumulative distribution of adoption times: 1900–1929. 57 2.11. Cumulative distribution of adoption times: 1930–1959. 58 2.12. Cumulative distribution of adoption times: 1960–2006. 59 3.1. Cumulative distribution of adoption times: All policies. 83 3.2. Cumulative distribution of adoption times: Governance policy. 84 3.3. Cumulative distribution of adoption times: Morality policy. 85 3.4. Cumulative distribution of adoption times: Regulatory policy. 86 3.5. Cumulative distribution of adoption times: Children’s policy. 87 ix

Figures

x

3.6. 4.1. 4.2. 4.3. 4.4.

Cumulative distribution of adoption times: Licensing policy. Map of state receptivity to innovation, 1960–2006. Map of state receptivity to morality policy innovation, 1960–2006. Map of state receptivity to regulatory policy innovation, 1960–2006. Map of state receptivity to governance policy innovation, 1960–2006.

88 102 121 122 123

Tables

2.1. 3.1. 4.1. 4.2. 4.3. 4.4. 4.5.

Statistical Tests for Normality by Historical Era Statistical Tests for Normality by Policy Type State Receptivity to Innovation, 1960–2006 State Receptivity to Innovation by Historical Era State Predictors of Innovation Receptivity, 1960–2006 State Receptivity to Innovation by Policy Type Predictors of State Receptivity to Regulatory, Morality, and Governance Policy, 1960–2006 (Baseline Model) 4.6. Predictors of State Receptivity to Regulatory, Morality, and Governance Policy, 1960–2006 (Full Model) 5.1. Interest Group Variation in Organization and Strategic Behavior

page 56 83 101 104 114 119 131 132 150

xi

Acknowledgments

This work explores the diffusion of public policy innovations in the United States. I became interested in the topic as a graduate student at the University of Washington, where I was introduced to research on policy making in federations. Although I appreciated the purported benefits of decentralized policy making, I could not reconcile ideal models of innovation and diffusion with my experiences growing up in California, where many of the prominent policies that the state adopted in the 1990s did not match the neat, cost-benefit decision-making processes outlined by researchers of an earlier generation. As I began to read research on agenda setting, it became clear that the process of public policy innovation and diffusion is dynamic, one in which incremental decision making is often interrupted by sudden moments of attention-driven policy change. I became interested in identifying what determines the pattern of innovation diffusion, whether by gradual increments or by sudden outbreak. In pursuing this interest, I could not help but notice that a similar pattern emerged in my own work – long periods of gradual improvement interrupted by new insights and sudden productivity. Unlike models of the policy process, the causes of these breakthroughs are easy to explain. They came after meetings and conversations with those friends and colleagues who graciously showed an interest in this project, and who took the time to offer suggestions for improvement. I wish to thank each of them for their attention and support. I am fortunate to have worked with an extraordinary group of people at the University of Washington, and thank each of them for the time and energy they invested in the development of this research. This book xiii

xiv

Acknowledgments

would not have been possible without the mentorship of my committee, Bryan Jones, Mark Smith, and Erik Wibbels. Bryan Jones encouraged my interdisciplinary approach to modeling diffusion dynamics and provided feedback and support for this research from its conception. My interest in policy diffusion grew out of challenging discussions with Mark Smith and Erik Wibbels, whose respective mastery of the literature on interest groups and federalism provided an invaluable resource as I began to explore the questions of why and how innovations spread across states. A number of others provided key advice at various stages of the project. Peter May volunteered feedback as I began to explore how the initiative process shaped the diffusion of innovations. I benefited enormously from the friendship and advice of David Olson, whose knowledge of state and local politics is limitless. Anne Ganley provided insight on how to organize and execute a major research project. John Ahlquist, Christian Breunig, and Josh Sapotichne were valuable critics and good friends. They pushed me to expand my conceptualization of the processes leading to innovation diffusion and often took time from their own research to help me work through various technical challenges that emerged as I worked on this project. I was surrounded by a group of colleagues and friends who made it a joy to study in the Department of Political Science at the University of Washington. Chris Koski, Rose Ernst, Sebastien Lazardeux, Ashley Jochim, Michelle Wolfe, and Samuel Workman each provided encouragement and assistance at important stages of this project. My thinking on public policy diffusion grew better from the exchanges we had over coffee in Gowen and Smith Halls. A number of others provided important comments on various stages of this research. Andy Karch read one of the first drafts of Chapter 2 and later provided feedback when I presented a completed manuscript at the University of Texas–Austin. Frank Baumgartner – a coauthor and former student of Jack Walker – made valuable suggestions that helped clarify how interest groups influence the process of policy diffusion. Frances Berry, John Fulwider, Michael Mintrom, Christopher Mooney, Karen Mossberger, Craig Volden, and Dick Winters offered criticism and comments at various panels over the past few years. I would also like to thank seminar participants at the University of Texas–Austin, who read the manuscript and provided lively feedback during a workshop and panel discussion.

Acknowledgments

xv

I was welcomed to San Francisco by a wonderful community of scholars. Max Neiman, of the Public Policy Institute of California, took an early interest in this project and graciously gave detailed comments on an entire draft of the manuscript. Richard DeLeon commented on a full draft, adding the perspective of a scholar who has made a career of studying the innovative politics of San Francisco. Jesse Cohen provided technical support for ArcGIS and helped produce the maps that appear in Chapter 4. The faculty and students in the political science department at San Francisco State University were supportive audiences as I worked on this research. I especially want to thank the students in my graduate seminar in American politics, whose comments on the manuscript gave me a fresh perspective as I neared completion. I am grateful to the School of Behavioral and Social Science and the Office of Faculty Affairs and Professional Development at San Francisco State University for their support of this research. Lew Bateman of Cambridge University Press has been a supportive editor and a valuable critic. He skillfully kept the project moving, providing important feedback at each stage of the review and revision process. I am in debt to the anonymous reviewers for their generous and useful comments. Each clearly invested a great deal of time and effort reviewing the manuscript, and I am certain that responding to their concerns improved this book. Their comments were careful and comprehensive and not only highlighted issues and ideas in need of improvement, but also provided concrete and useful directions that made revisions much easier. Any errors or omissions that remain are entirely my own. I am blessed to have a family that has provided unwavering support over the years that I have worked on this research. As I collected data for this project, my parents Homer and Virginia Boushey became active students of policy innovation and diffusion. My mother forwarded newspaper clippings about interesting new policies and the problematic legacy of California’s initiative process. My siblings Geoff and Sarah Boushey were close confidants when I became excited by a new idea or frustrated by a setback. My father, a professor of medicine at the University of California San Francisco, was always willing to read drafts of my manuscript, and provided perspective on examples I chose from studies of epidemiology. Finally, I would like to thank my wife, Sara Levine, whose friendship, support, encouragement, and patience have sustained me as I have worked

xvi

Acknowledgments

late evenings and long weekends on this project. I always found it easier to return to writing after the long hikes we took on the coast or in the redwood forests of Northern California, where I gained perspective in the calm that comes during a long walk with a good friend. I cannot imagine completing this book without her extraordinary love and support.

1 Contagion in the Laboratories of Democracy

In July of 1997, Dallas area child protection activists appealed to local police and media broadcasters to launch the nation’s first Amber Alert system, a crime prevention program enabling law enforcement agencies to activate regional emergency broadcast systems to announce missing children alerts.1 From these origins, the Amber Alert system evolved into one of the most successful interstate innovation campaigns in recent history. With strong support from child-protection and victim’s-rights advocates, every state in the union adopted the Amber plan between 1999 and 2005.2 The Amber Alert proved to be such an appealing response to kidnapping that identical versions of the child protection law were soon adopted internationally. Between 2002 and 2004, every Canadian province adopted the Amber program.3 In 2006, the United Kingdom launched its own version of the Amber plan called the Child Rescue Alert.4 1

2 3

4

Demands for the Amber Alert grew out of local outrage following the brutal kidnapping and murder of nine-year-old Amber Hagerman in 1996. Although a neighbor had witnessed the child’s kidnapping and contacted the police with a description of the vehicle, there was no way to broadcast the event to the broader public. For a brief history, see http://www.iowabroadcasters.com/ambrhist.htm; accessed August 2007. Oklahoma became the first state to adopt the Amber Alert in 1999. By 2003, the Amber Alert had been adopted by every state save Alaska and Hawaii. In the United States, Amber legislation stands for America’s Missing Broadcast Emergency Response. The Amber Alert legislation is therefore both a memorial tribute to Amber Hagerman and a description of the program. Interestingly, Canadian provincial Amber plans retained the tribute to Amber Hagerman in its legislation, speaking to the power of the image associated with the policy innovation. A summary of the efforts to internationalize Amber Alert legislation can be found on the website for the Center for Missing and Exploited Children www.missingkids.com; accessed August 2007.

1

2


Although the Amber Alert was exceptional in the sheer speed and scope of its implementation, such abrupt patterns of policy adoption are far from unique in American politics. The reenactment of the death penalty, prohibition, term limits, tax revolts, state auto lemon laws, English Only language legislation, “three strikes” sentencing guidelines, mandatory child auto-restraint requirements, and sex-offender registries stand as prominent examples of policy innovations that moved rapidly and extensively throughout the nation. Most of these innovations were championed by well-organized interest groups, and appealed broadly to voters across the states. In many cases, the innovation was adopted by more than 30 states in fewer than six years.5 In other cases, innovation spread suddenly over a subset of states before abruptly stopping. The sudden and rapid diffusion of innovations challenges traditional conceptions of policy making in the United States. Students of American government argue that federalism should exert a conservative pressure against rapid policy change.6 The implementation of identical public policies across states should be slowed by the multiple veto points of policy making in a federation, because innovation adoption requires an independent legislative decision by 50 state governments. Yet as the Amber Alert demonstrates, new innovations can and do spark positive feedback cycles leading to the sudden implementation of identical policies across states. Although such rapid standardization of state policies is often stimulated by intervention of the federal government through grants and other inducements,7 there is little evidence to suggest that rapid diffusion depends on the power and resources of the national government. In the case of the Amber Alert, 32 states had adopted the program before the federal government passed enabling legislation providing grants for state Amber Alert programs.8 In the case of the term-limitation movement, during which government reform activists imposed strict legislative term 5

6

7

8

This requirement for the scope and speed standard for unusually rapid diffusion was proposed by Savage (1985a) in his study of the rapid diffusion of public policies whose “time has come” (111). Baumgartner and Jones (1993; 2005) provide a thorough review and critique of models of policy change in federations. For a summary, see Agendas and Instability in American Politics, Chapter 11. National Interaction models of public-policy diffusion explore how federal intervention shapes public-policy diffusion. For a recent study of national interaction effects in policy diffusion, see Andy Karch’s “National Intervention and the Diffusion of Policy Innovations.” American Politics Research 34(4): 403–426 (2006). In 2003, the same year the federal government passed legislation to fund Amber Alert programs across the states, an additional 15 states enacted Amber Alert programs.


3

limits on politicians across 20 state legislatures through the first half of the 1990s, interstate policy diffusion occurred absent the involvement of either the federal or state governments.9 Surprisingly, the rapid and sudden adoption of innovations across states is not well explained by extant studies of policy diffusion – the formal study of how ideas move from one jurisdiction to another in federations. Political scientists have generally explained policy diffusion as resulting from a process of incremental political learning by state governments (Walker 1969; Gray 1973; F. Berry and Berry 1990). The diffusion of innovations occurs through the “science of muddeling through” (Lindblom 1959), as government officials identify and emulate those policy innovations that present convenient or popular solutions to existing social or economic problems (Walker 1969; F. Berry and Berry 1999; Volden 2006). In their most common form, theories of public-policy diffusion anticipate that state decision makers identify policy problems and policy goals; engage in a limited solution search by exploring the policy solutions of peer jurisdictions; evaluate competing policy experiments for their efficacy; and, finally, select the “best” available policy solution. Diffusion research therefore gives primacy to the decision making of formal elected and appointed officials in state government, who identify, evaluate, and adopt emerging innovations that meet the challenges presented by interstate economic competition or address pressing social policy problems. Current research in state policy diffusion overlooks the causes of varying rates of innovation diffusion. Whereas the earliest studies in policy diffusion assumed an expressly comparative orientation to the study of policy innovation and adoption,10 modern research has assumed a narrower approach to documenting the processes leading to public-policy

9

10

The diffusion of state legislative term limitations overcame the significant opposition of elected representatives in state governments, who were reluctant to vote themselves out of office. A number of states (MA, WA, OR, ID, UT, WY) later repealed their laws. The two articles responsible for focusing political science research on the diffusion of innovations adopted a comparative approach to the study of policy innovation and adoption. Walker’s groundbreaking article, “The Diffusion of Innovations Among the American States,” (1969) explored general patterns of policy adoption across 88 distinct innovations that diffused across the states. Gray’s (1973) “Innovation in the States: A Diffusion Study” compared temporal and spatial patterns of policies across a range of different issue areas. These two articles sparked an important debate about the validity of generalizations drawn from comparative research on policy diffusion that continues to shape diffusion research today.

4


diffusion.11 The modern standard in diffusion research focuses on single case studies that are used to document the political and decision-making processes underlying a demonstrative case of innovation diffusion.12 This perspective fails to capture the complexity of policy making in American federalism. Often, a series of states adopt nearly identical policies in a very short time frame, suggesting decision making driven by sudden policy imitation rather than gradual incremental learning. Just as importantly, innovations often spread through the channels of direct democracy, beyond the direct control of state legislatures and without the input of bureaucrats or elected officials. Finally, diffusion research has understated the role of nongovernmental actors in policy diffusion. The diffusion of innovations is driven not simply by sequential emulation across state governments, but rather by carefully orchestrated pressure campaigns of organized interests that strategically work to see policies adopted in as many states as is feasible. The term-limitation movement of the 1990s demonstrates this dynamic. Term-limit activists operated outside of state legislatures and were uninterested in evaluating the impact that term limits would have on the future quality of state policy making. Term-limit activists instead coordinated initiative campaigns to push for governmental reform in as many states in as short a time as possible. The diffusion of state term limits shows little evidence of the incremental learning process familiar to state politics researchers. Perhaps because most research anticipates incremental learning as a centerpiece of the diffusion process, many of the most prominent and compelling cases of policy diffusion in recent U.S. history simply do not conform to the existing theoretical frameworks for how ideas move from one state to another. Although research in state policy diffusion has produced excellent descriptive studies of how economic competition or social-policy learning lead to innovation diffusion across state governments, this approach has been insensitive to the political and 11

12

This standard is partially shaped by the limitations of the event history models currently favored in innovation and diffusion research. These logistic time-series models permit researchers to model how changes in state internal dynamics and interstate interactions increase a state’s probability of adopting a single innovation at a given time. Perhaps the most widely known and cited piece of recent research in policy diffusion is a detailed study of how economic competition and geographic proximity spurred the diffusion of state lottery programs across states from 1960–1987 (F. Berry and Berry 1990). The popularity of the article is in no small part due to its groundbreaking introduction of the event history framework for diffusion research. However, it is telling that this preeminent piece of research on interstate policy diffusion addresses the gradual diffusion of a significant but not highly salient economic policy innovation.


