The Pennsylvania State University The Graduate School College of the Liberal Arts

PARTY COALITIONS, PARTY IDEOLOGY, AND PARTY ACTION:

EXTENDED PARTY NETWORKS IN THE UNITED STATES

A Dissertation in Political Science by Kevin Reuning

© 2018 Kevin Reuning

Submitted in Partial Fulfillment of the Requirements for the Degree of

Doctor of Philosophy

August 2018 The dissertation of Kevin Reuning was reviewed and approved∗ by the following:

Lee Ann Banaszak Professor of Political Science Dissertation Advisor, Chair of Committee

Michael Berkman Professor of Political Science

Bruce Desmarais Associate Professor of Political Science

Michael Nelson Associate Professor of Political Science

John McCarthy Professor of Sociology

Glenn Palmer Professor of Political Science Director of Graduate Studies

∗Signatures are on file in the Graduate School.

ii Abstract

American political parties are not singular entities, but webs of interests that come together to gain power and implement policy. This has been noted by recent work, but there has been little theoretical focus on the implications of this parties as networks approach. My dissertation unpacks what it means for political parties to be networks and what the implications of this view are. I argue that because political parties are networks, the relationships that exist between groups within the network are critical in explaining variation in party ideology across the state parties. In addition, I argue that fracturing of a party network outside the legislature leads to a similar fracturing of the party caucus inside the legislature. To test these theories I use state legislative donation data from 2000 to 2016 to develop state donation networks. Using these networks I first show that relationships help to explain party ideology even when controlling for resources. I then use Exponential Random Graph Models to measure the degree of cohesion/fracturing within a party network. I find that this is an important predictor of legislative cohesion for Democrats and not for Republicans. My findings have important ramifications for democracy in the United States. In particular it demonstrates that solutions over unequal representation cannot just focus on the role of money in politics, as relationships are just as important, and are not solely a function of resources. In addition it helps to explain how parties in the United States often have unsteady paths forward, moving quickly to change policy positions after a long time of stasis.

iii Table of Contents

List of Figures vii

List of Tables ix

Acknowledgmentsx

Chapter 1 Introduction1 1.1 Why Parties Matter ...... 2 1.2 What Are Parties? ...... 5 1.2.1 Parties in the States ...... 9 1.3 The Organization of Party Coalitions...... 11 1.4 Outline of the Dissertation...... 12

Chapter 2 What are Political Parties? 14 2.1 Parties as Networks...... 15 2.1.1 Demands for Policy...... 16 2.1.2 Coming to Agreement ...... 18 2.2 Party Networks and Ideology ...... 22 2.2.1 What Ideology Achieves ...... 23 2.2.2 What is the Party Ideology Going to be?...... 24 2.3 Parties In and Outside the Legislature ...... 27 2.3.1 From Outside In ...... 28 2.3.2 Party Role Inside the Legislature ...... 32 2.4 Conclusion...... 33

Chapter 3 Identifying the Party 35 3.1 Identifying the Party Network...... 36

iv 3.2 State Level Donation Data...... 39 3.2.1 Inferring Relationships...... 43 3.3 Validation...... 51 3.3.1 Republican and Democratic Networks...... 51 3.3.2 Groups with Similar Interests ...... 54 3.4 Conclusion...... 57

Chapter 4 Party Ideology 59 4.1 Introduction...... 59 4.2 Party Coalitions and Ideology...... 61 4.2.1 Differences Between Republicans and Democrats...... 63 4.3 Testing the Effects of Network Position...... 65 4.3.1 Policy Demander Relationships and Resources...... 65 4.3.2 Party Position...... 69 4.3.3 Modeling Relationship Between Party Networks and Ideology . . . . 71 4.4 Results...... 73 4.4.1 Predictive Accuracy...... 77 4.5 Discussion and Conclusion...... 80

Chapter 5 Party Cohesion 83 5.1 Introduction...... 83 5.2 Division in Kansas ...... 84 5.3 Where Does Cohesion Come From?...... 86 5.3.1 How Cohesion Matters...... 89 5.4 Testing the Effects of Cohesion ...... 94 5.4.1 Measuring Cohesion Outside the Legislature...... 94 5.4.2 Measuring Cohesion Inside the Legislature...... 97 5.4.3 Controls and Modeling Assumptions ...... 98 5.5 Results and Discussion...... 101 5.6 Conclusion...... 104

Chapter 6 Conclusion 107 6.1 Summary ...... 107 6.2 Broader Implications...... 109 6.2.1 Representation and Democracy ...... 109 6.2.2 Connecting Parties...... 112 6.3 Limitations ...... 113

v 6.3.1 What Contributions Miss ...... 114 6.3.2 Party Networks: Cause or Effect?...... 115 6.4 Future Work...... 116 6.4.1 The Creation and Destruction of Party Networks ...... 117 6.4.2 Activist Networks Versus Party Networks ...... 118

Appendices 120 A Technical Details ...... 121 B Controlling for Donation Limits...... 127 C Exponential Random Graph Models...... 130 C.1 Nodal Terms ...... 131 C.2 Geometrically Edgewise Shared Partners and Extensions ...... 132 C.3 Maximally Weighted Edgewise Shared Partner ...... 134 C.4 Estimation...... 136

Bibliography 138

vi List of Figures

3.1 Number of Candidates vs Groups in Each State-House-Full-Cycle ...... 42

3.2 Percentage of Non-Zero Donor-Candidate Pairs ...... 42

3.3 Median Donation in Each Dataset...... 43

3.4 Examples of Full Donation Networks ...... 47

3.5 Examples of Democratic Donation Networks...... 49

3.6 Examples of Republican Donation Networks...... 50

3.7 Assortative Mixing by Party...... 53

3.9 Measures of Homophily...... 56

4.1 Network Relationship Strength Scores...... 68

4.2 Distribution of Party Ideology...... 70

4.3 Predicted Ideology ...... 79

4.4 Repeated 10-Fold Predicted RMSE...... 80

5.1 Subset of Kansas Lower House Donation Network (2013-2014) ...... 96

5.2 Density of Party Network Cohesion...... 98

vii 5.3 Density of Legislative Cohesion ...... 99

5.4 Predicted Legislative Cohesion...... 105

A.1 Real Donations from a Sample of States ...... 123

A.2 Estimated Coefficients for Sample Models ...... 124

A.3 Predicted versus Real Donations from a Sample of States ...... 125

A.4 Number of Edges Pre and Post Backboning ...... 126

B.1 Repeated 10-Fold Predicted RMSE with Donation Limits ...... 129

C.1 Examples of Edgewise Shared Partners...... 134

C.2 MWESP Examples...... 135

viii List of Tables

2.1 Collaboration Game for Intense Policy Demanders...... 21

2.2 Collaboration Game with Ideology ...... 24

4.1 Analysis Party Caucus Ideology...... 76

4.2 Effects Conditioned on Party...... 78

5.1 Analysis of Party Cohesion...... 102

5.2 Model for Party Cohesion Conditioned on Party ...... 103

B.1 Analysis Party Caucus Ideology with Donation Limits ...... 128

ix Acknowledgments

Dissertations are group projects with a single name attached. This project would not have been successful without the work of so many. I want to first thank my committee for reading many drafts, guiding me through a complicated process, and helping me better understand how to fit my research into the academic world. A special thank you to Michael Nelson for reading a lot of versions of the content here and for rapidly providing comments, along with always allowing me to walk into his office and ask random questions. A thank you as well to Lee Ann Banaszak for always pushing me to think more about the theory and pulling my head out of the data when it needed to be pulled out (and for allowing me to spend a lot of time with the data). And finally, thank you to John McCarthy for bringing me onto his projects when I still did not really understand research and showing me the sociology side of social movements. The dissertation is the capstone of graduate school which is itself a team event. I owe a debt of gratitude to many of my fellow travelers through graduate school: Ted Chen, Nick Dietrich, Mike Kenwick, Sarah Liu, Dan Mallinson and Wonjun Song to name only a few. They listened to me complain, while helping to work through both the academic and non-academic problems that graduate school poses. I want to thank as well all the leaders in the Coalitions of Graduate Employees. We did not win a battle, but I hope they will keep fighting. The Coalition of Graduate Employees helped me to not only remember a world outside of the confines of academic political science, but to shape my views towards how to better do political science. There are also so many who helped in small, and easy to overlook ways. Thank you to my Aunt Jill for reading over my job market materials. Thank you to my parents for allowing me to explore things. Thank you to my sisters for teaching me so much. Thank you to all the baristas of State College for not kicking me out of your cafes. Finally, none of this would have been possible without Anne Marie Whitesell. I am so glad we found each other. Thank you for listening me ramble about networks, and latent variables. Thank you for keeping me thinking about things outside of the narrow confines of my projects. Thank you for allowing me to have a JoJo in my life. And thank you for being there with me through everything. I cannot wait to begin our next journey together.

