The European tradition of network analysis in political science differs some- what from its American counterpart. It was inspired by work in political sociology in the 1970s and 1980s. In the German and Swiss case, the same research design was employed by generations of researchers: identify the relevant organizations for a policy process, administer a network survey on information exchange or collaboration, influence attribution, venue participa- tion and other network relations, and identify the most central organizations as well as subgroups in order to reveal interest group influence on policy making. In many respects, this is valuable because numerous studies with nearly identical survey questions exist and are now amenable to inferential network analysis, a more recent methodological development (e. g., Leifeld and Schneider 2012; Ingold and Leifeld 2016). On the other hand, the inferences one can generate based on such an approach are limited because only a specific aspect of policy making is captured. A parallel development in the United States in the 1990s and 2000s was concerned with the structure of policy subsystems and the role of policy beliefs and ideas for their structure. This implies that actors' policy beliefs and verbal interactions matter for a collective understanding of a complex policy problem, an idea that is akin to the notion of political discourse. Yet, more recently, these approaches were influenced by a more collaboration- and collective-action-centered perspective and lost much of their original focus on policy beliefs. In short, the literature on policy networks and the literature on belief systems and advocacy coalitions have been increasingly merged, and the study of advocacy coalitions is now often perceived as interchangeable with the study of policy networks. This book is an attempt to overcome the methodological limitations of policy network analysis and operationalize the relational elements hidden in political debates. As it turns out, policy debates are complex and dynamic systems that need to be analyzed with scientific scrutiny. The time has come for a more rigorous approach to studying political discourse than the hermeneutic approaches that have been prevalent in the last decades. Only quantitative, relational methods, coupled with a (possibly qualitative) bridge from text to data, will permit a systematic study of policy debates. After receiving my master's degree in Politics and Public Administration at the University of Konstanz in 2007, I had some experience with policy networks and related approaches. Before I started my doctoral studies at the Max Planck Institute, I co-edited a volume on policy networks (Schneider et al. 2009). For one of the chapters, Volker Schneider at the University of Konstanz advised me to look into ways that network analysis could be combined with the notion of discourse. This was a very vague idea that needed to be developed into something that other people could actually use in their own research. For the time being, I contributed ideas to a joint review chapter of existing work with my co-editors (0anning et al. 2009). In the same year, I joined the PhD program of the Max Planck International Research Network on Aging (MaxNetAging) at the Max Planck Institute for Demographic Research and the Max Planck Institute for Research on Collective Goods. I soon developed an interest in the politics of demographic change and old-age provision, a topic that was both compatible with my newly developing interest in aging research and demography, and my background in the study of politics and policy networks. After talking to a variety of experts on demography and politics, I realized that organized interests were playing important roles in the politics of demography and old-age security, and that one of their main strategies was the deliberate use of the media and other venues to frame the pension debate in ways that supported their material interests. At the time, demographers thought senior citizens' interest groups and peak associations were some of the most influential players in the politics of demographic change. At some point, however, I realized that demographic change was only a phenomenon that caused the debate, but the debate was actually about the future design of a sustainable pension system, one of the subsystems most severely affected by demographic change. It turned out that other types of interest groups like financial market actors and employers' associations were apparently playing a more important role than senior citizens' interest groups in the important reforms of the last decades. The problem was that existing methods like the survey-based policy network approach or approaches related to policy beliefs were not sufficient to fully capture the dynamics of the debate. I turned to my previous work on discourse networks and started working on a more comprehensive methodological approach. What I wanted was a methodology that would tell me what competing advocacy coalitions or discourse coalitions looked like at any point in time, how they changed over time, and how some actors left their coalitions and joined the political opponent. Later, I also became interested in the behavioral mechanisms that were driving these changes at the micro-level of a debate. Therefore I started combining my existing knowledge on network analysis, policy networks, political discourse, policy beliefs, and programming in order to come up with such a methodology and apply it to German pension politics in order to explain the policy changes that came about in recent years. The results of these developments, which are also the results of my PhD work, are presented in this book. On the way from the initial idea to the product presented in this book, I received valuable input from a number of people and organizations. Volker Schneider, Professor of Empirical Theory of the State at the University of Konstanz, triggered my original interest in the role of ideas and policy beliefs in policy networks. He also became my doctoral advisor. Christoph Engel and Martin Hellwig, the Directors of the Max Planck Institute for Research on Collective Goods in Bonn, realized my potential when they chose to hire me as a PhD student over