5

decision-making processes leading to both incremental policy adjustment and sweeping policy change. Diffusion research currently provides no framework to distinguish between the processes leading to the sudden diffusion of innovations like the Amber Alert, the gradual and steady diffusion of innovations like state lottery programs, and the episodic and periodic diffusion of public policies, such as term limits. To understand why policies follow such remarkably different temporal and spatial patterns of diffusion requires a new and explicitly comparative approach to the study of diffusion – one which moves away from single case studies and instead studies factors leading to variation in patterns of diffusion. If states truly are “laboratories of democracy,” then diffusion research must begin to account for the important causes of variation in the contagion and virulence of innovations that lead to such different patterns of policy diffusion. Overview of Research This book explores the underlying causes of diffusion dynamics – the processes underlying the stable, gradual diffusion of innovations over time and the sudden policy shocks precipitating positive feedback cycles and rapid policy mimicking across states. It advances an epidemiologic framework to understand the factors leading to variations in the rate of innovation diffusion, the relative susceptibility and immunity of states to innovation, and the critical role that interest groups play as carriers or vectors of innovations from one state to another. To understand the causes of diffusion dynamics, this project addresses two primary areas of inquiry. First, it updates the behavioral model of political decision making underlying the distinct patterns of incremental and nonincremental policy diffusion. In so doing, it distinguishes between the decision-making processes leading to gradual policy emulation and the pressures leading to sudden policy imitation. Second, it explores the characteristics that propel certain innovations across some states much more rapidly than others. In addressing these two areas of inquiry, this book provides a framework for the study of the contagion and virulence of innovations, and advances a theory for understanding the causes of policy outbreaks – a process characterized by a positive feedback cycle leading to the extremely rapid adoption of policy innovation across states. The book begins by generating a theoretical and empirical critique of theories of incrementalism in public-policy diffusion. The first section demonstrates that the popular model of incremental decision making

6


provides only a partial understanding of the behaviors leading to innovation diffusion. To model the joint processes of gradual policy change and sudden positive feedback cycles, this section develops an agenda-setting model of attention-driven political choice to capture the decision-making processes leading to diffusion dynamics. The second section of the book identifies factors leading to both positive feedback cycles and incremental patterns of diffusion. This section borrows from the study of epidemiology to conceptualize the distinct factors leading to the diffusion of innovations across states. This study argues that epidemiology can serve as a useful guide for the study of innovation diffusion. Policy innovations are the specific agents that are being transmitted through the population of states. States are the susceptible hosts that can adopt innovation. Finally, interest groups are the carriers or vectors that transmit policies from one state to another. A model of diffusion dynamics cannot be built around a single causal factor or process, but rather must account for how variation in the agents, carriers, and hosts of innovation shapes the process of diffusion. In three separate chapters, this research develops how systematic variation in the characteristics of policy innovations, the political and institutional traits of states, and differences among interest-group carriers all contribute to nonincremental patterns of policy diffusion. These two stages of the project are closely connected, and taken together the epidemiologic framework yields considerable insight into learning and decision making in federal systems. Diffusion dynamics are shaped by variations in the interactions of individual policies, state sociopolitical institutions, and interest-group organizations to produce different decision-making responses to policy innovations in the federal system. Different policy ideas produce nonincremental patterns of policy diffusion because they affect distinct decision-making processes by state decision makers, by elevating issue salience and arousing a sense of urgency, or by limiting issue salience and encouraging satisficing – a decision-making shortcut in which decision makers adopt the first available solution that is “good enough.” Differences in state receptivity to innovations are shaped by variations in the political and institutional capacities of state governments to process simple or complex political information. States are not uniformly receptive to all forms of innovation. Instead, variation in state political and institutional attributes makes them systematically more or less prone to adopt different forms of innovation. Finally, the interest groups that act as carriers or vectors


7

of innovation produce drastically different patterns of diffusion, in part because they adopt different strategies when organizing pressure campaigns for innovation adoption. Variations in both the resources and the rhetorical strategies of interest groups agitating for innovation influence how state decision makers respond to calls for policy change. Political Decision Making and the Diffusion of Innovations The idea that policy change results from two distinct decision-making processes has recently gained traction in public-policy studies. Founders of an important research program documenting policy dynamics in American politics, Bryan Jones and Frank Baumgartner (1993; 2002; 2005) observe that “dramatic policy change occurs regularly in American politics, even if most issues most of the time are characterized by routine developments” (2002, 1). Periods of policy stasis and dramatic policy change are caused by changes in the allocation of government attention (B. Jones and Baumgartner 2005). Incremental policy change occurs through a negative feedback process, as risk-averse decision makers operating under severe time constraints make marginal adjustments to policy regimes in order to maintain the status quo. Sudden and dramatic policy change occurs through positive feedback cycles, as an event focuses mass political attention to a specific issue area, leading to increased demands and support for dramatic policy change.13 A growing body of policy research confirms the dynamics of negative and positive feedback cycles in policy making across an impressive array of American political institutions. Research in presidential decision making (Larsen 2006), congressional attention (Baumgartner and Jones 1993), and state budgeting (Koski and Breunig 2006), has demonstrated that policy making in American political institutions displays both extended periods of policy stasis and sudden moments of policy change. Similar dynamics have been documented in policy areas as diverse as gun laws (True and Utter 2002), crime control legislation (A. Schneider 2006), and environmental and nuclear energy policy (Baumgartner and Jones 1993). Policy change in each of these issue areas has occurred

13

A focusing event can be produced by a number of different factors, ranging from a natural catastrophe, an alarming shift in accepted policy indicators, or increased media attention on a policy problem. It need not be an exogenous shock to the political system.

8


through long periods of gradual policy adjustment interrupted by dramatic moments of sweeping policy reform.14 Despite anecdotal evidence that the process of policy diffusion is likewise prone to positive and negative feedback cycles, research has yet to connect the process of public-policy diffusion to the decision-making processes documented in the study of policy dynamics. To explain the causes of diffusion dynamics, this book begins by connecting the behavioral model underlying the larger study of policy dynamics to the decisionmaking processes leading to the diffusion of innovations. As with other research linking agenda setting to policy outputs, the processes of policy diffusion cannot be explained through a single static decision-making model, but rather must account for attention-driven pressures leading to both incremental policy adjustments and sudden nonincremental policy outbreaks. Chapter 2 develops an agenda-setting model of public-policy diffusion to account for attention-driven pressures leading to both gradual incremental diffusion and policy outbreaks. Drawing on research in individual and organizational decision making, this analysis demonstrates that differences in diffusion dynamics occur because state decision makers prioritize information differently based on issue salience, perceived importance, and issue complexity. State political institutions disproportionately respond to innovations that activate emotional considerations or address relevant and highly salient issues. They give scant attention to issues with less immediacy. Such disproportionate information processing by decision makers across the federation leads to different diffusion dynamics, and can be used to reconcile models of incrementalism with policy outbreaks. The incremental model of information processing proves accurate for a large subset of innovations that encourage neither elevated issue attention nor a sense of urgency; however, in certain instances innovations channel mass political attention leading to mass policy mimicking. In these cases, diffusion occurs as a positive feedback cycle, absent any form of instrumental program evaluation familiar to incremental models of diffusion. Chapter 2 introduces a stochastic process model to empirically evaluate the degree of incrementalism in the diffusion of innovations. This

14

Similar decision-making dynamics have been identified by a number of researchers studying individual and systemic decision making. For example, most social customs spread gradually through populations; however in social fads, many members of a group imitate a behavior nearly simultaneously.


9

chapter introduces a unique data set measuring the state years of adoption for 133 policy innovations covering a wide range of innovations from the nineteenth, twentieth, and twenty-first centuries. This data set is used to evaluate general patterns of policy adoption in state policy making, asking specifically whether patterns of policy diffusion can be characterized as resulting from a process of incremental or nonincremental political decision making.15 The findings presented in Chapter 2 reveal that policy diffusion displays punctuated dynamics that are inconsistent with incremental policy learning and emulation. Instead, policy diffusion has occurred more rapidly than expected in incremental learning models, indicating a process of incrementalism interrupted by sweeping policy outbreaks. Modeling Diffusion Dynamics: Are Public Policies Some Kind of Disease? Taken by itself, analysis of models of decision making can provide only cursory insight into the causes of diffusion dynamics in American federalism. The distinct patterns of policy diffusion instead suggest some compelling questions about the process driving the diffusion of innovation. Why do some policy innovations spread much more rapidly than others? Why are some states receptive to certain forms of innovation when others appear policy-resistant to even the most popular state-level reforms? How does the involvement of nongovernmental actors shape the diffusion of innovations from one state to another? Resolving each of these questions leads to a greater understanding of the dynamics underlying the diffusion of innovations. To model the causes of diffusion dynamics, the second section of this book conceptualizes the diffusion of innovations from the perspective of epidemiology, a discipline expressly dedicated to evaluating how changes in the environment, the virulence of agents, the behavior of vectors, and the attributes of susceptible and resistant hosts interact to shape the distribution and determinants of disease.16 The basic approach of epidemiologic research encourages comparison to the study of the diffusion of 15

16

Chapters 2 and 3 rely on distributional analysis to compare empirical patterns of policy diffusion to simulated patterns associated with incremental diffusion. A more detailed discussion of the approach is described in each of these chapters. Although there are clear limits to comparing the diffusion of innovations to the communication of disease, it is worth noting that diffusion researchers have long drawn inspiration from epidemiology to describe the underlying processes of diffusion. The

10


innovations. Whereas the study of policy diffusion is focused on understanding the determinants and distribution of policy ideas and innovations across policy-making jurisdictions, epidemiologists have focused on understanding both the distribution and determinants of “health-related states or events in specified human populations” (Last 2001, 62). The epidemiologic framework is especially appropriate for evaluating the incremental and nonincremental diffusion dynamics of central interest in this book. Epidemiologists have explored factors contributing to the speed and scope of outbreaks over time: the comparative virulence of the causes of disease (bacterium, virus, toxin, etc.); the distribution and activity of the carriers of “vectors” transmitting the pathogenic agent; and the susceptibility of the populations exposed (see Text Box 1.1). These areas of inquiry are similar to studies of public-policy diffusion that have explored how the internal dynamics of states make them more or less susceptible to innovation (Walker 1969; Savage 1978; Canon and Baum 1981; Carter and LaPlant 1997); how the distribution, activity, and interactions of interest-group carriers shape the diffusion of innovation (Gray 1973; Mintrom 1997); or how changes in the policy idea itself can lead to the sudden spread of policy innovation (Savage 1985a; Mooney and Lee 1995). Importantly, the epidemiologic framework encourages researchers to move away from descriptive studies documenting individual policy diffusion and toward new questions about comparative diffusion dynamics and the joint processes of incremental and nonincremental policy change. Mapping the Diffusion of Disease in Epidemiology Figure 1.1 shows how public health researchers explore variation in each of four general factors to understand the dynamics of disease in human populations. A researcher interested in modeling the incidence, severity, and rapidity of transmission of disease in a population must account for change in environmental conditions, the characteristics of the carriers or vectors of disease, the genetic or behavioral traits of the host determining susceptibility to infection, and the unique attributes of the agent.

event history models currently favored in diffusion research were pioneered by epidemiologists interested in understanding when and why individuals in a community succumb to illness. In the emerging discipline of memetics, researchers argue that cultural norms and common policy ideas may actually replicate and spread like viruses (Dawkins 1989; Aunger 2002).


11

Host Characteristics

Environment

Agent Characteristics

Vector Characteristics

figure 1.1. The epidemiologic triad of disease. Source: Gordis, Leon. 2004. Epidemiology. Philadelphia: Elsevier Saunders, p. 16.