x Chapter 1 | Introduction

Political parties are an inescapable part of American politics (Schattschneider 1942). They define the terms of debate over almost every issue. Political pundits increasing point to them as the source of almost everything that is wrong with American politics. Politicians are said to only care about winning partisan battles. And many candidates run campaigns that focus on their willingness to buck the party establishment. But, it is often difficult to identify what a party is. Elected officials can be identified with their party of choice, but it seems unlikely that political parties can only be found in legislative chambers. Every four years the national parties come together for party conventions, and those deeply involved in politics can list off an alphabet soup of acronyms associated with parties. The Democratic Party has the DNC, DCCC, DSCC, DGA, and the DLCC, to name just some of the national organizations. Again, this identification of party organizations or practices does not seem to fit with what we mean when we talk about parties. Just as we know that parties are inescapable it seems increasingly obvious that parties are internally fractious (Drutman 2017). In 2010, the Tea Party set off a civil war within the Republican Party, while in 2016 the nominally Democratic Bernie Sanders was seen as setting off a similar war within the Democratic Party. What are political parties then? How are they structured? And how does that structure matter?

1 Those are the questions that this dissertation attempts to answers. I argue that the relationships between a wide coalition of groups that make up the political are important for understanding the positions of parties, and the actions of their elected co-partisans. Groups within parties build relationships with each other and use these relationships to push and pull the party to their own positions. Similarly, when fractures build within parties they reveal themselves by impacting how elected officials of that party work together. By answering these questions I am not only trying to explain what political parties are but how they shape the political world we exist in. The most important take away for this is that the political world is not only a function of resources but also a function of relationships. Relationships allow groups to build broader coalitions and draw focus on issues that others might not be concerned about. The importance of relationships in politics can also mean that those with divergent preferences are excluded from the political process even when they have resources. Before discussing these ideas in full it is useful to first touch on why political parties matter, how they shape the political world that we currently inhabit. After doing this I will layout the general idea of political parties that is used here. The idea is not new, but the implications I draw from it are. I discuss in brief what these implications are, before discussing the outline of my dissertation.

1.1 Why Parties Matter

There are two ways to approach the importance of political parties in modern American democracy. Empirically it is hard to fathom American politics without political parties (Schattschneider 1942). The language of politics is often defined through a language of parties. Partisan polarization is a defining trait in the public and elected officials (McCarty, Poole and Rosenthal 2016). Parties are so deeply embedded that candidates often frame campaigns

2 in opposition to the Republican or Democratic establishment. In understanding the empirical role of parties it is useful to use Key’s (1949) tripartite definition of parties: parties in government, parties in the electorate and parties as organiza- tions. In each of these places parties are inescapable. Parties in government can be easily identified just by examining the number of independent or third-party candidates. In 2016, of the 1,972 state senators only 3 had run as neither Democrat or Republican. 27 of of the 5,411 state representatives were similarly independent or third party candidates (The Council of State Governments 2017). Even in Nebraska where elections are non-partisan, party politics has increasingly defined what happens within the state legislature (Masket and Shor 2015). Parties in the electorate remain influential as well. One way of looking at this is by examining how party identification explains vote choice: does knowing someones party mean you can guess who they will vote for? The answer is yes, and since World War II the relationship between party identification and vote choice has only increased (Campbell et al. 1960; Bartels 2000; Weinschenk 2013). In addition more and more voters vote straight-ticket, selecting only Democrats or only Republican candidates to vote for (Abramowitz and Webster 2016). The impact of parties as organizations is hardest to quantify, and is the focus of this dissertation. Briefly, just an examination of formal party organizations shows important influence on the electoral process (Gibson et al. 1983; Mayhew 1986). In the 2016 election party campaigns registered at the federal level raised $941 million, dispersing this to candidates and spending it themselves on get out the vote efforts. Empirically it is hard to imagine a world without political parties. This leads to the second way to approach the role of political parties in the United States: are they good for democracy? Political parties in the United States have had a long and often inglorious history. Famously, President Washington spoke out against political parties in his farewell address saying: “the common and continual mischiefs of the spirit of party are sufficient to make

3 it the interest and duty of a wise people to discourage and restrain it” (Washington 1796). Similar concerns of parties permeate the Federalist Papers. Federalist Number 10 is in part a defense of the constitution by arguing that it can protect the nation from the “mischief of factions” (Madison 1787). Political parties were seen as corrupting influences that would shift policy away from what was the best for the nation.1 This early work generally viewed political parties as involving elite politics, although it did not focus on clearly delineating what political parties are. This negative view of parties was challenged in the 1940s by both political scientists and economists. Political parties were argued to be necessary for modern democracy as they competed with each other for votes. This competition was critical as most voters were unlikely to be directly involved in the political process and did not have well developed opinions (Schattschneider 1942; Schumpeter 1942). At this time one of the dominant views of parties was also developed. Schattschneider defined parties as teams of politicians that wanted to gain political power. He did not argue that there was any particular reason that they wanted power and that it was perfectly fine for those within the party to want power for their own personal ends. This was influential as it explained how parties could be good for democracy without requiring that parties be noble. In framing debates around issues in the light of parties, voters use heuristics to understand the different positions. This is crucial as modern democracy can demand a lot from voters, and so a party label can ease the burden and allow under informed voters to participate without the expensive cost of gathering information. In this defense of parties though is an assumption that political parties will work to represent majorities of voters and so party brands will be useful. At their best parties are aggregators of interests that make it easier for voters to participate by defining debate around those aggregated interests.

1This position was not universally held. Edmond Burke was an important exception. He understood political parties as oriented toward the national interest and so thought that they ought to be a positive force in politics (1770).

4 This defense of parties requires parties to aggregate interests well, and to clearly define themselves around those interests. Historically there has been questions about parties ability to live up to both of these demands (Committee on Political Parties 1950). Polarization between parties has led parties to do a better job of providing clear definitions of themselves to voters (Rae 2007). Yet there is reason to believe that parties better represent the interests of affluent voters (Rigby and Wright 2013; Schlozman, Verba and Brady 2012). Parties might be aggregating interests and providing clear delineation between themselves and others, but these interests might not reflect everyone equally. In order to argue for parties it is necessary to explore how parties aggregate the interests of voters. This is a question of organization or structure, how parties are organized around the interests of voters. Are all voters equally represented in the organization of parties or are some voters better represented? And how is the organization of parties reflected in what political parties do? If it is found that the interests within a party are equally represented then there is more evidence that parties are good for society. In contrast, if power structures within parties impact what parties do, and if these power structures reflect inequalities in society, then parties only heighten the political inequality within American politics.