candidates from economics and other disciplines to which the Max Planck Institute usually provides a home. Christoph Engel became my doctoral co-advisor. I am greatly indebted to him for this exciting and sometimes challenging opportunity to grow up in a truly interdisciplinary environment. Christoph Knill, then professor in Konstanz and now Professor of Political Science at the University of Munich, served on my committee as the third reviewer. Without the support of staff and colleagues in the MaxNetAging program, as well as generous funding of my research through MaxNetAging, this research would have taken a different, possibly less ambitious direction. In the context of MaxNetAging, I appreciate the extensive discussions on my topic and the connections and institutional resources I was able to use. A bottleneck of any discourse network analysis is the manual coding effort required to annotate thousands of political statements. My student assistant Frank Kaiser supported me with this challenging task and provided excellent research assistance to this project. Research findings can only be important if there is a demand for them. I wish to thank the numerous people who have used my methods and companion software DISCOURSE NETWORK ANALYZER in their own research and who have provided feedback and reported bugs, especially Dana R. Fisher (University of Maryland, College Park), 0effrey P. Broadbent (University of Minnesota), and other members of the Comparing Climate Change Policy Networks (COMPON) project, where discourse network analysis could be employed in a comparative setting. The dissertation won two prestigious prizes in 2013. I am deeply grateful for this unexpected honor: In April, I received the Südwestmetall Award, which was sponsored by Südwestmetall, the Employers' Association of the Metal and Electrical Industry in Baden-Württemberg and one of Germany's largest employer federations. The award is presented annually for academic theses of particular importance to the industrial workplace and/or its social- political conditions. In December, I received the Dissertation Award of the Foundation Science and Society at the University of Konstanz ('Stiftung Wissenschaft und Gesellschaft an der Universität Konstanz'), which is a prize for the best dissertation defended at the University of Konstanz in the previous academic year. Frank Nullmeier, Professor at the Center for Social Policy Research (Zentrum für Sozialpolitik) at the University of Bremen, encouraged me to publish this monograph after inviting me to Bremen to give a talk about the German pension debate. I appreciate his support and the opportunity to publish this work in the Series 'Studies in Social Policy Research.' The Research Network of the Statutory Pension Scheme ('Forschungsnet- zwerk Alterssicherung - FNA') provided generous funding for the publication of this monograph and gave me the opportunity to use their outreach and dissemination channels to make my work accessible to a broader public. The contribution of FNA helped me to cover parts of the costs associated with the publication. Finally, in every PhD project, there are ups and downs. I particularly wish to thank my wife Miriam and my parents for their invaluable support during this time. Konstanz, February 2016Philip Leifeld I.The Theory and Methodology of Discourse Networks 1. Introduction There are many explanations for political outcomes like reforms or status-quo orientation in a policy sector. A subset of these public policy theories is based on ideas, interests, and language. The phenomenon leading to political outcomes in this ideational branch of literature is often called 'political discourse'. Other names with slightly different connotations are 'policy debates', 'policy deliberation' and 'policy learning'. Hereafter, all of these names shall be used interchangeably. A critical element in many approaches to political discourse is endogeneity. Preferences of political actors, such as interest groups or politicians, are not exogenously given. They rather 'emerge' as a result of communication processes and are as such endogenous. Endogenous preferences are in stark contrast to many economic models of preference aggregation or political action, which assume that actors base their strategies rationally on their predefined preferences. In this book, I present an empirical model of political discourse that does not make any prior assumptions about endogeneity or exogeneity of policy preferences. It rather serves as a measurement device for assessing what a specific discourse looks like. From this starting point, theoretical and empirical explorations into the mechanisms behind the observable macro phenomenon are possible. The approach I have developed is called 'discourse network analysis'. It is based on the premise that discourse is a relational phenomenon, which means that actors mutually influence each other (in the endogenous conception of discourse) or at least show certain degrees of similarity or dissimilarity regarding their preferences (in the exogenous version of the phenomenon under scrutiny). A straightforward methodological toolbox for the analysis of discursive structures is therefore social network analysis (Wasserman and Faust 1994). The book is structured as follows. There are three parts, each associated with a specific goal in mind. The first part deals with the empirical opera- tionalization of existing public policy theories. Chapter 2 gives an overview of all relevant public policy theories that are both relational and concerned with ideas or interests. Chapter 3 presents an array of methodo!ogica! approaches to the relational measurement of ideas or interests. I argue that there is a void between the theoretical approaches, which focus mostly on actors and their behavior, and the methods, which mostly deal with the contents of a discourse and neglect actors. The conclusion from the literature review (Chapter 4) therefore synthesizes existing approaches and proposes a canon of requirements for an improved methodology that can operationalize relational and ideational public policy theories. Chapter 5 proposes such a new methodology for the analysis of political discourse. For any policy debate, a set of six different network