Of course, the study of public health processes is complex. Although health scientists organize their research to isolate and evaluate each of these elements of epidemiology, these factors interact to produce distinct patterns of disease. For example, the introduction of a new strain of the influenza virus in a dense urban setting can lead to a sudden outbreak of flu, as the denser population exposes a greater number of susceptible individuals to the more virulent strain of the disease. Whereas introduction of the same strain into a sparsely settled rural population may result in only a few isolated cases, with the agent most often “dying out” before it can be transmitted. Text Box 1.1 illustrates how an epidemiologist might organize research into each factor of disease dynamics by discussing a hypothetical emergence of a new, dangerous form of the malaria parasite in the South. The Epidemiologic Framework of Interstate Policy Diffusion The epidemiologic triad maps nicely as an organizing framework for the study of policy diffusion. A survey of recent research in the diffusion of policy innovations suggests the epidemiologic framework can be applied to isolate and classify factors contributing to the diffusion of policy innovations across jurisdictions. Researchers interested in how state geographic proximity (F. Berry and Berry 1990; Rincke 2004) or media agenda setting and public opinion (Hays 1996) shape patterns of diffusion have focused on how changes in the political environment spur policy

12


text box 1.1 Thinking about Diffusion: Tracing an Outbreak of Malaria Consider the potential causes of an outbreak of malaria in the United States. An epidemiologist would recognize that the epidemic could be spurred from a number of factors, including a change toward environmental conditions more favorable to malaria, a change in the carrier of the malaria parasite, a shift in the virulence of the malaria parasite, or a shift in the genetics or behaviors of the human population threatened by disease. The researcher would begin by exploring the many well-known environmental factors that have long been associated with the spread of malaria. In the 1940s and 1950s, the Centers for Disease Control and Prevention (known at that time as the “Communicable Disease Center,”) (CDC) eradicated malaria by eliminating favorable environmental conditions for the breeding of mosquitoes, first draining stagnant water pools where the Anopheles mosquito breeds and then spraying pesticide and petroleum to exterminate mosquito larvae (CDC 2004). As a first step, researchers interested in the resurgence of malaria would begin evaluating changes in the environmental conditions associated with the disease. For example, they could ask if recent flooding increased the number of stagnant pools of water favored by mosquitoes for breeding. Perhaps a shift away from pesticide use or other mosquito-control techniques contributed to a marked increase in the malariacarrying mosquito population. Finally, a change in climate or an increase in temperature could have increased mosquito breeding and forced human populations outside, and thus into contact with the disease carriers more frequently. Each of these lines of inquiry stems from a common observation. Environmental conditions are closely related to the spread of malaria, and a small change in environmental conditions could precipitate a sudden uptick in the incidence of disease. Of course, because malaria has been dormant in the United States for more than 50 years, it is unlikely that these environmental conditions would be the only factor contributing to hasten the spread of the malaria parasite. Students of mosquito vectors have observed considerable genetic and behavioral variation in the different Anopheles species that carry the malaria parasite (CDC 2004). Some Anopheles mosquitoes are more effective at carrying and transmitting the malaria parasite than others. Although there are 430 species of Anopheles mosquitoes, only 30 to 40 are known carriers of the parasites (CDC 2004). Of these, many Anopheles mosquitoes are relatively poor hosts for the malaria parasite, whereas others are especially effective at transmitting the parasite. Even when the ability of the mosquitoes to carry the malaria parasite is equal, some vectors are more likely to transmit the disease than others while


13

feeding. Many Anopheles mosquitoes are zoophilic. Some species prey aggressively on humans, whereas others feed primarily on other animals. Finally, some Anopheles vectors have developed a resistance towards pesticide or repellent and thus are more able to overcome existing abatement programs to address the disease. In addition to measuring changes in the environment, public health officials dealing with an outbreak of malaria would begin documenting any changes in the vectors or carriers of disease. The introduction of a new species of mosquito could lead to an influx in mosquito-borne disease, whereas a change in the behavior of a native mosquito could also lead to a reemergence of malaria transmissions. In addition to documenting changes in the carriers of malaria, researchers would also be interested in measuring changes in the contagion of the agent, and begin researching potential changes in the malaria parasite itself. As with the vectors of malaria, there is considerable variation across different forms of the malaria parasite. The P. falciparum malaria parasite is associated with higher incidence of human mortality, while the P. vivax and P. ovale parasites remain dormant in humans for months and years before proliferating, making the best-laid malaria-control programs difficult (CDC 2004). A mutation in any malaria parasite could make transmission from vector to host more likely than before, especially if malaria parasites become immune to treatment methods. An introduction of a new form of the parasite could easily account for a sudden increase in the rate of infection in a region where malaria had previously been eradicated. Finally, a student of public health might become interested in the changes in the hosts of malaria, and begin to explore genetic and behavioral traits of people who became infected with the parasite. Many individuals of African ancestry have sickle cell trait and are resistant to common forms of the malaria parasite. A demographic shift increasing the population without this trait could lead to an upsurge in transmissions of the mosquito-borne disease. Community health officials might also explore the behavioral factors associated with exposure to the malaria parasite. A decline in preventative behaviors such as the use of bed nets or mosquito repellent could also shape the rate of transmission leading to an outbreak of malaria in the region. The example of malaria provides some context to the epidemiological approach to understanding the dynamics of disease. Rather than being limited to a single explanation or cause, the framework encourages researchers to generate theory about a number of factors leading to the outbreak of malaria, including how environmental conditions, the traits of the vector, the characteristics of the agent, and the behavior and characteristics of the host itself all contribute to the transmission of disease.


14

State Characteristics

Issue Salience National Mood

Innovation Characteristics

Interest Groups & Professional Organizations

figure 1.2. The epidemiologic framework of innovation diffusion.

diffusion. Researchers documenting the role of professional or political networks in the diffusion of innovation have explained the role of carriers in the innovation and diffusion process (Gray 1973; Mintrom 1997). Studies on the internal political, social and institutional characteristics of early and late adopting states have identified how the characteristics of hosts shape state receptivity to innovation (Walker 1969; Savage 1978; Canon and Baum 1981; Carter and LaPlant 1997). Finally, a select few researchers have focused on the virulence of the idea itself, and have offered theory for how specific characteristics of policy agents shape the rate and extent of diffusion (Savage 1985a; Mooney and Lee 1999). As illustrated in Figure 1.2, the movement of policy innovations across jurisdictions is shaped by the changes in the political environment, the carriers of innovation, the distinct traits of each American state or jurisdiction, and the unique characteristics of the innovation agent. This framework has significant advantages for capturing the dynamic processes of policy diffusion. The epidemiologic framework connects research from students who have focused on such varied factors leading to policy innovation, adoption, and diffusion as the strategic behavior of interest-group communities, the characteristics of states, changes in national mood, or the emergence of a new policy idea into a single conceptual framework. Moving towards an epidemiologic framework further encourages researchers to consider intriguing questions about the virulence of policy ideas, the different strategies and behaviors of policy entrepreneurs and interest groups, and the varying resistance and susceptibility of states to innovation. It is an explicitly comparative approach. It attempts to understand the dynamics of diffusion by differentiating


15

between the attributes of states, interest groups, and policy innovations that determine the distribution of innovation through the federation. Using the epidemiologic framework as a loose guide for organizing research in diffusion dynamics allows us to draw a number of compelling implications for the causes of policy outbreaks. Text Box 1.1 demonstrates how a political scientist might follow the epidemiologic framework to conceptualize key factors associated with the sudden emergence and spread of anti-crime policies in the United States. text box 1.2 Thinking about Diffusion: Tracing the Spread of Anti-Crime Policies To illustrate how the epidemiologic framework can be transferred to the study of policy diffusion, consider how a political scientist familiar with the basic framework of epidemiology would study the sudden spread of anti-crime policies like the Amber Alert or three-strikes sentencing guidelines in the 1990s. The researcher would recognize that the sudden diffusion of these anti-crime innovations could be examined from a number of different perspectives, ranging from a change in national or state political conditions more favorable to “tough on crime” legislation, an evolution in the type of anti-crime policy favored by state legislatures, the emergence of new interest groups advocating for “tough on crime” legislation, or a change in political, cultural, or institutional makeup of the states adopting new crime policies. In organizing their research around these questions, the political scientist would be following a framework familiar to epidemiologists. This research would explore how the variation in the agents, carriers, and hosts of anti-crime innovation shaped rates and patterns of public-policy diffusion. As a starting point, a crime policy analyst would begin by looking at how changes in the political environment shaped the sudden emergence of state anti-crime policies. For example, analysis of Gallup’s national “most-important problem” (MIP) data reveals that Americans were disproportionately concerned with the crime problem in the mid 1990s (www. policyagendas.org/datasets/index.html#mips).∗ While historically fewer than 5% of Americans have identified crime as one of the most important problems facing the nation, from 1994 through 2001 an average of 25% of respondents (continued) ∗

Trends in Gallup’s national “most-important problem” (MIP) data can be accessed through the Policy Agendas website, at www.policyagendas.org/datatools/toolbox/ analysis.asp. To view trends in national opinion on crime and crime control, navigate to this webpage and select the “Most Important Problem (1947–2007)” and “all subtopics in Law, Crime and Family issues” options. These MIP data are included as part of a larger collection of data made available through the Policy Agendas Project, directed by Bryan Jones, Frank Baumgartner, and John Wilkerson.

16


in the Gallup MIP poll viewed crime as a pressing problem facing the nation, suggesting political conditions were ripe for crime-policy innovation. Other political scientists might explore whether changes in crime rates or unemployment statistics were associated with periods of state anti-crime innovation. Each of these questions are related to the assumption that state policy-making activity is in some way responding and reacting to the surrounding policy environment. Yet it is unlikely that elevated attention to the problem of criminal justice policy would be the only factor driving the rapid spread of anti-crime policies like the Amber Alert or the three-strikes law. Students of the policy process have commented on the important role of policy activists and organized interests as the carriers of innovation in the United States. Political scientists have documented a good deal of variation in the size, strength, and strategies of interest groups operating in the United States. Interest groups specialize according to lobbying strategies, membership size, financial resources, and organization. Some nationally prominent interest groups benefit from a reserve of financial resources or possess the advantages of mass membership and grassroots participation. Other interest groups have limited membership and struggle to capture resources needed to generate political support for their policy goals. These distinctions relate to the ease with which different carriers of innovation could orchestrate a diffusion campaign across the 50 states. The involvement of well-organized and well-funded groups – such as the National Center for Missing and Exploited Children or Mothers Against Drunk Driving – could accelerate the diffusion of an anti-crime initiative by actively carrying it from state to state using existing networks and traditional pressure campaigns. Interest groups that suffer from a relative dearth of financial or human resources – such as the National Organization for the Reform of Marijuana Laws – would struggle to simultaneously pressure for policy change across states. Other political scientists might explore how the characteristics of the innovation agent themselves shape diffusion, exploring how the cost, complexity, or rhetorical framing of an innovation shaped patterns of policy adoption. Child protection policies like the Amber Alert or Megan’s Law directed public attention towards the visceral need to protect young children from violence. The three-strikes law cast mandatory sentencing laws in populist and easily accessible language that appealed broadly to American voters. Although these policies varied according to the ultimate cost, none involved terribly complex or sophisticated expertise to understand how the policy was intended to diminish crime. A similar dimension can be used to explain the slow spread of criminal justice reform policies in recent history. Efforts to address social or public health problems associated with drug use face the challenge of convincing voters that drug use is a problem of public health rather than criminality. In each of these cases, researchers can assess how


17

changes in the scope, cost, or presentation of individual policies shaped their rates of diffusion. A final set of questions revolves around how variations in the political, institutional, and cultural traits of states could lead to their resistance or susceptibility to anti-crime policies. These questions implicitly ask how the unique makeup of each of the 50 state hosts of policy innovation shape patterns of policy adoption. For example, 24 states in the union allow citizens to pass either constitutional amendments or legislation through direct democracy (www. iandrinstitute.org). These states thus have an additional institutional avenue for innovation adoption that is especially responsive to interest-group influence and majoritarian politics. Conversely, states without direct democracy might be generally resistant to populist anti-crime movements. Crime policy researchers have operationalized a number of ways for how variations across each of the fifty states (racial demographics, political culture, wealth, professional legislatures, etc.) shape anti-crime policy innovation and adoption. All of these studies explore common questions for how variations in state innovation hosts shape the probability of public-policy adoption. This discussion provides some context for how an epidemiologic framework can be applied to the study of public-policy diffusion. Instead of being limited to a single explanation for the diffusion of innovation, the framework encourages public-policy researchers to think broadly about how changes in the political environment, interest-group carriers, the unique attributes of the innovation, and the political, cultural, and institutional characteristics of states all contribute to public-policy diffusion.

Innovation Characteristics and the Diffusion of Innovations A first major set of implications drawn from the epidemiologic framework of policy diffusion is that the characteristics of innovations or policy agents themselves contribute to diffusion dynamics.17 As with strains of malaria or the flu, not all innovation agents interact the same way with states in the federation. Instead, policy ideas possess attributes that make them more or less likely to produce policy outbreaks. For example, the Amber Alert possessed special qualities – targeting the protection of children through the use of existing emergency broadcast technologies – that made it especially appealing to voters across states and countries. 17

The question of how variation across different innovations shapes rates and patterns of diffusion has been largely overlooked in policy diffusion research. For a summary of research and future directions in this area, see Andrew Karch’s (2007b) “Emerging Issues and Future Directions in State Policy Diffusion Research.” State Politics and Policy Quarterly 7(1): 54–80.

18


Chapter 3 identifies a number of key innovation attributes that contribute to diffusion dynamics. First and foremost, policy innovations represent issues with different natural levels of issue salience. Because diffusion dynamics are in part a function of attention allocation in political decision making, high-salience issues should be especially prone to positive feedback cycles across states, whereas low-salience issues should encourage incremental decision making, as they fail to focus mass political attention. Many policies – like those related to education, defense, or government fiscal policy – have naturally elevated issue attention, whereas others are traditionally lower priorities for governments. However, issue salience is also shaped by the changing ways governments describe a policy problem, and shifting rhetoric around the way people understand a policy problem can shape mass political attention and elevate issue salience. Beyond salience, a number of other characteristics shape the probability that a given innovation will interact with states to produce a policy outbreak. Innovations vary according to issue complexity and program cost. Innovations that require professional technocratic program analysis or the allocation of significant government resources will tend to diffuse more slowly than cheap innovations that require little in the way of policy expertise. These attributes are not unique to individual innovations and can be assessed to classify policies that are more or less likely to produce policy outbreaks in the federation. Researchers who have studied common attributes of distinct classes or types of public policies have been interested in how “policy determines politics” (Lowi 1972, 300). Recent research connecting policy types to the diffusion of innovation suggests that public policies with common attributes follow similar patterns of policy diffusion (Mooney and Lee 1999). For example, morality policy is characterized by elevated issue attention, low complexity, and high emotional appeal. If elevated issue salience and diminished issue complexity are connected to rates of diffusion, morality policies should be especially prone to policy outbreaks. On the other hand, state regulatory policy – a policy form typified by high technical complexity and low salience – should conform closely to incremental patterns of policy diffusion, as this class of policies rarely engages mass political attention. Finally, governance policy – a policy type where citizens regulate the behavior of elected government officials and political institutions – should be especially prone to nonincremental patterns of diffusion. Governance policy is not only marked by high salience and low complexity, but is also strongly associated with the political institution of the direct statutory citizen initiative.