1.2 What Are Parties?

To answer these questions about the structure of political parties we need to return to the broader question about what political parties are. The tripartite structure provided by Key (1949) provides a useful way of examining political parties in different political arenas but does not provide the necessary analytical traction for understanding the internal functions of political parties such as how they select candidates or distribute resources. Currently there are are two major understandings of political parties. The first comes from Aldrich and Rohde(2001) and expands on the the idea of political

5 parties as teams seeking power. His focus was again on how parties are the result of candidates and politicians that are interested in gaining office. Aldrich argues that political parties solve problems commonly faced inside and outside the legislature by these individuals. Parties therefore develop out of the demands of candidates and are responsive to the needs of candidates. This candidate centered view of parties has been a dominant paradigm in how parties in America are understood and analyzed. A decade later an alternative understanding of parties was put forth by Cohen et al. (2008). Cohen et al.(2008) argued that groups with interests in policy were the constituent pieces of parties. These groups, interested in implementing policy, realized that they could work together and so formed parties. Candidates are selected by the coalition of interests and these coalitions work to enforce party discipline. The relationship between groups and candidates can be fraught as candidates try to defect from the party coalition. In contrast to Aldrich this reverses the relationship between parties and candidates, political parties are prior to the candidates. The framework of Cohen et al.(2008) came to be known as the extended party network or parties as networks approach to parties due to its focus on the network of interests that exist around political parties. This is the paradigm that I will be working with throughout the book, and will explain in full in the later sections of this chapter. It is worth discussing now the different implications of these two paradigms and why I opt to use the parties as network approach. The chief observable difference between the two theories of parties is whether candidates use parties or if parties use candidates. In the context of Aldrich(1995) parties are a tool of candidates and so will respond to the needs of candidates. Parties should never restrain candidates but, candidates can, when they want to, restrain parties. The theory of parties as networks instead implies that political parties force candidates and elected officials into positions they don’t necessarily want to take (Masket 2011). At times candidates and

6 anti-party movements have restrained parties through legislation but political parties evolve around these rules (Masket 2016). A growing body of evidence supports the idea that political parties play a critical role in selecting candidates, that they push candidates and elected officials into positions they would not take, and that when politicians can defect from party discipline they do. Bawn et al. (2012) outlines much of this evidence. Two pieces are worth discussing here. First, they show that party elites play a major role in selecting candidates. This has been bolstered by other work of theirs and others that show the role of party elites in selecting candidates (Cohen et al. 2008; Desmarais, La Raja and Kowal 2015; Hassell 2016; Masket 2016). The original focus in this line of research was looking at elites in the presidential election and showed that in most, but not all, the candidate with the most elite support before any primary elections take place is selected as the eventual nominee (Cohen et al. 2008). Other work has expanded on this to other elections and used more technical means to test if there are other driving forces leading to candidate support. Hassell(2016) tests the role of party donors in Senate elections. He defines these as individuals that are donating to both party organizations and Senate candidates. Hassell(2016) finds that candidates that are coordinated on by party donors early have a significantly better chance of winning primary elections than those without the coordinated support of party donors. In addition he shows that party donors are not just coordinating on the most viable candidates. This demonstrates that the party network, in this case defined as party donors, play an out-sized role in selecting candidates. Second, Bawn et al.(2012) discuss polarization in Congress. In particular they argue that Members of Congress from both parties hold more extreme positions than their constituents. It is difficult to explain this from Aldrich’s perspective, especially as candidates that take extreme positions are more likely to lose their seat (Canes-Wrone, Brady and Cogan 2002). When these extremist candidates lose they are not replaced by moderates, but by extremist candidates from the other party (Bafumi and Herron 2010). In contrast, extremist candidates

7 are not surprising given the paradigm of parties as networks. Candidates reflect the interests of the party network, and are only partially constrained by the preferences of voters. Party activists have been previously shown to have extreme positions on many issues (Layman, Carsey and Horowitz 2006). Although party activists are not the sole component of the party network they are an important part. A third piece of evidence, not discussed as much by Bawn et al.(2012), is how candidates often try to defect from political parties when they have the tools to do so. Masket(2016) shows in multiple instances how candidates have, when given the opportunity, escaped from political parties. For example, non-partisan state legislative elections in Minnesota provided a temporary tool for legislators to defect from parties and vote based on their own or constituent interests. They did this for years until party elites found ways to work around the electoral rules. A similar learning process has taken placed in Nebraska which also have non-partisan elections for the state legislatures. The implementation of term limits there allowed party networks to start enforcing party discipline on candidates which they had previous defected from Masket and Shor(2015). The fact that candidates and elected officials so often defect from their own party brand is not easily explainable from the Aldrich perspective. If candidates create parties, then they should not try to avoid parties. In contrast, if parties are foisted on candidates, as they are in the party network perspective, then it is unsurprising that candidates look for ways to escape from parties. I rely on the party network approach in this work because of the growing evidence that shows that party networks play an important role in selecting candidates; that the Aldrich model cannot explain certain aspects of observed phenomenon, such as polarization; and that candidates often attempt to defect from parties when they have the tools to do so. This is not to claim that candidates have no influence on the party, as will be discussed in full they also play an important role, but I do emphasize the position of party networks in understanding political parties.

8 1.2.1 Parties in the States

Most of the work discussed above focuses on parties at the national level in the United States. Parties though are federated structures with a presence at the federal, state and local level. State parties play a similar role to national parties, structuring state level political debate, organizing legislative caucuses, and electing candidates. They do all of this in varying contexts which has lead to differences across political parties. Key(1949) noted this in his investigation of the one-party South. The Democratic state parties differed drastically from the federal Democratic Party, not just in their positions but also in their organizational make up. Mapping differences in state party organizations has been task taken up by different political scientists in the decades following Key(1949). Two significant pieces of research emerged in the 1980s, one focused on party typologies and the other focus on formal party organization. The first is from Mayhew(1986), where he developed the idea of traditional party organization (TPO) to identify particular types of party organizations in states. TPOs had to be independent from other interests, participate actively in nomination processes, have a well developed internal structure and rely “substantially on ‘material’ incentives, and not much on ‘purposive’ incentives, in engaging people to do organization work” (Mayhew 1986, p 20). Mayhew then analyzed the 50 states and scored them on how well they fit this TPO ideal and so identified an extensive degree of variation. Although there might be concerns with how Mayhew identifies party organization, his work demonstrates how varied party organizations can be within the United States all at the same time period. The second piece of research is from Gibson et al.(1983). They focus their attention on the organizational strength of parties, arguing that the resources and structures of formal organizations are important for understanding political parties. Through surveys of state and sub-state party leadership they identify variation in organizational strength across states,

9 time, and between parties within the same state. They find that party organizations were stronger in the 1980s than they were in previous decades. This is an unexpected result given that most political scientists at the time saw parties as becoming weaker. In later work they examine how party organizational strength affects political outcomes. They find that party organizational strength is not strongly correlated with general electoral success, although there is a relationship with gubernatorial success, and that there is a negative relationship between voter identification with a party and the strength of that party organization (Cotter et al. 1989). In the decades since there have been a handful of attempts at tasks of identifying and categorizing differences in party organizations or structures across the states (Gimpel 1996; Gimpel and Schuknecht 2009; Roscoe and Jenkins 2015). They find, in line with the previously discussed work, wide variation across states in party structures and resources. Unlike the work I present here, the previous work on state parties has almost solely focused on formal party organizations, and not on the broader network of interests that exists around state parties. This work is useful nonetheless as it provides tentative evidence that there is meaningful variation across the states. There has been a parallel, although disconnected, track in research that has focused on the fact that the Democratic and Republican Party labels have divergent meanings across the states. The conservative stances of the Southern Democratic Party have been well documented, but this is only an extreme case of a common characteristics. Researchers have documented variation in ideological preferences of citizens and elites across states, including elites of the same party (Berry et al. 1998; Shor and McCarty 2011). This research has primary focused on difference across state legislatures, narrowing in on examining the causes of polarization and the effects of primary nomination systems (Barber 2016; McGhee et al. 2014). Work using the paradigm of party networks, the paradigm I embrace here, has mainly focused on how changes to network structure changes the dynamics within legislatures over

10 time (Masket 2016; Masket and Shor 2015). In this dissertation I look to contribute to the research on variation in parties across states but use the framework of party networks to take a broader understanding of parties. By using party networks as a starting point I provide more analytical clarity. I am able to identify what factors of party networks are important, and how these factors can explain differences in partisan ideology as well as differences in other actions by party elites.