represen- tations can be computed: affiliation networks, actor congruence networks, conflict networks, concept congruence networks, time window networks, and attenuation networks. All methods are discussed in detail, and a software implementation in a program called DISCOURSE NETWORK ANALYZER (DNA) is briefly introduced. Some of the core findings of this chapter have been published in Leifeld (2016). The second part of this book contains a showcase for the methodology elaborated in the first part. German pension politics constitute an ideal case study. It has been subject to a vast amount of hermeneutic public policy analysis. The 2001 Riester reform departs significantly from previous policy trajectories. The political science literature contains many ideational explanations for this 'paradigm shift', the explanatory power of which can be assessed by conducting a discourse network analysis. Chapter 6 introduces the case of pension politics in Germany. It first discusses several theoretical dimensions of pension systems and then classifies the pre- and post-Riester pension system on these dimensions. By drawing on theories of demographic change, the complexity and uncertainty that political actors face is illustrated. These uncertainties constitute the actual significance of ideational explanations. The final section of this chapter draws on the political science literature regarding the pension system and particularly the Riester reform. Several propositions about the development of the discourse and the actor structure over time are distilled in order to formulate them in an empirically testable way by means of discourse network analysis. Chapter 7 discusses the data source, the coding process and potential validity issues with the news media data used for the analysis. Moreover, the categories used for the content analysis are explained in great detail, and summary statistics are given for each category. Chapter 8 proceeds with the analysis of German pension politics between 1993 and 2001. After summarizing some general trends over time, a cross- sectional analysis of the cleavage lines in the discourse is conducted. The discourse is clustered both at the actor level and at the level of the contents of the discourse, eventually combining both perspectives. There is evidence for the existence of two specific cleavage lines ('public pay-as-you-go system' versus 'private pension system', and 'labor' versus 'capital'), while other cleavage lines like 'old' versus 'young' are not visible in the data. Moreover, three distinct ideologies can be identified, which correspond to specific actor groups at the individual level. Finally and most importantly, the pension discourse is analyzed in a dynamic way. There is clear evidence for the existence of a closed policy community in the mid-1990s, which is eventually 'cracked up' and replaced by a new advocacy coalition centering around actors from the financial sector around the year 2000. Discourse network analysis and the software implementation are able to provide qualified answers to the questions posed by the theoretical or purely hermeneutic approaches presented in the previous chapter. Parts of this chapter have been published in Leifeld (2013). The advantage of the methodology presented in the first part of the book is its openness for several paradigmatic conceptions of preferences (endogenous versus exogenous). At the same time, this generality implies a potential disadvantage: while the aggregate structure of a discourse can be reliably measured and analyzed, the data-generating process largely remains a black box. The third part of the book therefore tries to abstract from the specific case study and infer general properties of political discourses. The overarching goal is to develop theoretical models that accurately reflect the aggregate structure of the discourse as it could be observed empirically in the previous part. Macro-outcomes can be explained by describing micro- and meso-level mechanisms, and micro mechanisms are in turn embedded in a macro structure (Bunge 1996: 264 ff.). Following this spirit, the micro- and meso-level mechanisms behind political discourse are modeled in a bottom-up fashion such that a macro structure emerges that is indistinguishable from the macro structure found in the empirical case study. Chapter 9 employs the attenuation algorithm, one of the methods de- veloped at the beginning of the book, as a measure of ideational contagion in the pension discourse. By controlling for preferences and institutional actor roles, and after matching the dataset with the policy network dataset of Pappi et al. (1995), ideational contagion can be modeled as a function of interest group influence and regular information exchange between actors. An exponential random graph model with dyadic dependence is employed to study who reacts to whom in the pension discourse. Endogenous belief adoption is present, even when controlling for various other effects, and the channels over which mutual reinforcement occurs between actors are identified. Chapter 10 abstracts almost completely from the pension case study. A formal model of political discourse with exogenous preferences and endogenous belief adoption is presented, and new metrics for the analysis of the model over time are introduced. The implications of the agent-based model are simulated over 10,000 rounds. A comparison between the macro structure of the simulated discourse and the empirical discourse presented in the preceding chapters allows for a validity check. While any single mechanism in the utility function of an agent yields unrealistic results, only a combination of exogenous preferences and endogenous belief adoption provide a sufficiently good match between the theory and the empirical data. This finding suggests that the recurring controversies between proponents of rational-choice theory and proponents of constructivist explanations are largely unwarranted because both elements are necessary to explain the structure of political discourses. A revised version of this chapter has been published in Leifeld (2014). Finally, Chapter 11 provides a summary of the most important findings in this book and discusses some promising avenues for future research.