19

Governance policy should therefore be entirely unrelated to incremental patterns of policy diffusion, and especially prone to policy outbreaks, although the outbreak may be limited to states with citizen initiatives. Chapter 3 evaluates how the characteristics of the policy agent encourage diffusion dynamics. Building upon the statistical method advanced in Chapter 2, Chapter 3 demonstrates that different policy types produce dramatically different diffusion dynamics. Some classes of innovation produce patterns of diffusion that closely match incremental decisionmaking processes, while others are much more likely to produce policy outbreaks. State Characteristics and Diffusion Dynamics A second major implication of the epidemiologic framework is that the characteristics of the states should, insofar as they shape susceptibility to policy innovations, lead to considerable variation in patterns of diffusion across states. The diffusion of innovations is not simply determined by whether or not a given state came into contact with innovation at a certain moment. Instead, states are themselves more or less receptive to distinct forms of innovation. The diversity of state political, institutional, and socioeconomic characteristics causes considerable variation in state receptivity to innovation. Chapter 4 assesses state receptivity to classes of morality, regulatory, or governance policy.18 Because these policy forms encourage different processes of political decision making by state governments, state receptivity to morality, governance, and regulatory policy innovation will vary based on each state’s capacity to engage with decision making associated with specific types of innovation. A quick profile of state political and institutional characteristics provides good reason to expect that state receptivity to morality, governance, and regulatory policy will differ depending on differences across key state political and institutional attributes. States with the direct statutory initiative process are most receptive to governance policy reforms, as 18

Chapter 3 elaborates on the distinction between regulatory, morality, and governance policies. State regulatory policy encompasses economic, environmental, and professional regulatory regimes where government uses its coercive authority to shape the behavior of private businesses, organizations, and individuals. Morality policy refers to social regulatory policies where government uses legal authority to dictate social, community, and moral values and practices (Tatalovitch and Daynes 1998). Governance policy addresses those policies which delineates the terms and limits of governmental authority itself.

20


the initiative process provides an additional venue for direct citizen regulation of government institutions and elected officials. States without the initiative process should be resistant to governance policy, as citizen activists must appeal directly to government officials to regulate their own behavior. State receptivity to regulatory policy (addressing economic, environmental, or professional regulation) is shaped by an entirely different set of characteristics. State receptivity to regulatory policy is directly related to a state’s capacity to engage in complex policy analysis and experiment with technically complicated regulatory design and implementation. Thus, states with more professional legislatures are more receptive to regulatory policy innovation, as legislators are provided with the resources and the staff support to engage in technical program analysis. Likewise, those states with citizen legislatures are resistant to regulatory policy innovation, as they have neither the resources nor the capacity to engage in technocratic regulatory policy analysis. Instead, states with citizen legislatures lag in the adoption of regulatory policy innovation, as they must rely on the analysis and experiences of other states to suggest appropriate state regulatory regimes. Finally state responsiveness to morality policy is shaped by citizen ideology and state levels of electoral competition. Because morality is high salience and encourages mass participation, state receptivity to morality policy is shaped by state legislative responsiveness to citizen demands. Chapter 4 evaluates how variation in the internal political, institutional, ideological, and sociodemographic characteristics of states shapes receptivity to classes of innovation. Chapter 4 builds upon the decisionmaking model to evaluate how state decision-making capacities lead to differences in receptivity to morality, regulatory, and governance policies. This chapter first ranks states according to the speed with which they adopt emerging regulatory, morality, and governance policies. It then generates a statistical model to evaluate those attributes that make states receptive or resistant to different classes of public policy. Policy Vectors: Interest-Group Networks and Diffusion Dynamics A final implication of the policy contagion model is that the carriers of policy innovation produce distinct patterns of diffusion dynamics. Although state governments hold a central place in studies of state policy diffusion, elected officials are not necessarily the primary vectors of innovation in the United States. Instead, policies are communicated across states by interest-group activists, who carry innovation from one state to another


21

through organized interest-group networks. Variation in the behavior and organization of policy vectors has important implications for patterns of policy diffusion. Some groups possess the resources and membership to aggressively pursue interstate diffusion campaigns across all states in the federation, whereas other interest groups are limited in their ability to advocate for the diffusion of innovations beyond a small subset of receptive states. Chapter 5 explores how variation in the organization and behavior of interest-group networks and professional organizations shapes diffusion dynamics. This chapter draws upon a series of historical case studies to illustrate how variation in interest-group organization and behavior precipitates both incremental and nonincremental patterns of policy diffusion. This section augments empirical findings in the previous sections of the book by documenting how activists and interest-group organizations are integral to interstate policy diffusion. These case studies consider how political activists or “policy entrepreneurs” strategically exploit the multiple venues of policy making in America, revealing how strategic issue framing and reframing, venue shopping, and the organizational structure of interest-group networks interact to shape diffusion dynamics. Implications This book contributes to the understanding of policy making in federations by exploring how innovations emerge and spread from one jurisdiction to the next. The project advances a model for understanding the considerable variation in the patterns of policy diffusion across states. The book first demonstrates that policy diffusion exhibits punctuated dynamics that are indicative of a mixed model of policy change. This approach confirms prior findings of incremental learning across state legislatures, but then further demonstrates how attention-driven policy prioritization causes positive feedback cycles leading to policy outbreaks. The book then introduces an epidemiologic model of public-policy diffusion in order to provide theoretical and empirical leverage for understanding how variations across states, innovations, and interest-group networks produce positive feedback cycles leading to policy outbreaks across states. This approach moves the focus away from a narrow study of state legislative decision making, and instead reveals the importance of the interactions between interest-group activists, the nature of the policy innovations, and the receptivity of states to the communication of innovations in federal policy making.

2 Incrementalism and Policy Outbreaks in the American States

This chapter builds upon recent advances in the study of policy dynamics to examine the decision-making processes associated with the diffusion of policy innovations. It demonstrates that patterns of policy diffusion display punctuated dynamics inconsistent with a single process of incremental learning, but instead indicate multiple underlying decision-making processes. This is perhaps shown most forcefully by the abrupt and rapid spread of policy innovations across states in American history, a diffusion that cannot be explained through the process of incremental learning but must rather reflect decision making under extreme pressures for change placed on state governments. A more nuanced theoretical understanding of the process of publicpolicy diffusion can be gained by integrating research on innovation diffusion with studies of agenda setting in political decision making. The agenda-setting perspective demonstrates that government attention is unequally allocated in political decision making. State decision makers prioritize and respond differently to competing streams of information based on perceived issue importance, salience, and urgency. The diffusion of public policies often conforms neatly to the process of policy identification, evaluation, and emulation central to most modern research on innovation diffusion. However, a significant subset of policy innovations attracts immediate and widespread attention. These policies encourage immediate political responses, leading to the outbreak of nearly identical policy innovations across states. If diffusion dynamics are truly driven by two or more paths of organizational information processing, then the existing incremental decision-making model is insufficient to explain the processes of diffusion dynamics. 22

Incrementalism and Policy Outbreaks in the American States

23

To evaluate how agenda-setting dynamics shape innovation diffusion, this chapter explores distinct patterns of policy diffusion, drawing upon a data set of state adoption times for 133 different policy innovations. The chapter introduces a method – familiar to students of agenda setting – to compare the distributions of innovation-adoption times in state policy making. It compares historical patterns of policy diffusion in the American states to simulated patterns that replicate diffusion processes if diffusion were purely driven by incremental learning. This comparison is especially appropriate for modeling the diffusion of innovations because observational data of both incremental decision making and the diffusion of innovations are expected to produce an S-shaped normal curve of policy adoptions over time (Gray 1974; Rogers 2003). A simple comparison of real and simulated diffusion curves indicates how well diffusion processes fit incremental learning patterns (Padgett 1980; Rogers 1983; Baumgartner and Jones 1993). This chapter proposes that the degree to which a given empirical pattern deviates from the simulated S-shaped pattern indicates the degree of incrementalism versus positive feedback cycles in policy diffusion.1 This chapter then extends this analysis by exploring diffusion dynamics in three historical eras spanning the twentieth century. A number of researchers have observed that advances in communications technology have shaped the pace of public-policy diffusion in the modern era (Savage 1985a; Mossberger 2000). The chapter extends the analysis by comparing the underlying patterns of adoptions for policies diffusing in the early, middle, and late twentieth century. Disaggregating policies by historical era allows for a simple test of whether nonincrementalism in public-policy diffusion is shaped by changes in communications technology across historical eras. The findings presented in this chapter complement recent research in political decision making (B. Jones and Baumgartner 2005). Diffusion dynamics result from two distinct processes of state institutional decision making – one process representing incremental policy adjustments and another, representing sudden moments of policy imitation. These two 1

This chapter employs distributional analysis to compare the distributions of simulated and real diffusion data. This approach is facilitated because of the properties associated with data collected from incremental diffusion processes. When plotted, the pattern of policy adoptions over time is expected to be not only S-shaped, but also normally distributed. Because policy diffusion is traditionally believed to be normally distributed, a series of statistical tests determining how well historical diffusion data matches a normal distribution allows speculation about models of policy diffusion.

24


processes result in remarkably different patterns of diffusion over time. Many policies (like state lotteries or the diffusion of living-will legislation) produce S-shaped patterns of adoption consistent with boundedly rational information processing and policy emulation by state legislatures (Rogers 1983; Mooney and Lee 1999). However, as outlined in the previous chapter, many of the most sudden and interesting cases of diffusion – movements such as term limits, “three strikes” laws, and child protection laws such as the Amber Alert – produce punctuated diffusion patterns that indicate decision making driven by elevated attention, emotional reasoning, and policy imitation. These policies depart from the traditional S-shaped curve in a number of interesting ways. Some, like the Amber Alert, produce steep S-shaped curves as policy adoption spreads extremely rapidly across states. Others, like the diffusion of the death penalty, produce R-shaped patterns of adoption, indicating policy diffusion driven by sudden agenda-setting pressures. Still other policies produce step patterns as policy adoptions occur through short bursts of state attention rather than gradual policy adjustment. These findings qualify conventional wisdom for understanding the diffusion of innovations by demonstrating that policy diffusion frequently departs from the process of incremental learning. States may well emulate successful policies, but the careful evaluation of costs, benefits, and outcomes is less common than theory suggests. Incrementalism and Diffusion Models of Policy Adoption It is one of the happy incidents of the federal system that a single courageous state may, if its citizens choose, serve as laboratory; and try a novel social or economic experiment without risk to the rest of the country. (Justice Louis D. Brandeis, 1932).

Central to nearly all studies of policy diffusion2 is an effort to understand the processes of policy evaluation and adoption implied in the “laboratories of democracy” metaphor. Here, an organizing assumption of the study of policy diffusion is that patterns of adoptions are not independent events but rather the product of interstate influence in a federal 2

By convention in the diffusion literature, a policy innovation is defined as any public policy that is “new to the state adopting it” (Walker 1969; Gray 1973, 1174) no matter how much time has passed between the original invention and subsequent adoption. Policy diffusion can be defined as “any pattern of successive adoptions of a policy innovation” (Eyestone 1977, 441).


25

system. The primary research questions underpinning most studies of policy diffusion revolve around why and when states adopt a given policy innovation (F. Berry and Berry 1999). Answers to these questions have been located in two general approaches. Following Walker’s pioneering work (1969), one school compares characteristics of early and late adopters to evaluate the internal dynamics leading states to innovate (Walker 1969; Savage 1978; Canon and Baum 1981; Nice 1994; Carter and LaPlant 1997). Other researchers are interested in modeling the effects of social learning through interstate networks on patterns of adoption (Gray 1973; F. Berry and Berry 1990, 1992; Mintrom and Vergari 1998). The event history framework introduced by F. Berry and Berry (1990) permits researchers to simultaneously model both internal dynamics and interactive processes. This approach has produced stimulating studies of how interstate economic competition (F. Berry and Berry 1990, 1992; Boehmke and Witmer, 2004), social policy learning (Mintrom and Vergari 1998; Boehmke and Witmer, 2004; Rincke 2004), and ideological similarities (Grossback, Nicholson-Crotty, and Peterson, 2004; Volden 2006) encourage the diffusion of innovations across states. An interesting recent addition to the study of diffusion addresses the phenomenon of policy reinvention, exploring how the scope and language of legislation evolves over a diffusion cycle (Glick and Hays 1991; Hays 1996). Although these studies identify different factors for explaining patterns of policy diffusion, they build upon a common behavioral model of political decision making. Following the study of organizational decision making advanced by Lindblom (1959), researchers have accepted that policy diffusion results from a process of incremental learning (Walker 1969; Gray 1973; F. Berry and Berry 1999; Mooney and Lee 1999) and have argued that the diffusion of innovations stems from the satisficing behavior of state decision makers operating under time constraints and considerable uncertainty (Walker 1969; Gray 1973; F. Berry and Berry 1999; Mooney and Lee 1999). Rather than taking a comprehensive solution search for each policy problem, governments borrow heavily from their neighbors or ideological peers (F. Berry and Berry, 1990; HaiderMarkel 2001; Grossback, Nicholson-Crotty, and Peterson, 2004; Rincke 2004; Weyland 2005). For example, a state government faced with problems of increasing identity theft may look to other states for model laws addressing privacy and consumer protection. This is, however, only a refinement of the incremental decision-making model of policy diffusion; it still has state legislatures looking to the experience of other states as

26


a way of reducing both the information costs and uncertainty regarding the outcomes of a new policy (Lindblom 1959; Glick and Hays 1991; Weyland 2005; Volden 2006). The incremental decision-making model holds for innovation diffusion driven by economic competition or social-policy learning. A state government losing revenue as a result of a neighbor’s lottery program will look to regional programs as a starting point for crafting new legislation. A politician pressured to introduce living-will legislation often will turn to neighboring states with similar demographics for guidance on drafting legislation. Although diffusion scholars have documented the process of policy reinvention, in which policies are modified over time as they are adopted by states, research in innovation diffusion nonetheless suggests that states begin by borrowing from the legislation of their peers. Walker (1969, 881) describes an instance where a policy was copied with so few amendments and revisions by other states that many accidentally reproduced a typo in the original legislation. This incremental learning model confirms the benefits of decentralized policy making in federations, as states are more likely to emulate policy successes than failures (Volden 2006). Diffusion occurs as state decision makers identify, evaluate, and adopt the successful policy experiments of their neighbors. In this formulation, even the adoption of a dramatic new policy innovation that is a stark nonincremental departure from the status quo is viewed as the result of an incremental decisionmaking process. As F. Berry and Berry (1999) note, “By showing how emulation of other states’ innovations can aid in simplifying complex decisions, policy diffusion theorists have demonstrated how the adoption of non-incremental policies can be consistent with the logic underlying incrementalism” (171). Challenges to Incrementalism in Policy Diffusion The movement of policy innovations across states clearly does not always follow the clean trajectory hypothesized by students of diffusion. Often a series of adoptions occur within such a short time frame that the type of policy evaluation and lesson-drawing implied in incremental learning models becomes nearly impossible. As with interest-group-sponsored initiative campaigns, some innovations are enacted with limited involvement of elected governments. Although diffusion studies have done well to fit incrementalism to case studies of diffusion driven by economic competition and social-policy learning, they offer an incomplete understanding of how policy ideas move across states. This dynamic is illustrated in

27

50 40 30

Death Penalty ---

20

--- State Lotteries

10 2005

2000

1995

1990

1985

1980

1975

1970

1965

0 1960

Number of Adopters


Year

figure 2.1. Distinct patterns of policy diffusion: The death penalty and state lotteries.