1.3 The Organization of Party Coalitions

I embrace the theory of parties as networks or coalition of interests, but what does this imply for parties? I argue that in order to understand parties we need to understand the relationships between groups within the party coalition. These relationships are what make up the party network, and what groups within the coalition use to influence each other. This affects two aspects of political parties: the ideology of the party and the coherence of the party. I explain these implications in full in the next chapter, and here I summarize the important aspects of them. Party coalitions exist in a latent state. Researchers do not observe the coalition as a whole, but they instead can observe relationships and shared actions between coalition members. The relationships between coalition members form a social network, where groups can be understood based on what sorts of relationships they have with other members of the coalitions. Identifying these relationships is itself complicated. Groups do not declare who they have relationships with, but it is possible to infer relationships between groups based on their actions. Relationships are important as they can be used to pass on private informations, sway the preferences of others, and coordinate around candidates. The relationships between groups then defines the preferences of the political party, as groups use their relationships to bring

11 others within the party closer to their own preferences. The connection between relationships and political ideology will be an important argument of this dissertation. Relationships are also important when they are absent within a party coalition. Coalitions where there are not dense webs of relationships across many groups will have more difficult time working together. This will be most apparent when there are factions within the party. These factions within the party coalition will impact the type of legislative caucus they nominate. Fractious parties will nominate legislative caucuses that are not well united around issues. This create a party caucus that is potentially incoherent which will have implications for the strength of the party in the legislature. The fact that relationships matter so much gives direction in answering the normative questions above. Relationships are hard to regulate, and groups with more resources or have a more privileged placed in politics will likely be able to have more and stronger relationships. In addition, groups within a party will be cautious in building new relationships with groups that are outside the party. This could threaten their own place in the party, or possibly lead to fractures within the party. As mentioned above, in this case the party will not be able to elect as coherent legislative caucus and so will find it difficult to legislate. This means that the informal relationship based structure of parties will make them slow to change and respond to new interests.

1.4 Outline of the Dissertation

In Chapter 2 I outline my theory for how the structure of party coalitions will affect the actions of party caucuses in the legislature. I begin by elaborating on the network paradigm of parties. From this paradigm I argue that the structure of the party network will impact two important characteristics of legislative parties: ideology and cohesion. This is a result of the importance of relationships within the party network and how they will translate into

12 the positions and actions of elected partisans. Testing my theory requires first identifying the party network. I take up this challenge in Chapter 3. I use state legislative donation data to identify relationships between groups by examining shared donations. Identifying relationships using donation data can be fraught as there are many reasons for why a group might donate to same candidates. I use a recently developed method to identify only relationships that are meaningful, opposed to relationships that happened by accident. In the remainder of Chapter 2 I validate the networks identified through this process. I test to see if they are polarized along partisan lines and if groups with known shared interests tend to have relationships with each other. I find compelling evidence for both. With the networks identified I can test the relationships between the structure of the networks and its impact on partisans in the legislature. In Chapter 4 I look at the effects of network structure on ideology. I show that party networks where labor unions have more relationships have legislative caucuses that are more liberal, and that the opposite is true for the amount of relationships that pro-business groups have. These relationships hold across both the Democratic and Republican parties and are robust to a variety of controls, including examining just the amount of donations coming from these groups. In Chapter 5 I move to testing the relationship between party cohesion outside and inside the legislature. I find that Democratic caucuses become more homogeneous when their party networks become more cohesive but that this relationship does not hold for the Republican Party. In the Chapter 6 I conclude by discussing what these findings mean for our understanding of political parties and democracy in general. The fact that relationships are so important has implications for the ability of underrepresented groups to gain influence. They need to not only mobilize to gain attention from those already involved in the party network, but also convince older organizations to build relationships with them. In addition I discuss how the informal nature of party networks poses significant challenges to regulating them.

13 Chapter 2 | What are Political Parties?

“A political party is first of all an organized attempt to get power” (Schattschneider 1942, 35). This quote, from E.E. Schattscneider’s classic defense and explanation of political parties, lays out a simple truth, political parties exist to gain control of government. At the same time it obscures a deeper and more important questions about political parties. Left out is why parties look to gain power, and how parties are organized to do so. Answers to these deeper questions have important ramifications for how policy is created, who gets represented in this process, and what political parties mean for a democracy. In answering these questions I use the paradigm of extended party networks which views political parties not as singular entities, but as complex and at times fractious coalitions of organizations. These organizations have separate and possibly conflicting answers to why they want political power, but they are willing to work together in order to gain access to the levers of power (Bawn et al. 2012). I argue that because political parties are coalitions to understand them we need to understand the relationships that structure these coalitions. Not all groups within party networks are created equal. Some groups build relationships across the coalition and use these relationships to pull the party closer to its own preferred positions. Just as important as when groups are able to work together within a party is when they clash with each other. Groups that have trouble building relationships will, no matter

14 their resources, have more difficulty in swaying the party. Along with these micro-structures the macro-structure of the party network is critical. A divided network will lead to divided policy and so make it difficult for a party to govern even when it holds large majorities. In this chapter I outline my theory for the implications of party networks in the United States. I start with a discussion of the constituent parts of party coalitions, the organizations that make up the party networks. The logic of politics and elections leads these organizations to form coalitions with other organization. In forming long term coalitions they gain a better chance at influencing policy. Throughout the chapter I use examples of groups and their involvement in particular party networks to illuminate my theory. After explaining how party networks form I then build my own theory to explain what effects party networks will have on policy. First, I argue that relationships are important for identifying what organizations are strong within a party network. As a group is more connected to others within the party network they will be able to convince those other groups to support similar policy positions. Second, I discuss why there can be conflict within party networks. Limited choices lead party networks to be expansive entities where not all groups agree all the time. I argue that the lack of perfect internal cohesion will lead to similar disagreement among the elected officials nominated by the party. Conflict can exist at the state and national level, but I focus especially on conflict within state parties. Variation across states lead to similar variation in the degree of conflict within parties. This allows me to directly test my theory in later chapters.

2.1 Parties as Networks

In order to understand party networks I start at the beginning by discussing the role of interests within politics. This follows a similar logic to that used by Bawn et al.(2012) in describing the source of party networks. I first discuss the constituent parts of party

15 networks, intense policy demands, and then what forces them to come together into a party coalition/network and the process of them working together. I focus particularly on the costs and benefits of policy demanders coming to agreement within the party coalition. This has been acknowledged in previous work, but the theoretical implications, as outlined in the sections that follow, have not been explored.