Figures 2.1 and 2.2, which illustrate the diffusion of death penalty reenactments, state lotteries, charter school legislation, and the Amber Alert. Each case demonstrates not only a different pattern of adoption, but also shows tremendous variation in the rates of innovation diffusion. As Figure 2.1 illustrates, there was a pronounced difference in the rate of adoption of the death penalty and state lotteries in the second half of the twentieth century. A similar dynamic emerges in Figure 2.2. Charter school legislation was adopted by 40 states over the 1990s. The Amber Alert was adopted in all 50 states within six years. Factors beyond rates of speed of adoption challenge the incremental learning assumption underlying diffusion research. Consider the episodic patterns of policy diffusion displayed in the adoption of state “English

50 Number of Adopters

45 40 35 30

Charter Schools ---

25 20

--- Amber Alert

15 10 5 0 1990

1995

2000

2005

Year

figure 2.2. Distinct patterns of policy diffusion: Charter schools and the Amber Alert.


28 40 Number of Adopters

35 Gubernatorial Term Limits ---

30 25 20 15 10

English Only Laws ---

5 0 1780

1805

1830

1855

1880 1905 Year

1930

1955

1980

2005

figure 2.3. Episodic patterns of policy diffusion: Gubernatorial term limits and English Only legislation.

Only” legislation or gubernatorial term limits3 (Figure 2.3). Rather than resembling the S-shaped distribution expected of a process of gradual but consistent issue uptake and policy adoption, the diffusion of these two public policies has been marked by temporal bursts of adoption activity from subsets of states interspersed with long periods of state policy inactivity. Louisiana passed an English Only law in 1811, whereas the next state to follow suit was Nebraska in 1920. The remaining 28 states to adopt English Only legislation did so between 1980 and 2000. New Hampshire passed gubernatorial term limits in 1787, followed by a select few states in the mid-nineteenth century, and a large group of states in the early 1990s. These innovations indicate remarkable nonincrementalism in policy diffusion, as these policies are characterized by extraordinary periods of inactivity interrupted by sudden periods of policy change.4 Agenda Setting and Diffusion Dynamics Research in agenda setting suggests that studies of policy diffusion have overemphasized the role of incremental decision making in state 3

4

Both state English Only laws and gubernatorial term limits were excluded from the statistical analysis in this research because of the abnormal length of time that elapsed between the first and second adoptions of these innovations. Interestingly, a similar dynamic emerges in the historical diffusion of state lottery programs and the death penalty. Although researchers generally treat these policies as new innovations, they have in fact been the object of significant policy-making activity throughout U.S. history. In various forms, state lotteries and the death penalty have been banished and reinstituted by states throughout U.S. history, suggesting that even the diffusion of these programs are driven by shifts in governmental attention.


29

legislatures while excluding other important factors shaping the process of policy diffusion. First, studies across agenda-setting perspectives demonstrate that changes in the policy environment can shape patterns of policy making. Elevated issue attention caused by a focusing or a mobilizing event that galvanizes mass public and political attention can open a “window of opportunity,” creating conditions for rapid and nonincremental policy change (Kingdon 1984; Baumgartner and Jones 1993; Glick and Hays 1997). Even absent a focusing event, research on framing effects (Zaller 1992; Kahneman and Tversky 2000), issue definition and redefinition (Baumgartner and Jones 1993; B. Jones and Baumgartner 2005), and policy targeting (A. Schneider and Ingram 1993) demonstrates that drawing public attention to certain dimensions of a policy innovation can lead to swings in public opinion and support for new policy ideas. Like other policy-making processes, patterns of interstate policy diffusion are sensitive to shifts in mass political attention, especially when new political attention short-circuits an incremental learning process leading to sudden public-policy diffusion. Perhaps more importantly, research in policy-process theory suggests that the idea that states are laboratories of democracy is a useful but limited framework for understanding the processes of innovation and diffusion in America. These “laboratories of democracy” are linked venues and are poorly insulated from those interest groups that sponsor innovations and pressure policy adoption. Although elected officials clearly play a central role in evaluating and implementing public policies, they are not the only sources of change in the political system. Federalism encourages venue shopping, a process in which activists and interest groups strategically exploit the multiple venues of government to secure support for their legislative programs (Baumgartner and Jones 1993; Holyoke 2003; Pralle 2003). This process increases the number of sites where new ideas can enter the political systems and can create conditions in which “new ideas or policy images may spread rapidly across linked venues, thus setting in motion a positive feedback process” (Baumgartner and Jones 1993, 240). The notion that states act as independent policy laboratories therefore presents a somewhat unrealistic understanding of policy learning and diffusion, as the pressures generated by activists and interest groups can lead to innovations being adopted across states before decision makers have had the opportunity to evaluate the costs, benefits, and implications of a policy experiment. Several recent and controversial innovations seem to stand as prominent examples of this phenomenon. In the 1990s, “three strikes” sentencing guidelines were adopted by many states in response

30


to heightened public concern over chronic violent felons. Three-strikes policies – which require repeat violent offenders to serve mandatory life sentences after a third offense – were enacted well before states could assess the impact such policies would have on prison crowding, crime rates, or the legal system. Likewise, term limits were adopted across 20 states before the benefits and costs of term limitations could be evaluated. Only recently have the first state legislatures felt the impact of the retirement of a generation of politicians, and only recently have states been positioned to evaluate the political consequences of high turnover and short tenures in government. This is not to say that states necessarily erred when passing these laws – only that they passed each piece of legislation with an incomplete understanding of efficacy or costs. Such patterns emerge partly because of the influence of interest groups in the political process. The legislative objectives of interest-group activists and policy entrepreneurs who invent and advocate for the passage of new innovations are often distinct from the goals of elected state representatives. Whereas policy makers may have a sincere interest in evaluating the costs and benefits of a particular policy innovation (e.g., evaluating the effectiveness and costs of three-strikes sentencing laws or statewide smoking bans), interest-group activists have issue-specific agendas and will look to capitalize on a window of opportunity to galvanize public support for policy change. These nongovernmental actors are not constrained by the same considerations of policy feasibility, complexity, and program cost as are elected officials, nor are they bound by the pressures of reelection. Instead, interest groups are driven to pass legislation that best meets the interests and preferences of their members and are sometimes capable of mounting sophisticated campaigns to win legislative support for new innovations (Balla 2001). When activists encounter political opposition to innovation, they adopt a number of strategies to encourage policy adoption. Wealthy and well-organized trade, professional, or peak associations can apply insider pressure across state legislatures, providing expert testimony, sample legislation, and campaign contributions for politicians. Mass-member citizen advocacy organizations, such as Mothers Against Drunk Driving, can organize pressure campaigns, mobilizing voters to demand policy adoption. When working with the state legislature fails, interest groups may seek policy changes at other venues of government, pursuing reforms in municipal governments or through the courts. In nearly half of the states, policy entrepreneurs can bypass statehouse governments entirely, using direct citizen initiatives to change public policy.


31

Research that folds agenda-setting variables into studies of policy diffusion have provided some promising support for understanding how political attention and issue characteristics shape the rates of diffusion. In a study linking political salience and mass public opinion to policy diffusion, Scott Hays (1997) considers the possibility that innovation under heightened salience will result in a positive feedback cycle, arguing that “if political conditions within a given state political system favor adoption at the same time that an issue gains attention and is placed on the agenda, rapid – if not immediate – policy adoption is imminent” (497). His research on the temporal diffusion of living-will legislation revealed a moderate interaction between elevated national salience and policy diffusion. Studies of the electoral connection in the diffusion of innovations have expanded on the link between mass preferences, electoral agenda setting, and policy diffusion. Andrew Karch (2007a) looks beyond the traditional focus of state legislative decision making to understand how broader stages of problem identification, policy enactment, agenda setting, information generation, and customization shape the process of policy diffusion. Karch hypothesizes that time-constrained officials will be drawn to innovations that are visible and politically salient (Karch 2007a). Politicians may advocate innovation adoption not simply as a result of problem identification and program emulation, but also because sponsoring nationally popular initiatives wins electoral support from voters. As public support for innovation increases, so too does the pressure and expected benefit for innovation adoption. Research on the diffusion of morality policy supports the notion that the broad emotional appeal of a policy innovation can accelerate patterns of diffusion. Christopher Mooney and Mei-Hsien Lee (1999) discover that the reenactment of the death penalty followed a much more rapid pattern of adoption than commonly expected in diffusion research. Although standard diffusion patterns are expected to follow an S-shaped cumulative adoption curve, the diffusion of the death penalty exhibited a dramatic, sudden issue uptake across a large subset of states (see Figure 2.1). Mooney and Lee suggest that the high salience and technically unsophisticated characteristics of the policy result in faster adoptions across receptive states. Theories of policy learning in public administration and public policy likewise propose that the processes leading to the diffusion of innovation expand beyond a process of incremental problem identification and policy emulation. Peter May (1992) argues that the conceptualization of

32


goal-oriented instrumental policy learning central to the most stylized studies of policy diffusion represents only one narrow form of analysis and decision making in the policy process. Instead, May (1992) identifies three distinct forms of learning in the policy process: instrumental policy learning, social policy learning, and political learning. He explains: Instrumental policy learning entails lessons about the viability of policy instruments or implementation designs. Social policy learning entails lessons about the social construction of the policy problem, the scope of the policy, or policy goals. Political learning entails lessons about policy processes and prospects (332).

These three forms of policy learning are not mutually exclusive; instead they indicate that what diffusion researchers often conceive of as learning – the evaluation of program success or the expected political gains of program implementation – involves different dimensions of lessondrawing. The considerations leading to a political choice are not necessarily linked to questions of policy design, program complexity, or cost. Instead, political choices are informed by considerations as distinct as a desire to achieve previously identified policy goals, a strategic choice to ensure electoral victory, or a new way of framing a policy problem to advance a political agenda. Interestingly, there is a degree of inconsistency in political learning even in the domain of instrumental policy learning, where the goal-oriented, rational, and analytic process of program evaluation and imitation is believed to be central to the selection and implementation of policy instruments. May (1992) differentiates between true instrumental learning indicated by “policy elites’ increased understanding of policy instruments and designs,” (337) and superstitious instrumental policy learning, in which “beliefs about effectiveness of particular actions or individuals dominate any understanding of evaluation of performance” (336–337). This distinction has important implications for understanding the process of learning underlying the diffusion of public-policy innovations. As May (1992) points out: As with trial and error learning, learning can simply entail judgments about whether a given course of action or a given policy tool is still preferred to the alternatives currently being promoted. Copying or mimicking entails adoption of policy ideas without such understanding. For example, competition among states for certain kinds of industry may spur economic development fads (e.g., tax credits, enterprise zones) that are inappropriate for the situation some states face (333).


33

May’s research on learning challenges the incremental diffusion model’s simplification of policy learning. The factors compelling states to adopt innovation frequently do not match the process of gradual incrementalism identified in recent studies of diffusion research. Instead, state decision makers superstitiously imitate new policies either because they are popular, or because they are persuaded by expected future benefits. Studies of public-policy diffusion have identified cases where policy learning approximated both superstitious and true instrumental policy learning. In his study of the diffusion of the Children’s Health Insurance Program (CHIP), Volden (2006) discovered that state governments emulated those programs that proved most successful in raising the proportion of poor children covered. Interestingly, the decision-making process leading to program adoption was a mixture of instrumental and political policy learning, as the incentive for state decision makers to identify the most successful program was driven by the election-seeking behavior of politicians. Mossberger’s (2000) excellent evaluation of political decision making in the diffusion of free enterprise zones finds less compelling evidence for true instrumental policy learning. In interviews with policy experts involved in the design of state free-enterprise zones, she discovered that “although learning occurred, it can be characterized as limited. The knowledge held by most participants involved only one or two generalizations about state zones, and the participants conducted no active search for information” (Mossberger 1999, 49). Mossberger instead speculates that the technical information diffusing was limited and that it was insufficient for true rational policy learning to have occurred. Instead, the information diffusing resembled first-order “policy labels” rather than a broader set of detailed, program-specific policy information (2000). Thus decision makers gained knowledge of a program’s innovation and objectives, but obtained less policy-relevant information about policy instruments, program design, and actual performance. Taken together, research in agenda setting and policy learning offers an interesting possibility for modeling the diffusion of innovations. Rather than following a single, uniform learning model, policy diffusion is better described by several distinct, underlying decision-making processes. A large subset of policies moves across states through a process of incremental learning – in which state decision makers encounter and emulate a successful policy innovation (Rogers 1983; Glick and Hays 1991; Volden 2006). However, theory and research suggests a second set of policies moves not through any formal learning process per

34


se, but rather through elevated issue attention, emotional reasoning, and imitation. Decision-Making Models in Policy Diffusion To clarify how policy diffusion can be conceived of as resulting from distinct decision-making processes, it is useful to briefly review how researchers have explained the decision-making process leading to publicpolicy diffusion. The process of incremental learning leading to policy diffusion mirrors the process of trial-and-error learning in individuals (Rogers 1983; F. Berry and Berry 1999).5 Everett Rogers (2003) argues that the decision to adopt an innovation occurs through five stages of knowledge, persuasion, decision, implementation, and confirmation6 : 1) Knowledge occurs when an individual (or other decision-making unit) is exposed to an innovation’s existence and gains an understanding of how it functions. 2) Persuasion occurs when an individual (or other decision-making unit) forms a favorable or unfavorable attitude towards the innovation. 3) Decision takes place when an individual (or other decision-making unit) engages in activities that lead to a choice to accept or reject the innovation. 4) Implementation occurs when an individual (or other decisionmaking unit) puts a new idea into use. 5) Confirmation takes place when an individual seeks reinforcement of an innovation-decision already made, but he or she may reverse this previous decision if exposed to conflicting messages about the innovation. Rogers’s model of decision making in the diffusion of innovations is a more general form of the incremental learning model favored by students of public-policy diffusion. Policy diffusion occurs through a process of innovation and emulation as decision makers encounter a novel policy 5

6

This conceptualization is entirely consistent with the broader research tradition of bounded rationality in individual and organizational decision making. An explicit aim of bounded rationality is to connect the process of organizational decision making to the preferences and decisions of individuals who populate an institution or an organization (B. Jones, Boushey, and Workman 2006). Rogers’s conception of the innovation diffusion process is similar to the heuristic stages of the policy cycle in public-policy research, in which policy makers identify a policy problem, form competing solutions, arrive at a political decision, implement the public policy, and evaluate the efficacy of the policy solution.

Incrementalism and Policy Outbreaks in the American States Intendedly

35

Rational Band

Information Impression Emotion Attention Problem Representation Solution Search Choice

Cognitive Band Information Impression Impression Choice

figure 2.4. Newell’s bands of rationality. Source: Jones, Bryan D. 2001. Politics and the Architecture of Choice, Chicago: University of Chicago Press, p. 101.

solution, form preferences through evaluation, make a formal policy choice, implement that choice, and reevaluate the intended and unintended consequences of the innovation. It is a process identical to the process of problem identification and adoption taken by individuals in trial-and-error problem solving. Yet as elaborated in the previous section, this decision-making process represents only one path of preference formation leading to policy adoption and diffusion. Although studies of diffusion frequently identify cases that conform to the process of preference formation outlined in Rogers’s innovation and diffusion model, research in both individual and organizational choice in the policy process challenges this simplification of the decision-making process. Researchers across disciplines have noted how cognitive heuristics (Kahneman and Tversky 2000), emotional reasoning (B. Jones 1994), or the influence of prejudice and bias short-circuit the linear process of problem definition, evaluation, solution, and implementation in both individual and organizational decision making (B. Jones 1994; B. Jones, Boushey, and Workman 2006). Instead, there are multiple paths to preference formation and choice for both individuals and social systems. Such a mixed decision-making model of policy diffusion is consistent with models of information processing advanced by cognitive psychologists. Newell (1990) identifies two bands of human information processing: a cognitive band and an intendedly rational band. As Figure 2.4 demonstrates, the intendedly rational band closely resembles Rogers’s diffusion-of-innovations model, and represents a process of evaluation and learning as individuals generate alternatives and compare expected

36


outcomes (B. Jones 1994). However, the cognitive band involves relatively little formal reasoning, as it produces “a system that engages in a knowledge search but it does not engage in a problem search” (Newell 1990, 139). In the cognitive band, a decision is arrived at with little formal evaluation of expected costs or outcomes. To connect these two cognitive processes to systems-level decision making in policy diffusion, the intendedly rational band can be viewed as a process of incremental learning and emulation, whereas rapid, fad-driven policies are similar to the impressionistic, emotional reasoning typified by reasoning at the cognitive level. Such parallel dynamics of information processing in social systems are well documented in other diffusion study analogs. For example, epidemiologists have well-developed procedures for identifying and distinguishing between “point source” outbreaks, which occur when a subpopulation is simultaneously exposed to a common contaminant, and propagated person-to-person outbreaks, which appear when a contagious disease is communicated via person-to-person interactions (CDC 2009). In a point-source outbreak, a subgroup within a population shows symptoms of disease almost simultaneously as individuals respond to a common exposure. In propagated outbreaks, disease transmission occurs at varying rates, as the individuals first infected transmit the disease to others, who in turn transmit it more broadly within the population. Epidemiologists have developed a number of applications for modeling and interpreting the comparative speed of outbreaks from the first known observed case. In the social sciences, communications scholars and marketing analysts have also researched how mass-media and advertising campaigns shape the awareness of a media event or a new product across a population, and the extent to which consumer behavior is driven by person-to-person interactions (Bass 1969; Valente 1993). In diffusion driven by mass-media effects, awareness of innovation can occur nearly immediately across a population (Valente 1993). In diffusion driven by word of mouth, learning occurs more gradually (Valente 1993). In both the social sciences and epidemiology, researchers have developed distinct models of information processing to capture the dynamic process of diffusion. Punctuated Equilibrium Theory and Policy Diffusion The notion that the dynamics of individual and organizational decision making represent two distinct processes of preference formation and information processing is not foreign to students of the policy process. Students of punctuated equilibrium observe that policy making in


37

American politics is characterized by periods of policy stasis interrupted by pronounced policy punctuations (B. Jones, Sulkin, and Larsen 2003; Baumgartner and Jones 2005; Koski and Breunig 2006). These two distinct processes are explained by different forms of decision making. Policy stasis occurs through a process of incremental policy adjustment in a policy subsystem monopolized by a group of actors. Policy punctuations result when a focusing event elevates issue attention, changes the dominant attributes of a policy image, and focuses broader political attention on the politics of a particular subsystem. Elevated issue attention coupled with a new understanding of a policy problem leads to increased demands for policy change and results in a positive feedback cycle as a series of new policies are passed representing new preferences for a policy domain (True, Baumgartner, and Jones 1999, 102). Such dramatic patterns of policy change have been documented in studies of congressional attention (Baumgartner and Jones 1993; B. Jones and Baumgartner 2005), presidential decision making (Larsen 2006), federal budgeting (Jones, Baumgartner, and True 1998), and state budgeting (Koski and Breunig 2006). There is good reason to expect the same dynamic process to appear in the diffusion of policy innovations. In developing their theory of punctuated equilibrium, Baumgartner and Jones (1993) argue that policy diffusion is characterized by the same underlying dynamics of incremental adjustment and sudden positive feedback cycles captured in punctuated equilibrium theory. They write that “policy diffusion, with its S-Shaped curve, is remarkably like a punctuated equilibrium model in which the system shifts rapidly from one stable point to another” (Baumgartner and Jones 1993, 17). Empirical Models of Innovation Diffusion These theoretical models of information processing in social systems can be matched to distinct patterns of innovation adoptions. One nearly axiomatic finding across studies of diffusion-of-innovations theory is that patterns of adoptions follow a bell-shaped curve when plotted over time, and that the cumulative number of adopters in the life of innovation diffusion generally follows an S-shaped curve. In his classic Diffusion of Innovations, Rogers (2003) reviews over 5,000 research publications on diffusion and concludes that the S-shaped curve is so common that it is a general characteristic of innovation diffusion. The S-shaped curve of diffusion has been identified in a striking number of diffusion studies ranging from diffusion of farmers’ adoption of hybrid corn (Ryan and


38

Gross 1943), hate-crimes legislation (Grattet, Jenness, and Curry 1998; Rogers 2003), new-product adoptions (Rogers 1976), and across a range of education, welfare, and civil rights policies (Gray 1973). This generalization emerges from the processes that describe the cumulative adoptions over time as a product, behavior, or policy spreads across actors in a social system. The number of innovation adopters is initially limited, as only a few early pioneering individuals have identified and embraced an emerging innovation. As time progresses, the number of adopters accelerates as more and more individuals in a social system encounter and adopt a new innovation. This acceleration continues until a tipping point, when more than half of the actors in a social system have adopted an innovation. The adoption curve then continues to grow at a slower rate as fewer and fewer individuals in a system adopt a new behavior. Mathematically, this common diffusion process can be described by an internal influence diffusion model, where a new product or behavior is first introduced into a social system by a few pioneering individuals, and is then adopted throughout a population as more actors within a social system encounter and adopt that product or behavior. This can be described by the internal influence diffusion model, given by Mahajan and Peterson (1985) as dN(t) = bN(t)[N − N(t)] dt

(1)

where dN(t) is the rate of diffusion at time t, N(t) is the cumulative number dt of adopters at time t, N is the total number of potential adopters at time t, and b is the constant of imitation or internal influence. The cumulative adopters distribution function of the internal diffusion model produces an S-shaped curve, and can be represented by the following equation, given by Mahajan and Peterson (1985) as N(t) =

N (N − N0 ) 1+ exp[−bN(t − t0 )] N0

(2)

The internal diffusion model converges on an S-shaped curve because the model represents how innovation diffusion occurs through interpersonal communication. The initial period of adoption occurs gradually, as relatively few units in the social system can spread an innovation to others. As more actors in a social system encounter and adopt the innovation, the rate of diffusion increases rapidly, as there is a rapid growth in the


39

0.0

0.2

Percentage of Adopters 0.4 0.6 0.8

1.0

S-Shaped Curve of Diffusion

0

5

10

15 Time

20

25

30

figure 2.5. S-shaped adoption curve.

number of units that can spread the new behavior. Finally, as the number of prior adopters approaches the number of possible total adopters, the rate of diffusion invariably decreases, as fewer individuals remain that have failed to adopt the innovation (Mahajan and Peterson 1985). Figure 2.5 simulates the standard life cycle of innovation diffusion. The S-shaped curve is the cumulative frequency of innovation adoption over time. In Figure 2.5, innovation diffusion begins gradually, accelerates and decelerates proportionally as more actors in a social system adopt the innovation, and then tails off as fewer and fewer laggards adopt the new innovation. This S-shaped curve is called the diffusion curve because it provides a general pattern of innovation adoption over time. It is important to note that the rate of innovation adoption may vary according to the rate of internal influence, b, changing the slope and asymptote of individual adoption curves (Mahajan and Peterson 1985). However, regardless of the rate of internal influence, the diffusion-of-innovations curve will almost always follow the process of limited initial adoption, described by a gradual introduction, an increasingly rapid period of adoption, and a decreasing period of late adoption. Because innovation diffusion is routinely characterized by these Sshaped cumulative distribution curves, diffusion theorists have argued that innovation diffusion converges on a normal distribution (Rogers

40


2003). Rogers advanced a generalization that “adopter distributions follow a bell-shaped curve over time and approach normality” (2003, 275). According to Rogers, this generalization is supported by both theory and empirical data collected across an impressive range of innovation and diffusion research. Comparing the innovativeness of individuals and organizations to other human traits, such as height, intelligence, or the learning of new information, Rogers writes: We expect a normal adopter distribution for an innovation because of the cumulatively increasing influences upon an individual to adopt or reject an innovation, resulting from the activation of peer networks about the innovation in a system. This influence results from the increasing rate of knowledge and adoption (or rejection) of the innovation in the system. We know that the adoption of a new idea results from information exchange through interpersonal networks. If the first adopter of an innovation discusses it with two other members of the system, each of these two adopters passes the new idea along to two peers, and so forth, the resulting distribution follows a binomial expansion, a mathematical function that follows a normal shape when plotted over a series of successive generations” (Rogers 2003, 272).

The assumption that adopter distributions approach normality has been used to support a general theory of the characteristics of adopters in a social system. Rogers (2003) sorts innovation adopters into five distinct categories according to how rapidly they embrace a new innovation. Innovators represent the small subset of actors that first introduce a new innovation into a social system and are classified as the first 2.5% of adopters in the population. Early Adopters are opinion leaders or trendsetters that first imitate the policy after it is introduced and make up the next 13.5% to take up an innovation. The Early Majority are those that embrace a new innovation before the general population and represent 34% of all adopters. The Late Majority are skeptical adopters that resist a new behavior or innovation and make up the next 34% of the population, and Laggards are the remaining few traditionalists who are the last adopters of innovation and represent the final 16% of the population (282–285). Anomalies Models in the Diffusion of Innovations If the S-shaped curve of innovation adoption can be applied as a tool to explore the common traits of innovation diffusion, it is also useful as a starting point for locating and describing divergence from traditional patterns of policy diffusion. Although Rogers argues that the tendency


41

towards normality is a general characteristic of diffusion data, he also observes that innovation processes do occasionally depart from the standard S-shaped curve. Mathematical modelers have likewise employed a number of alternate distributions for fitting S-shaped diffusion curves. These diffusion theorists have used comparisons to the standard normal S-shaped curve to draw a number of inferences about the processes driving innovation diffusion. Researchers have examined the non-normality of innovation data to understand factors leading to unexpectedly rapid diffusion resulting from both external and internal pressures on social systems and to draw inferences about how crises, agenda-setting pressures, or contagion shapes innovation diffusion. Ryan and Gross (1943) compared the diffusion of hybrid corn to an expected normal distribution and discovered that the diffusion of corn seed occurred more rapidly than anticipated by a normal model. Valente (1993) demonstrated how the diffusion of three distinct innovations – hybrid corn, awareness of Eisenhower’s stroke, and physician prescriptions of new drugs – conformed to or deviated from an S-shaped distribution, demonstrating that diffusion of awareness from exposure to external events occurred more rapidly than diffusion through internal processes. Mooney and Lee (1999) demonstrate how the diffusion of the death penalty followed an R-shaped curve rather than the traditional S-shaped curve, speculating that the characteristics of morality policy led to more rapid innovation diffusion. More generally, Baumgartner and Jones (1993) speculate that innovation diffusion can spark positive feedback cycles in the United States, producing a more rapid and steeper-than-expected S-shaped logistic diffusion pattern. Findings from this body of research on diffusion “anomalies” have interesting implications for modeling public-policy diffusion. Many of these studies begin with an assumption that the standard model of innovation diffusion conforms closely to an S-shaped normal distribution. Researchers then compare empirical patterns of innovation in order to measure how strongly a given pattern matches an expected normal distribution of diffusion. These comparisons are used to draw inferences about innovation processes. Importantly, S-shaped distributions do not immediately indicate normal distributions. Virtually any unimodal distribution can produce an S-shaped curve, and researchers have employed logistic, log-normal, and Gompertz functions to model innovation diffusion (Mahajan and Peterson 1985). An extremely rapid pattern of innovation diffusion may