2.1.1 Demands for Policy

The theory of parties as networks take as a given that individuals have preferences for policy (Bawn et al. 2012). This is a fairly non-controversial starting point. In particular I do not assume that all individuals have preferences for policy, or that those individuals that do care about policy care about all policy. Instead, I assume that there exist some set of individuals and groups that care about a set of policies (Schattschneider 1942; Olson 1965; Downs 1957). These preferences might be rooted in economic concerns. Business owners prefer policies that are beneficial to their business, and so their preferences might be very specific (e.g. changes to tariffs on their products) or very general (e.g. reduced taxes on profits). Other interests may reflect pertinent identities, cultural values, or ideological understandings of the world (Wildavsky 1987; Laitin and Wildavsky 1988). A subset of individuals or groups with demands will be willing to take action in order to implement their policy demands. These engaged groups, organizations, and individuals make up a set of actors called intense policy demanders. Intense policy demanders are actors that regularly participate in political activity in order to pursue their demands (Cohen et al. 2008). They can be, as mentioned above, groups focused on relatively narrow set of interests. Intense policy demanders can also have wider foci. For example, Public Citizen, a progressive political organization has an expansive view of its policy concerns. Their 2016 annual report highlighted success on trade (stopping the Trans-Pacific Partnership), energy (new policies

16 on energy rates in a handful of cities and states), and consumer protection (lobbying to stop hidden ads on social media), among many other issues (Public Citizen 2017). Most of these intense policy demanders are groups, and not individuals, because of this I use both groups and intense policy demanders throughout the dissertation to refer to the same thing. Intense policy demanders are not limited to just those that participate in traditional lobbying and electoral activities. Intense policy demanders might also use more confrontational actions, designed to approach the polity through outside avenues. These types of intense policy demanders are often subsumed in the category of social movement organizations (McCarthy and Zald 2003). These two categories of groups are not mutually exclusive. For example, ADAPT, a disabilities rights organization, is well known for their theatric confrontations with decision-makers. They have taken over legislative offices, declared areas “nursing homes” to demonstrate the negative treatment of those in real nursing homes, and had wheelchair bound members crawl up the steps of capitol buildings. ADAPT has also been part of lawsuits, and consistently ask their members to call into legislative offices (Fleischer, Zames and Zames 2012). ADAPT, like other intense policy demanders, participate in politics through multiple avenues. It is important to acknowledge that not everyone with demands will be spurred into action. Individuals with little to gain from policy changes will be less likely to act based on their preferences. Large numbers of individuals with preferences will not always translate directly into an equivalent amount of action. Collective action, especially among large groups, is difficult, and most individuals will be incentivized to free-ride off the work of those who are willing to take action (Olson 1965). In addition, individual resources will be an important determinant in who takes action. Socioeconomic status is closely connected to political participation, with those who are highly educated and wealthy as also being the most active (Schlozman, Verba and Brady 2012). A similar bias will then appear among intense policy demanders, with active groups best representing privileged individuals and interests. Even

17 policy demanders representing underprivileged interests they will tend to represent the least underprivileged individuals (Strolovitch 2008). Individually these intense policy demanders make up the building of block of political parties as networks. To see how policy demanders go from working individually to working together as coalitions it is necessary to look at the demands of the political process. It is the logic of politics that push these atomized actors into party coalitions.

2.1.2 Coming to Agreement

The electoral rules in the United States, single member districts with first past the post voting, mean that it is unlikely for more than two candidates to be viable in any general election (Duverger 1959). As intense policy demanders look toward implementing policy they realize this and that they will need a majority of voters to support their interests in order to elect candidates. As an individual group this will be difficult as the issues that intense policy demanders care about are unlikely to draw interest from a majority of voters. This pushes intense policy demanders toward building broader based coalitions with other intense policy demanders (Bawn et al. 2012). Although they might not necessarily agree with each other on all positions, they can build coalitions if they all have relatively focused interests that do not clash. This is where party networks begin to appear, through the demands of the electoral process. The way that intense policy networks have to come together has important ramifications for how the party network operates and its implications for the policy process. Intense policy demanders within a party network will work to come to some common agreement about how to successfully implement policy. The most important task is for them to select candidates that are appealing enough to the electorate to be successful and can also be trusted to implement the party’s preferred policy once in office. This task is difficult given the culture and legal system established in the United States. Democratizing reforms put

18 in place over the 20th century, first by progressive reformers in the early 20th century and then by leftist during the 1960s and 1970s have led to an open candidate selection system (Cohen et al. 2008; Masket 2016). Candidates are selected through a process of elections. The specifics vary across states, but in general either all those registered to vote can participate or those who have registered to vote and indicated some party preference can participate. Intense policy demanders have to operate within this system in order to come to select candidates that others in the party network can agree upon. Intense policy demanders are not without strategies in getting the right candidates selected. Elections are costly and most voters are relatively uninformed so intense policy demanders can provide resources to candidates and cues to supportive voters (Grossman and Dominguez 2009). This support can have a strong impact in ensuring candidates win both the primary and the general election (Desmarais, La Raja and Kowal 2015; Yang et al. 2015). The actions of those in the extended party network can be strong enough to create ideologically cohesive parties within the legislature even in instances of non-partisan elections. In Nebraska’s non-partisan legislature implementation of term-limits allowed policy demanders within the state to finally overcome the challenge of non-partisan elections to create coalitions within the legislature that are non-partisan in name only (Masket and Shor 2015; Masket 2016). Intense policy demanders within the extended party network can work together beyond just endorsing candidates or providing donations. Another key resource for organizations is the provision of expertise in campaigning. Analysis of staffing networks have shown that these staff follow partisan flows, and that information is passed from campaign to campaign though shared staffing (Nyhan and Montgomery 2015; Skinner, Masket and Dulio 2013). This, by itself, does not directly connect to intense policy demanders, but a lot of the staffing support comes from groups affiliated with the intense policy demanders. For example, Wellstone Action is a progressive organization founded to honor the late Senator Paul Wellstone. They provide not only training for progressive campaigners but also disseminate job listings for

19 campaign and organizational staff needed on progressive campaigns. Wellstone Action and the associated Wellstone Action Fund is supported by a long list of participants in the Democratic network: AFL-CIO, America Votes, Democracy for America, Emily’s List, Jobs with Justice, and many more. They are also supported by formal party organizations like the Georgia House Democratic Campaign Caucus, the Idaho Democratic Party, the Minnesota Democratic-Farmer-Labor Party and Ohio Young Democrats.1 In the language of party networks Wellstone Action is an intense policy demander that has the support from other intense policy demanders and they all use their resources to elect progressive candidates under the banner of the Democratic Party. Along with the difficulty of operating within a legal framework, intense policy demanders face the challenge of participating in what is referred to as a collaboration game (Morrow 1994). Start with only two intense policy demanders—the Farmer’s Association and the Free Speech Association. They can support either Candidate A or Candidate B. Candidate A was a farmer and so will strongly represent the interests of farmers, but is willing to also support free speech advocates. Candidate B was a free speech lawyer, and is willing to support farmer’s interests as well. The Farmer’s Association has much more trust in Candidate A than Candidate B, but the opposite is true for the Free Speech association. If the two groups split though neither candidate is likely to get elected. Table 2.1 summarizes the payoff for each intense policy demander. This type of collaboration game is commonly known as the battle of the sexes. Importantly, both players (in this case the two intense policy demanders) can increase their playoff through communication prior to the game (Morrow 1994). The two groups can indicate who they plan on supporting and respond to each others plan in order to coordinate their actions. Without this communication they are best off playing a mixed strategy—sometimes supporting Candidate A and sometimes supporting Candidate B. If groups are able to communicate

1Information available on http://www.wellstone.org/about/partners and collected in March of 2017.

20 Table 2.1: Collaboration Game for Intense Policy Demanders

Free Speech

Support A Support B

Support A (2,1) (0,0)

Farmer’s Support B (0,0) (1,2) Note: This shows an example of the collaboration game for intense policy demanders. Both groups can select either to pick Candidate A or B. They receive the payoff based on either, and the other groups selection. If they both pick A then the Farmer’s group receives a payoff of 2 and the Free Speech of 1, if they both pick 1 the opposite payoffs are received and if they pick opposite candidates then both receive a payoff of 0. Both groups would prefer to work together but disagree on who to coordinate over.

before the game they can improve upon this by coming to agreement ahead of time over the best strategy (Cooper et al. 1989). These types of pre-game communications happen often among intense policy demanders. Intense policy demanders use the long election cycles to communicate with each other so that they can coordinate around a single candidate (Cox 1997). This simple game becomes more complicated as more players are added. Given the thousands of national, state, and local organizations that actively participate in party networks it might be expected that it would be impossible for any coherent structure to appear. Even with months of long discussions prior to elections finding agreement between hundreds of groups would be likely impossible. This is not the case though, coherent coalitions do form and they form because intense policy demanders participate in multiple games over time. Players are better able to coordinate when these types of games are repeated multiple times (Crawford and Haller 1990; Bhaskar 2000). Repeated participation in these games gives rise to social norms about what participants ought to do as well (Knez and Camerer 2000; Voss 2001). In the case of party networks the most important norm that is created is around what it means to be a Democrat or a Republican. Repeated interactions give rise to

21 the perception of what types of candidates will gain general support from the coalition and so make it easier to continue to coordinate. This is the formation of a party ideology which I elaborate on in the next section.

2.2 Party Networks and Ideology

In the development of parties, the policy demanders benefit from creating a “useful fiction” of ideological labels for the parties (Bawn et al. 2012, 254). Ideological labels indicate that those within the party are in agreement and are in opposition of those who stand outside the party. Although Bawn et al.(2012) and others have discussed that party ideology is based in the party networks, no one has elaborated on how party ideology reflects party networks. Bawn et al.(2012) explain that party ideology is a reflection of those involved in the party network, but they do not detail how the potentially conflicting interests within a party organization are translated into a particular ideology. They leave unanswered if ideology equally reflects all preferences within the coalition, or if there is differences in the influence groups have. I contend that groups that play a more critical role within the party network will have more control over the content of this ideological brand. The necessary next step is defining what it means to be a critical participant in the network. I argue that groups that are more central to the network, meaning that they have more relationships within the party coalition, will be more critical to that network and will therefore have a larger impact on party ideology. In this section I first discuss how ideology solves a problem of coordination, this helps explain how networks relate to party ideology. I then discuss how the placement of groups within the network will give them strength over the ideology of the party, that as groups have more relationships within a party they will also have more strength. Variation in the strength of groups across states explains the observed variation in state party ideology.

22 2.2.1 What Ideology Achieves

As discussed above, intense policy demanders have strong preferences about a limited subset of issues; their preferences across issues will vary in both what their preferred policy outcome is as well as how salient that policy outcome is to them. As the party coalition looks for a candidate each particular member wants to maximize how much policy they will get from the government. They prefer to work with other groups because they know this improves the chance of them electing a candidate but they do not necessarily care about the policy preferences of other groups. It is easier for groups to coordinate if there is a common understanding of the positions of the coalition as a whole. This makes it is easier for a candidate that wants to appeal to the party network to select the right bundle of positions. If such a candidate exists, where all members of the coalition find them to be an optimal choice, then there is no longer a coordination problem. In Table 2.2 I extend the battle of the sexes of the game above to include such a candidate. In Table 2.2 Candidate C will give equal payoffs to both the Farmer’s group and the Free Speech group and is no less preferred by either group than other candidates. The optimal strategy for each group is to support Candidate C, knowing that the other group will also select Candidate C. Although each group might individually receive the same pay off by selecting other candidates this would risk potentially getting nothing as well. But how do intense policy demanders ensure that there exists a Candidate C? Intense policy demanders can rely on partisan ideology to make this possible. As discussed above, ideology flows naturally from the repeated interactions of policy demanders. When there are clear partisan divides, party activists become socialized into the specific values and opinions of the party (Layman et al. 2010; Layman and Carsey 2002). This socialization process means that increasingly more activists will take on the necessary views to well represent the

23 Table 2.2: Collaboration Game with Ideology

Free Speech

Support A Support B Support C

Support A (2,1) (0,0) (0,0)

Support B (0,0) (1,2) (0,0)

Farmer’s Support C (0,0) (0,0) (2,2) Note: This is an example of the same collaboration game as found in Table 2.1 except with the addition of Candidate C. Candidate C is possible because a clear ideology makes it easy for Candidate C to select the right bundle of preferences. Unlike Candidates A or B, both groups can trust that Candidate C will support their positions and so it is easy for them to coordinate around her.

coalition. As more activists take on these views the pool of potential candidates increases and so the potential coordination game becomes easier.2

2.2.2 What is the Party Ideology Going to be?

Ideology reflects the coalitions coordinating around candidates. The ideology itself is depen- dent on who is participating in this coordination game. At a minimum this implies that party ideology will be a function of who is involved in the coalition. As intense policy demanders leave or enter the party coalition they will change what the ideology looks like. Although this is necessarily true it is not enough to understand party ideology simply by looking at who is involved. It is necessary as well to look at how powerful each intense policy demander is within that party coalition. I argue that the ideology that is presented by the party network will reflect the relationships of different intense policy demanders within the party coalition. Those that have more relationships will have more influence on the party ideology while

2Activists are a particularly good source of candidates as they can be most trusted to fulfill the needs of the coalition. If they are true-believers in the party positions, then the party coalition does not need to worry about them abandoning important issues (Masket 2016).

24 those who are more isolated within the party network will have limited influence on the party ideology. Before discussing why relationships matters it is useful to examine, briefly, some examples of when we know that intense policy demanders within a party coalition have had varying influence on the nomination of candidates. In the lead up to 2008, Senator John McCain made high profile overtures to the Christian Right including speaking at the 2006 commencement ceremony for Liberty University. This was an attempt to make amends for an incident during the 2000 election where Senator McCain called Jerry Falwell, Liberty University Founder and leader within the Christian Right, part of a set of “agents of intolerance” in the Republican Party (Balz 2006). It is not possible to be certain that John McCain’s approach to the Christian Right is the reason that McCain won the nomination in 2008 and lost it in 2000. It suggests, though, that Senator McCain believed that Jerry Falwell, a leader in the Christian Right, exerted a large amount of influence over the Republican nomination process. Now, imagine that the Christian Right were only a minor part of the Republican Party network in 2000. This shift in the party network could have opened a path for John McCain to have won the nomination in 2000 instead of George Bush. Bush, a compassionated conservative, was more conservative than John McCain on many issues (Jacoby 2004). A McCain win in 2000 might have indicated a different, more moderate or libertarian, brand for the Republican Party in the early 2000s. Although McCain did go on to win in 2008, he did so only after purposefully courting individuals on the right. After winning the nomination he also selected a much more socially conservative running mate, Sarah Palin, in part to strengthen his relationship with the Christian Right of the Republican Party (Caswell 2009). In contrast, African American groups and voters have often been described as a “captured interest” by the Democratic Party (Frymer 2010). Although they participate within the Democratic Party both at the voter level, the activist level, and the elite level, they do not necessarily have as much voice in the positions that the Democratic Party has taken. This is

25 a result of them having few alternative choices. The Republican Party coalition has rarely attempted to court African American voters. The options for African American activsts are to participate with the Democratic Party or to exit from politics. As in the case of John McCain, it might be possible to imagine a world where black voters had a stronger voice in dictating the types of candidates selected by the Democratic Party. Given that African American voters are significantly more liberal on a range of economic issues (Frymer 2010) they would pull the Democratic Party further to the left on not just racial issues but economic issues as well. The ability to influence a party coalition is contingent on not just a intense policy demander being present within the party coalition but the relationships of a group within that coalition. The Christian Right and African American voters are both part of party coalitions, but they are not equally influential within the coalition. It is also clear from this example that influence is not only a function of the ability to provide resources. Both the Christian Right and African American interest groups are able to mobilize large number of voters and potentially resources, yet they have differing influence within their respective parties. Instead, influence is rooted in the relationships that these intense policy demanders have with other groups in the party coalition, and how these groups see each other. African American interest groups were at their strongest in the Democratic Party during the civil rights movement, in part because of movement mobilization, but also because they built relationships with labor unions. Both unions and African American interest groups saw the Democratic Party as being potentially drawn further to the left and so worked together to make that possible (Baylor 2013). The Christian Right has also steadily built influence within the Republican Party not only by mobilizing voters but also by providing information to elites within the Republican Party (Clifton 2004). The Christian Right has also successfully worked with other groups in the Republican Party without stepping on toes. Christian Right voters