42


produce an S-shaped curve, but one which deviates strongly from a normal curve. One primary source of non-normality in adopter distributions emerges from the dynamics of internal influence in social systems, which can produce steep and non-normal S-shaped distributions. Ryan and Gross (1943) discovered that the diffusion of hybrid corn seed in Iowa deviated sharply from a normal distribution after an initial period of innovation. They speculate that normal frequency “does not appear to be a concept closely adapted to this condition where pressures, or reasons, for adoption become increasingly acute with passing time” (1943, 23). The diffusion of hybrid corn increased in speed as pressures to adopt the innovation increased dramatically on farmers over time. Baumgartner and Jones speculate that the rapid diffusion of policies across states would converge on a logistic distribution as demands for immediate policy adoption increase through a bandwagon effect or positive feedback cycle, as occurred with the term-limitation movement or the diffusion of the Amber Alert. The initial period of diffusion occurs slowly as new ideas are gradually evaluated and implemented; but then diffusion occurs extremely rapidly through a positive feedback cycle, as states rush to adopt emerging innovation. Such positive feedback cycles occur when innovation spreads rapidly through internal dynamics. Both Ryan and Gross and Baumgartner and Jones have implicitly connected variation in patterns of public-policy diffusion with processes familiar to epidemiologists and marketing researchers. Not all innovations produce like patterns in social systems. As the sheer contagion of an innovation increases, the likelihood also increases that it will lead to diffusion that is more rapid than traditionally expected. This dynamic can be captured in the internal diffusion model cited earlier. As the rate of innovation by imitation increases, policies will take off more rapidly than expected. These dynamics can also be modeled with the internal influence model. Figure 2.6 illustrates the S-shaped curve of diffusion of policies spreading at three different rates. As the contagion of internal influence increases, the speed of policy adoption and the slope of the S-shaped curve increase dramatically. While S-shaped distributions are by far the most common distribution of adoptions, diffusion researchers have observed alternate patterns of diffusion that are neither S-shaped nor normally distributed, which also merit discussion. As observed in Mooney and Lee’s (1999) research on the death penalty and Henrich’s (2001) exploration of the cultural

43

0.0

0.2


1.0


0

5

10

15 Time

20

25

30

figure 2.6. S-shaped adoption curves representing three different rates of innovation diffusion.

transmission of innovation diffusion, adoptions can be spurred not by gradual person-to-person interaction, but rather by exposure to an external common stimulus, such as a focusing event. Such phenomenon is known to researchers in epidemiology or communications who study how exposure to a common external parameter shapes innovation adoption. In these rare cases, adoption will occur more suddenly than expected, as units in a social system immediately encounter and respond to innovation. When plotted over time, the distribution of innovations in these instances will resemble an exponential curve, as innovation adoption across a large number of actors in the system is nearly immediate. Mathematical modelers of innovation diffusion have identified this process as an external influence model of innovation diffusion. Here, the innovation adoption is driven by immediate awareness of adoption, given by Mahajan and Peterson (1985) as dN(t) = a[N − N(t)] dt

(3)

where dN(t) describes the rate of innovation adoption, N(t) is the cumudt lative number of adopters at time t, and a is a coefficient or constant of external influence describing the influence of all factors other than interpersonal communication.


0.0

0.2


1.0

44

0

2

4 Time

6

8

figure 2.7. R-shaped exponential adoption curve.

As the constant a increases, the cumulative distribution function of the external influence model produces an R-shaped exponential distribution over time. This represents innovation adoption driven by responses to an external stimulus rather than internal dynamics. The cumulative distribution function is given by Mahajan and Peterson (1985) as N(t) = N[1 − exp(−at)]

(4)

When diffusion processes are shaped by external influences, such as an exogenous event or a common source influence on diffusion processes, innovation adoption can happen nearly immediately. When plotted over time, the external influence diffusion model produces an Rshaped curve, deviating strongly from an S-shaped innovation diffusion curve. Figure 2.7 illustrates the exponential adoption curve of diffusion driven by external influence. Of course, students of innovation diffusion have realized that both external and internal influences likely shape the rate of diffusion. Because of this fact, many researchers have preferred to represent the diffusion of innovations through a mixed-influence model developed by Bass (1969). The Bass diffusion model has been applied to estimate the rates of innovation diffusion driven by both external and internal influences (Bass 1969; Mahajan and Peterson 1985; Srinivasan and Mason 1986, Valente 1993). The Bass model returns estimates of two rate parameters of a diffusion


45

process: the constant a, representing the rate of innovation adoption due to innovation or external influence, and the constant b, measuring adoption driven by imitation. The Bass Model therefore provides a relatively simple representation of the processes driving the diffusion of innovations. The constant a describes the rate of initial innovation adoption, and the constant b represents the rate of new adopters of innovation at a given time as a fraction of those that have already adopted the policy. It is a preferred model of innovation diffusion because it provides a more realistic representation of the real-world processes of innovation diffusion than the internal or external influence model alone. The mixed-influence equation models diffusion as a response to both external and internal pressures for adoption. It is given here by Mahajan and Peterson (1985) as dN(t) = (a + bN(t))[N − N(t)] dt

(5)

with a cumulative distribution function of a(N − N0 ) exp[−(a + bN)(t − t0 )] (a + bN0 ) N(t) = b(N − N0 exp[−(a + bN)(t − t0 )] 1+ (a + bN0 ) N−

(6)

The mixed-influence model is useful for diffusion scholars for several reasons. Perhaps most importantly, it is a more theoretically satisfying model of innovation diffusion, as it incorporates how both external and internal pressures shape diffusion patterns. Diffusion can be an immediate response to an exogenous shock, followed by a period of internal learning, or diffusion can occur through internal dynamics at vastly different rates. The mixed-influence model provides a tool for distinguishing between external and internal dynamics driving innovation diffusion, and a method for capturing both dynamics simultaneously.7 This review of mathematical models provides two important takeaways for the subsequent analysis in the book. First, like the larger theory of incrementalism, the a priori expectation is that policy diffusion data will follow an S-shaped distribution when plotted over time. The distribution of policy adoptions will converge on a normal distribution. This 7

The Bass Model returns three constants, a, b, and M. An increasing constant, a, suggests more rapid diffusion from external influence. A larger constant, b, indicates more rapid innovation diffusion from internal influence. M is a measure of the expected (or total) number of adopters.

46


expectation is supported by both theory and evidence developed in the study of innovation diffusion writ large, and the study of public-policy innovation more specifically. Data will generally conform to an S-shaped normal distribution because most diffusion processes capture the process of trial-and-error learning in social systems. However, this discussion suggests that the diffusion of innovations can deviate from normality in two principal ways. First, exogenous agendasetting pressures can make innovation adoption more immediate, resulting in an R-shaped distribution resembling an exponential distribution. Second, the dynamics of internal innovation diffusion can shape the pattern of innovation diffusion. As the transmission of innovation increases through internal influence, the diffusion of innovation can deviate from a normal distribution. This occurs when there is a sudden acceleration in the number of adopters following an initially gradual introduction of innovation into the social system. Measuring Decision-Making Models in Policy Diffusion This review of theoretical and mathematical models of innovation diffusion presents an interesting question for research in policy diffusion. To what extent do patterns of diffusion match behavioral incrementalism? Are policy outbreaks uncommon aberrations, or does the rapid diffusion of innovations represent an important and distinct process of policy diffusion? Finally, is it possible to explain the rapid and sudden diffusion of innovations within the context of existing incremental diffusion theory? Prior challenges to incremental learning models outside of diffusion research have employed stochastic processes models to evaluate political learning and policy change (Padgett 1980; Baumgartner and Jones 1993; B. Jones, Sulkin, and Larsen 2003). This method is justified by an empirical nuance of incremental decision making. The observational data of an incremental learning process are normally distributed, providing researchers with a relatively simple method for verifying incrementalism (Padgett 1980; B. Jones and Baumgartner 2005). As Jones and Baumgartner (2005) explain, this method greatly simplifies hypothesis testing, as “any normal distribution of policy changes must have been generated by an incremental process,” whereas “any time we observe a non-normal distribution of policy change, we must conclude that incrementalism cannot have caused it” (123). This approach is especially appropriate for the study of policy diffusion. First, as the previous section described, there is a strong assumption


47

that the cumulative pattern of adoptions over time follows an S-shaped curve, with an initial period of early adopters pioneering an innovation, a middle period of more rapid uptake around an inflection point, and a final phase of uptake by late adopters (Rogers 1983; Mooney and Lee 1999). Perhaps more importantly, the patterns of adoptions are assumed to be normally distributed (Rogers 1983). Thus, a plot of the probability density function of a typical diffusion process should generate a distribution resembling a Gaussian curve, and derivation of the cumulative distribution function (CDF) should yield a plot following the CDF of a normal distribution (Rogers 1983). That observational data collected from processes of incrementalism and policy diffusion share the same underlying distribution is not coincidental. In Diffusion of Innovations, Rogers (1983) explains that the distribution of times to adoption in diffusion is normal exactly because it represents an incremental learning process. He writes: Psychological research indicates that individuals learn a new skill, or bit of knowledge, or set of facts, through a learning process that, when plotted over time, follows a normal curve. When an individual is confronted with a new situation in the psychologist’s laboratory, the subject initially makes many errors. After a series of trials, the errors decrease until a learning capacity has been reached. . . . If a social system is substituted for the individual in the learning curve, it seems reasonable to expect that the experience with the innovation is gained as each successive member in the social system adopts it. Each adoption in the social system is in a sense equivalent to a learning trial by an individual. (Rogers 1983, 244)

Thus it is not an abstraction or a matter of mathematical convenience to suggest that the underlying probability distributions of incrementalism and policy diffusion are the same. They are the same because they both measure the behavioral process of incremental learning. The diffusion of innovations model is an incremental learning model.8 Nonincremental Patterns of Policy Diffusion As described in the two prior sections, theory and empirical evidence suggest two distinct patterns of learning associated with the process of public-policy diffusion. Figure 2.8 simulates and plots the CDF of data 8

Interestingly, Rogers recognized that the diffusion of innovations did occasionally deviate from this expected cumulative frequency distribution. Likewise, Gray (1973) observed that the diffusion of public policy also occasionally departed from the S-shaped cumulative normal curve. However, neither author seemed to ask exactly what such departures indicated for the process of innovation and diffusion.


48

0.6

0.8

Policy outbreaks

0.4

Incrementalism

Hyperincrementalism

0.0

0.2

Percentage of Adopters

1.0

Theoretical Diffusion Curves

0

5

10

15 Time

20

25

30

figure 2.8. Simulated theoretical diffusion curves.

from three hypothetical diffusion processes: hyperincrementalism, incrementalism, and policy outbreaks. The two S-shaped curves show the cumulative incremental adoption curve at different rates, indicating variation in communication across units in a social system. The hyperincrementalism curve is represented by the gradually sloped thin line at the bottom of Figure 2.8. The incremental learning curve is the S-shaped distribution curve in the middle. The policy outbreaks curve is represented by the thick exponential distribution at the top of the figure. Both the hyperincrementalism curve and the incrementalism curves are simulated normal curves with different means and standard deviations indicating differences in time elapsed between adoptions.9 If the process of diffusion occurs through a process of incremental problem identification, policy evaluation, and emulation, then the observational data associated with public-policy diffusion should follow the S-shaped normal curve, although the steepness of the curve may vary.10 9

10

Both the hyperincrementalism curve and the incrementalism curves are simulated normal curves with different means and standard deviations indicating changes in time elapsed between adoptions. It is important to note that both rapid and slow policy diffusion can be normally distributed. The speed of diffusion is at least partially related to how frequently units in a


49

An alternate to the incremental curve is captured in the policy outbreaks curve, when an accelerated diffusion process results in serial adoptions resembling a policy outbreak. As with the pattern of death penalty readoptions displayed in Figure 2.1, this curve exhibits extremely rapid issue uptake as imitation leads to a sharp and immediate rise in innovation adoption. The policy outbreaks curve is represented here by a steep simulated exponential curve.11 Policies matching this distribution exhibit patterns of adoption consisting with policy mimicking rather than incremental learning. Figure 2.8 simulates and plots data from three hypothetical diffusion processes – hyperincrementalism, incrementalism, and policy outbreaks. The hyperincrementalism curve is represented by the gradually sloped thin line at the bottom of Figure 2.8. The incremental learning curve is the S-shaped distribution curve in the middle. The policy outbreaks curve is represented by the thick exponential distribution at the top of the figure. Expectations The extent to which a diffusion distribution matches the normal distribution can be used as an indicator of incrementalism in a diffusion process. Here, the assumption that the probability distributions of diffusion processes are normally distributed allows us to assess whether incrementalism is the appropriate decision-making model for policy diffusion. If incrementalism is the dominant decision-making process in diffusion, then the data from diffusion processes should conform closely to the S-shaped cumulative normal curve. If diffusion processes display punctuated dynamics, then we expect a distribution converging towards the exponential curve in Figure 2.8. Incremental Diffusion and Historical Eras In order to evaluate the consistency of nonincrementalism in diffusion research, it is important to control for temporal variation in diffusion patterns. Modern advances in communication technology have made it

11

social system come into contact with and encounter new policy innovations. As mentioned in the prior section, researchers have used Gompertz or logistic curves when S-shaped diffusion patterns have not matched the assumptions of normality. A number of alternate distributions could be used to match the rapid issue uptake of data collected from policy outbreaks. Although the exponential distribution is proposed here, other researchers might model such diffusion data using a Paretian distribution.