26 are willing to accept candidates that might not be perfect in order to build relationships within the Republican coalition as a whole (Oldfield 1996). To see why this is return back to the coordination game in Table 2.2. Currently there are only two players, and neither are harmed by going for Candidate C. In reality though there are many players and ideology is not rigidly defined so as to ensure there will be a single candidate that either all prefer or are indifferent to. Intense policy demanders will have to navigate this coordination process with others, and when an intense policy demander has strong relationships with others in the party network they will be in a better positions to convince others in the party network to coordinate on candidates that are more preferred to them. These relationships are important as they allow groups to provide information on candidates to others in the party network as well as to make decisions together about candidates. Groups with few relationships will have more difficulty in convincing other intense policy demanders on what candidate to coordinate on and so will be less influential in selecting candidates and dictating party ideology. I argue that party ideology is a function of the relationships that exist within a party coalition. Intense policy demanders with more relationships will move the party ideology closer to their own preferences. Relationships themselves might be a function of the resources of a group but not necessarily. Relationships between groups can also reflect other factors, especially how elites within the party view the issues that are central to the groups they are looking to build relationships with. They must perceive relationships as useful to their own immediate ends and the party as a whole.

2.3 Parties In and Outside the Legislature

Party networks aim to create a party caucus within the legislature that is in agreement over the preferences of that party. As discussed in the previous section, the party caucus ideology

27 reflect the relationships within the party network. But, what if there is not agreement within the party network? Here I argue that a lack of agreement within the party network has important ramifications on the party caucus in the legislature. As the party network breaks down outside the legislature the related caucus will also breakdown and will have trouble governing. I start this section by discussing my argument for the theoretical connection between the party outside the legislature and the party inside the legislature. Next I discuss, given our understanding of parties within the legislature, that we would expect that a more cohesive party outside the legislature would have an impact not only on the cohesion within the legislature but on the strength of parties within the legislature. In doing so I draw on the idea of the conditional party government (CPG) (Rohde 1991).

2.3.1 From Outside In

The theory of extended party network supposes that intense policy demanders outside the legislature select candidates in order to get policy passed. The immediate purpose of the party network is to select a unified set of candidates and elect them into office (Bawn et al. 2012). When those in the party network perfectly agree this ought to be an easily accomplished task. Groups nominate a slate of candidates that reflect their policy interests. Although not all of these candidates will be elected by the voters, those that are will work together to pass the legislation supported by the party network. This simple story does not capture important parts of the process. Parties often face a large amount of internal dissension as those within the party network compete with each other. The fact that networks are internally fractured has not been theoretically examined by previous work and has important implications for how party networks affect parties within the state legislature. As groups within the party network disagree they will have difficulty

28 coordinating on candidates. This breaks the assumption that the party network will select a coherent set of legislators, instead legislators in different districts may represent different aspects of the party coalition and so will bring the disagreement of the party network with them as they enter the legislature. We have observed significant variation in ability of intense policy demanders to come to agreement within a party network. The most readily apparent example of this is the Southern Democratic Party that formed after the Civil War and remained influential until late in the 20th century. The Southern Democratic Party was cohesive around only the issue of race and was internally divided into many factions (Key 1949). At the national stage the Southern Democrats clashed with the rest of the party. Smaller differences in agreement within the party network have been noted as the result of differences in the presence and size of varying socio-economic groups across different states (Gimpel 1996; Gimpel and Schuknecht 2009). For example, compare Illinois and Minnesota at the turn of the 21st century. These states are broadly similar in many respects, but different socioeconomic coalitions and patterns result in different electoral forces and coalitions. Illinois politics is dominated by Chicago with its relatively large African American population. Minnesota politics is less dominated by a single metropolitan area, is more racially and ethnically homogeneous and politics in the northeast part of the state reflects its history of union politics founded in the coal mines there (Gimpel and Schuknecht 2009). The differences in socioeconomic groups and division in the state can make it more or less difficult for party networks to cohere. The centralization of politics around Chicago could act as a centripetal force on the Illinois Democratic Party as groups and actors within the party network are in close contact with each other and likely have similar political concerns. In contrast, the Minnesota Democratic Party network spans more of the state and must unite both urban liberals with rural coal miners in the North. The differences in racial makeup may also have an affect on party network cohesion. In Minnesota policy concerns for both

29 Democratic and the Republican Party revolve around concerns of mainly white voters, this prevents potentially racialized disagreement from causing division within the party networks. In Illinois this operates differently. The Democratic coalition in Illinois is very racially diverse leading to potential division within the party network as groups split on racialized issues. The Republican coalition, which is predominantly white, will not suffer the same sort of division though. The potential for geographic variation in agreement is especially important given that candidates are elected from geographic constituencies. Legislatures represent specific areas, although these areas are embedded in the state or national political discourse, they are only directly accountable to the voters in their constituencies. Disagreement among the party coalition combined with variation across the state and elected officials responsive to local concerns leads to variation in the legislature. As intense policy demanders look to select candidates in Illinois and Minnesota they will have to navigate potential disagreements and elect a set of candidates that are relatively uniform across a wide set of districts. If the party network is poorly unified, then the slate of candidates they select will have similar disagreements, and will not have a single view of what sort of policies they need to be implementing. This brings the fractures that exist in the party network into the legislature. Along with differences in groups across states there has been longterm changes that will affect the ability of coalitions to be cohesive. Most critically there has been increasing sorting of voters into political parties, where preferences of Republican (Democratic) voters have become more uniformly Republican (Democratic) (Levendusky 2009). Evidence of this has been found at the state level as well, with some states experiencing partisan sorting at different degrees (Caughey, Dunham and Warshaw 2016). The growing amount of party sorting might reflect strengthening party networks (better developing party brands) but also might cause strengthening party networks as those within the party network are in better agreement about what concerns the party.

30 Party disagreements within the legislature do not necessarily have to be the result of disagreements within the party coalition. Party coalitions have imperfect control over the select of candidates. They do not control nominations but only exert influence over the nomination process. Candidates can gain the nomination of their party even without the favor of the party elites. Donald Trump is perhaps the best example of this in modern American politics. He received little support from party elites throughout the Republican Primary and even after his nomination party support was far from universal. Donald Trump’s nomination is a deviation from the normal selection process. In most cases though candidates that have the support of party networks and party elites win the nomination (Bawn et al. 2012). Hassell(2016) shows that in races for the US Senates candidates deeply embedded in the party coalition are usually selected as nominees. At the state level party networks are able to use the nomination process to enforce partisanship even when nominations are not directly partisan (Masket and Shor 2015). Party coalitions are very adept at overcoming the limitations of electoral rules to enforce some degree of partisanship on legislatures (Masket 2016, 2007). Given that party coalitions do a relatively good job controlling the nomination process, it is unlikely that most contention within legislative caucus is the result of unacceptable candidates winning nomination and being elected into office. Here we have seen that party coalitions appear to vary in their ability to be cohesive, and that this cohesion ought to travel from outside the legislature to the caucus inside the legislature. In the next section I explain why variation in party cohesion in the legislature is important by focusing on one traditional understanding of legislative party politics: conditional party government.