50


easier for decision makers to acquire and evaluate information about new policy innovations (Savage 1985a; Mossberger 2000). The proliferation of radio, television, Internet, and mass-communications technology permits the rapid flow of information, making positive feedback cycles and the rapid diffusion of innovations more common in the modern era. The importance of communications variation across historical eras suggests a second research proposition regarding the diffusion of innovations. Newer policies will display especially pronounced punctuated dynamics, whereas older policies will tend to display more gradual incremental learning patterns resembling an incremental, normal distribution. Data To evaluate patterns of policy diffusion in the American states, this investigation gathered information on state years of adoption for 133 different innovations, following a purposive sampling procedure designed to ensure a balanced representation of state public policies by historical era, policy type, and speed of diffusion. To identify innovations, this study followed Walker’s definition of an innovation as a public policy that is “new to the state adopting it” (1969, 881). The research therefore includes only innovations that were formally enacted by state governmental institutions, excluding informal policies, legal strategies, and other unofficial policy positions. Furthermore, because the study of diffusion is concerned with the timing of political decisions, coders entered information for the year of legislative enactment rather than the year of implementation. Despite these limitations, the data cover a wide range of policy types, representing economic, social, and procedural policies throughout American history. To ensure a balanced and representative sample of innovation diffusions, this research followed the lead of both Walker (1969) and Savage (1978), collecting policies representing a wide range of policy innovations. Walker (1969) collected his sample of innovations from a list of issue areas suggested by the Council of Governments’ Book of the States (Walker 1969, 882). He gathered information on state years of adoption for legislation from different issue areas: welfare, health, education; conservation; planning, administrative organization; highways; civil rights; corrections and police; labor; taxes and professional regulation (1969, 882). In constructing an even more expansive set of policies, Savage (1978) built a data set of 181 policy innovations drawn from a similar list of issue areas including “agriculture, business


51

regulation, conservation, crime and corrections, education, electoral regulation, governmental structure and operation, local government, health, professional licensing, race relations, taxation, transportation, and welfare” (214). This research followed the sampling protocol developed in each of these two earlier studies as a guide to construct a balanced sampling universe of state innovations. As explained below, it first replicated Walker’s (1969) data. When gathering new innovations to build the larger sample of 133 innovations, it followed the protocol outlined by Walker and Savage and identified policies across a similar universe of state issue areas, including civil rights, crime and policy, public health, highway and transportation safety, education, economic regulation, local government, and governance reform. The sampling procedure used to identify and collect innovations in this study is consistent with other sampling protocols used to construct large samples of state legislation in prior research on policy diffusion. Because the goal here is to evaluate how well incrementalism explains patterns of diffusion, a significant portion of the cases included in this research are drawn directly from innovations that have been featured prominently in prior research in public-policy diffusion. Of these, Walker’s replication data set provides a large portion of the data analyzed for this study. These data alone supply information on the years of adoption for 86 policies in the 48 contiguous states. The research also includes policies and years of adoption from more recent studies of policy diffusion. To identify policies for inclusion, this study identified policies through keyword searches for policy diffusion research in Expanded Academic Index and JSTOR. The data set includes research in state lotteries (F. Berry and Berry 1990); child abuse reporting and crime victims’ compensation (Hays 1996); smoking regulations (Volden and Shipan 2006); same-sex marriage bans (Eskridge 1999); death penalty reenactments (Lee and Mooney 1999); charter schools (Rincke 2004); living-will laws (Glick and Hays 1991); state medical savings accounts (Bowen 2005); no-fault divorce laws (Nakonezny, Shull, and Rogers 1995); statutory rape laws (Cocca 2002); the repeal of state sodomy laws (Disarro 2005); and hate-crimes legislation (Disarro 2005). When appropriate, it extended and updated this information through searches of state legislative statutes. Information for the remaining public policies was gathered through information provided by issue organizations, government agencies, the National Conference of State Legislatures, and keyword searches of


52

individual state legislative statutes. These policies include issues such as term limits, three-strikes laws, medical marijuana legislation, and bottle deposit legislation. A full list of policies can be found in Appendix A. To facilitate comparison of timings of adoption across decades, the data have been organized in a duration format, indicating the time each state took to adopt a given public policy. Here, the first adopters are assigned a zero (indicating immediate adoption), and subsequent adopters are indicated by the number of years elapsed between the first adoption and their own adoption of the innovation. A censoring variable is included for states that have not yet adopted a policy. To classify policies by historical era, this researcher followed the procedure suggested by Savage (1978), placing a policy in the period when the first 10 states adopted the innovation.12 Following Savage’s research on the temporal consistency of innovation scores in modern U.S. history (1978), policies were then grouped into one of four historical eras spanning the late nineteenth, twentieth, and early twenty-first centuries. This process identified 17 policies that diffused prior to 1900, 28 policies in the 1900–1929 period, 35 policies in the 1930–1959 era, and 52 policies in the 1960–2006 period. A list of policies in each of these categories can be seen in Appendix B. Method These data were employed to model the underlying distribution of times to adoption, and to evaluate whether policy diffusion can be characterized as resulting from an incremental decision-making process. This approach follows prior stochastic process studies of public policy making, which evaluate probability distributions to understand the underlying decision-making process in public budgeting (Padgett 1980) and governmental attention (Baumgartner and Jones 1993; B. Jones and Baumgartner 2005; Koski and Breunig 2006). These studies suggest that incremental processes “invariably lead to a normal outcome change distribution,” whereas nonincremental decision-making processes leading to sudden change should deviate strongly from a normal curve, following an exponential or Paretian distribution (B. Jones and Baumgartner 2005, 123;

12

To check the validity of this sorting protocol, a secondary classification sorted policies by historical era by first taking the mean year of adoption for all states,and then subtracting one standard deviation from the mean to find the year of early diffusion. This process provided an almost identical classification system to the process proposed by Walker.


53

Padgett 1980). Transferring this method to the study of decision making in policy diffusion is appropriate given the strong prior assumption that temporal patterns of policy diffusion resemble a normal S-shaped curve (Gray 1973; Rogers 1983). To measure how well diffusion data matched an incremental learning curve, this research compared the empirical cumulative distribution function (ECDF) for the aggregated pooled data of policy diffusions to a simulated Gaussian cumulative distribution of the same mean and variance as the empirical data. Here, the degree to which a plotted empirical cumulative distribution function conforms to the expected incremental or Gaussian curve can be evaluated through a visual plot. The plotted CDF of the hypothetical normal curve provides the familiar S-shaped distribution of an incremental learning process. If the empirical duration data collected in this research follows a similar distribution, then it should closely follow the simulated normal cumulative distribution function. This analysis of distributions is then extended to identify distinct patterns of diffusion across historical eras. An identical procedure is used to compare the empirical cumulative distribution functions across three groups of policies: those diffusing from 1900 to 1929, 1930 to 1959, and 1960 to 2006. Each of these distributions is then plotted alongside their expected Gaussian distribution. Again, the research hypotheses can be tested through a visual plot. Distributional Tests of Normality and Kurtosis In addition to the visual plots, several goodness-of-fit tests were employed to assess the normality of diffusion data. To examine the normality of the distributions of adoption times, this research relied on two standard, onesample tests of normality. First, this research assessed the normality of the diffusion data with the Shapiro-Wilk (S-W) test, which provides a powerful evaluation of normality and is appropriate for small-to-medium sample sizes (Conover 1999; Koski and Breunig 2006). The S-W test allows researchers to measure normality by returning both a p-value and W statistic, with smaller values for W indicating non-normality in the data. To confirm this test, the research also evaluated the distribution of diffusion data using the Anderson-Darling (A-D) test of normality. The A-D test is a powerful variation of the nonparametric Kolmogorov-Smirnov test, but gives more weight to the tails of a distribution in evaluating normality. The A-D test returns both an A statistic and a p-value for confirming whether or not a given distribution is consistent with a normal

54


distribution. Together, these goodness-of-fit tests provide a straightforward method for assessing the normality of diffusion data. To measure the general shape of diffusion distributions, this research also evaluates diffusion data for positive and negative kurtosis. Kurtosis (K) measures the “peakedness” of a given distribution, and with diffusion data can be used to indicate a positive feedback cycle from a sudden uptick in the number of individuals adopting an innovation as it gains momentum. A normal distribution is described as mesokurtic and is typically smooth and symmetrical around the mean. Deviations from a normal distribution can be represented in two common shapes. A platykurtic distribution has thin tails and is relatively flat in the center, whereas a leptokurtic distribution is indicated by a sharp central peak and fat tails. As a first measure of kurtosis, this research first employs a measure of Lkurtosis (LK), which represents the fourth moment of a probability distribution around the mean (Hosking 1990; Breunig 2006). Recent research applying distributional analysis to study the shape of distributions from the policy process has preferred LK to the more traditional measure of kurtosis, as it is less influenced by outliers and results in a more stable estimate of shape (Hosking 1990; Breunig 2006; Baumgartner, Breunig, Green-Pedersen, Jones, Mortensen, Neytemans, and Walgrave 2009). LK provides a relatively simple method for interpreting kurtosis, returning a value in intervals between 0 and 1, with values of 0.1226 representing the LK of a normal distribution. Marginally higher values are indicative of departures from a normal distribution, with an exponential distribution having an expected LK score of 0.1667 (Hosking 1990). Lower values likewise suggest non-normality, and are indicative of abnormally flat or bimodal distributions common in binomial or uniform distributions. In keeping with prior research in distributional analysis, this research also presents the standard measure of kurtosis, measuring the shape of a distribution (DeCarlo 1997). Normal distributions have kurtosis values of 3, with larger numbers indicating leptokurtic distributions and smaller, negative values indicating platykurtic, or flatter bimodal or unimodal distributions. As a final measure of the shape of diffusion data, each distribution of adoption times was measured for L-skewness. This measure of skew represents the third moment of a probability distribution around the mean and provides a final method for evaluating the shape of diffusion data. Here, normal distributions typically have L-skewness of 0. Exponential distributions produce L-skew values of 0.333 (Hosking 1990). Higher values of skew indicate a greater departure from a normal distribution.


55

Together these measures of distribution shape provide cursory measures for understanding the normality and shape of diffusion data. As a rule, distributions with higher K and LK are indicative of punctuations in the middle of a diffusion process. A higher measure of L-skewness may be indicative of more immediate policy uptake, followed by a series of later adoptions, indicative of an exponential distribution. To illustrate how values of K, LK and L-skewness change depending on the distribution of diffusion data, it is useful to refer back to Figures 2.5 and 2.7 of this chapter. The curve in Figure 2.5 represents a cumulative normal curve. The distribution underlying the curve has a kurtosis of 2.73, LK of 0.103, and L-skewness of 0.030. Figure 2.7 represents a simulated exponential distribution, with kurtosis of 6.65, LK of 0.151, and L-skewness of 0.333.13 Results Figure 2.9 plots an empirical cumulative distribution function of all policy adoptions against a simulated normal cumulative distribution curve of the same mean and variance. Here, the x-axis represents years to policy adoption, and the y-axis represents the probability of adoption. As indicated in Figure 2.9, the empirical cumulative distribution of the pooled diffusion duration data deviates sharply from the expected normal distribution. Policy adoptions occur more rapidly than expected, resembling an exponential distribution rather than the simulated normal distribution. The separation between these two curves is indicative of a nonincremental learning process and lends preliminary support for the hypotheses that policy diffusion deviates from an incremental learning process. Across cases and historical periods, policy diffusion is marked by nonincremental decision making. The non-normality of the aggregate diffusion data is confirmed by the goodness-of-fit statistics presented below in the first row of Table 2.1. Both the Anderson-Darling and Shapiro-Wilk tests permit us to reject the assumption that diffusion data is normally distributed. The two measures of kurtosis (LK = 0.120, K = 4.08) do not indicate unusually high levels of positive kurtosis; however, the measure of L-skewness suggests that the skew around the mean of 0.32 is close to the expected skew in an exponential distribution. 13

These measures of kurtosis and skew shift slightly as the mean, standard deviation, or rate of the normal and exponential curve changes.


56

table 2.1. Statistical Tests for Normality by Historical Era Historical Era

A-D Test (A) P

Policy Diffusion Dynamics in America

Theory, Policy and Dynamics in International Trade

Bounded Rationality and Policy Diffusion: Social Sector Reform in Latin America

Diffusion in polymers

Trade Policy Reforms in Latin America

Diffusion in Polymers

Diffusion in Semiconductors

Diffusion in Condensed Matter

Diffusion in Solids

Anisotropic diffusion in image processing

Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion-Controlled Processes

Surface Diffusion

Anisotropic diffusion

The Dynamics of Foreign Policy Analysis

Diffusion phenomena

The Dynamics of Social Welfare Policy

Democratic Laboratories: Policy Diffusion among the American States

The Dynamics of Social Welfare Policy

Institutional Dynamics in Environmental Governance (Environment & Policy)

America Debates United States Policy on Immigration

Nonprofits in Urban America (Policy Studies Organisation Book)

Work in America: An Encyclopedia of History, Policy, and Society

Social Policy Reform and Market Governance in Latin America

America Unbound: The Bush Revolution in Foreign Policy

Political Survival: Politicians and Public Policy in Latin America

Land & Development in Latin America: Openings for Policy Research

U.S. policy toward economic nationalism in Latin America, 1917-1929

U.S. policy in Central America: the endless debate

Institutional and Policy Dynamics of Eu Migration Law (Immigration and Asylum Law and Policy in Europe)

Policy Diffusion Dynamics in America

Theory, Policy and Dynamics in International Trade

Bounded Rationality and Policy Diffusion: Social Sector Reform in Latin America

Diffusion in polymers

Trade Policy Reforms in Latin America

Diffusion in Polymers

Diffusion in Semiconductors

Diffusion in Condensed Matter

Diffusion in Solids

Anisotropic diffusion in image processing

Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion-Controlled Processes

Surface Diffusion

Anisotropic diffusion

The Dynamics of Foreign Policy Analysis

Diffusion phenomena

The Dynamics of Social Welfare Policy

Democratic Laboratories: Policy Diffusion among the American States

The Dynamics of Social Welfare Policy

Institutional Dynamics in Environmental Governance (Environment & Policy)

America Debates United States Policy on Immigration

Nonprofits in Urban America (Policy Studies Organisation Book)

Work in America: An Encyclopedia of History, Policy, and Society

Social Policy Reform and Market Governance in Latin America

America Unbound: The Bush Revolution in Foreign Policy

Political Survival: Politicians and Public Policy in Latin America

Land & Development in Latin America: Openings for Policy Research

U.S. policy toward economic nationalism in Latin America, 1917-1929

U.S. policy in Central America: the endless debate

Institutional and Policy Dynamics of Eu Migration Law (Immigration and Asylum Law and Policy in Europe)

Recommend Documents