31 2.3.2 Party Role Inside the Legislature

Conditional party government (CPG) is one of the dominant theories that explains the relationship between individual legislators and the legislative party. In brief, the theory of CPG outlines two conditions necessary for political parties to exert influence in the legislature. First, the two political parties must be distinct from each other. Legislators from two different parties should not have similar positions on the issues. Second, within each party legislators must be cohesive and agree on the issues. In short, for parties to be strong in a legislature they must be distinct and internally in agreement (Rohde 2013, 1991). When these conditions hold legislators are willing to give up individual power so that the party as a whole can accomplish more. Whether these conditions are met is determined by who is elected into office. Conditional party government was therefore based on partisanship within the electorate (Aldrich and Rohde 2001; Ladewig 2005). An electorate that met the conditions of conditional party government would elect candidates that met these conditions as well (Aldrich and Battista 2002). This assumes that the preferences of the electorate are directly translated into the candidates that they elect. Party coalitions though, as described above, are an important linkage in the electoral process. Through the use of primary elections they narrow the options for the electorate during the general election. As long as the party coalition is able to exert influence on the primary process candidates will not perfectly reflect the preferences of their constituents. The party coalition is an important factor in determining the strength of parties in the legislature. As both party coalitions within a state become internally more cohesive and externally more divided, they will nominate candidates across the state that reflect the views of their party well and have very few preferences that are in line with out-partisans. Those candidates that are elected from the set nominated will fulfill the necessary conditions of

32 CPG and will elect to strengthen parties within the legislature. The electorate at-large still plays a role, as the party coalition is constrained by needing to nominate candidates that can be elected in the general election. Voters in general elections though often rely on partisan cues, especially when they have relatively low information (Jessee 2010). In primary elections voters tend to have trouble selecting among similar candidates which one better represent their views, especially on down-ballot elections where there is substantively less media attention (Hirano et al. 2015; Lau 2013). The party coalition plays an important role as they are able to ensure that the right candidate is nominated and work to ensure that that candidate wins the general election.

2.4 Conclusion

Political parties are complicated contentious webs of interests, but through repeated interac- tions overtime they build a structure. This has been noted by other researchers in recent years, who have argued that these structures impose control on politicians. Here I have gone beyond that by specifying what about the party network matters for understanding the actions of politicians. In doing so I have elaborated on how the structure of complicated networks, and the placement of intense policy demanders within that network, can have specific effects. In particular I argue that because parties are networks the important variable of interest are the relationships between intense policy demanders. Groups with more relationships within a party network will be more influential in the preferences of that party. They can use those relationships to pass important information and to persuade co-partisan to support their ideal candidates. Through this process they shift the party ideology. In addition, the legislative caucus that is put into office by the party will reflect the divisions within the party network. As a party network becomes more divided internally the caucus will also be divided

33 and so will have problems governing, no matter the majority they have. In the following chapters I take up this theory and test it by examining state legislature ideology and cohesion. This requires first identifying the party coalition. This task is made difficult by the informal nature of the party coalition. In the next chapter I develop a way to identify the party network. I devote an entire chapter to this because of how difficult this task is, and the importance of correctly identifying the party coalition in order to test theories about it. In the following chapters I directly test how the party coalition effects the legislative actions of the party.

34 Chapter 3 | Identifying the Party

Identifying party networks leads to both conceptual and practical problems. Conceptually, as discussed in the previous chapter, there is no formal designation for an intense policy demanders position within a party network. Groups do not always declare that they are part of one or the other party and many groups claim to be non-partisan while dominantly supporting candidates within a single party. Even when groups do declare themselves as part of a party it is not clear where they are within that party coalition. This is important as we know that there is often conflict and jockeying for influence within political party coalitions (Baylor 2013). Relationships between intense police demanders are fluid and groups often vie to increase their position within the party network—looking to have their policies become more central to the party in order to gain policy wins. Since designation within a party network is not readily available it must be inferred from other actions, which leads to a practical problem of how to infer relationships from actions. In this chapter I first examine how the party network has been previously identified, in particular the use of donation data. I extend and refine this approach to look at state party networks. In order to do this I use a social network methodology called backboning which infers significant relationships among donor groups. Backboning has often required rigid assumptions across the network to identify when a relationship is significant. Instead of

35 doing this I employ a newly proposed method by Neal(2014). This method uses a variable threshold for each pairs of donors based on the observed in the data. A variable threshold is important as it is not necessary to assumed homogeneity across group resources. I apply this approach to state data from 2000 to 2016. I validate the networks by examining how groups cluster by party, the place each group has within a network, and finally how groups that are expected to work together are found to be connected.

3.1 Identifying the Party Network

The party network is made up of a variety of intense police demanders including elected politicians, candidates, groups, activists, and individual donors. Together these groups and individuals act in order to gain policy through first winning elected office and then legislating. They may participate in formal organizations together, but their relationships are not bound by formal connections. Using the definition of institutionalization from Polsby(1968) and Squire(1992), parties are not well institutionalized. There are no formal gatekeeper structures on the party network, instead the individual actions of each intense policy demander dictates who is allowed into the party network. Even the formal party organization can change rapidly in response to demands from groups. This does not mean that examining and testing theories of parties is intractable, only that one must look beyond formal party organizations to fully understand their actions. In order to examine the extended party network, it is necessary to use the actions of actors that could potentially be within the network. Previous research has looked at a variety of different activities. Donations (Hassell 2016; Desmarais, La Raja and Kowal 2015; Yang et al. 2015; Grossman and Dominguez 2009) and endorsements (Bawn et al. 2012; Grossman and Dominguez 2009) have been the most common, but others have used consultants (Nyhan and Montgomery 2015), employment (Skinner, Masket and Dulio 2013) and the sharing of

36 contact lists (Koger, Masket and Noel 2009). The different activities are used to indicate overlapping interests among actors. For example, groups willing to share (or sell) contact lists are expected to be in broad agreement on political issues and this one relationship is assumed to indicate deeper connections between them (Koger, Masket and Noel 2009). Because of their usefulness, shared donations are the preeminent ways to identify party networks (Hassell 2016; Desmarais, La Raja and Kowal 2015; Yang et al. 2015; Grossman and Dominguez 2009). This approach has several advantages. First, it is clear that raising money and garnering donations are important for candidates winning elections. In state races, candidates often raise hundreds of thousands of dollars. In 2013-2014 cycle, the average amount raised by the top-funded legislative candidate in a race was ≈$128,000. Even in elections with little competition, candidates still raise sizable sums of money. In Georgia, where only 2% of seats up for election in 2014 had a race with multiple candidates in the general election, the average raised by the most well funded candidate in each legislative election was ≈$77,000 (Holden 2016). Campaigns concern themselves with raising cash even if they do not have a challenger because raising funds might be seen as a way of preventing future challengers. Research on the effects of candidate funds on preventing challengers is mixed (Carson 2005; Goodliffe 2007; Lazarus 2008), but there are strong reasons to believe that elected officials perceive fundraising as an important way to prevent challengers (Goodliffe 2005) or just as preparation for when a challenger does emerge (Goodliffe 2001). Second, there is variation in the amount donated by each group to each candidate. Intense policy demanders do not decide just whether to donate to a candidate or not, they also decide how much to donate to a candidate. This provides ample information about the relationship between the candidate and the group, which leads to ample information about the relationships between intense policy demanders. For example, if two intense policy demanders donate the maximum amount to a candidate, that indicates a stronger agreement between

37 them then if, all else equal, they were just to donate a small amount to that candidate. The information from donations is thus more fine-grained than other options like endorsements, where a group either endorses or does not endorse. Third, donation data is widely available and transparent to other donors. This is important both for practical and theoretical reasons. Theoretically, the fact that donation data is made available helps other intense policy demanders track each other. Although most communications between closely tied actors will happen privately, publicly available donation data provides an avenue for communication between groups that are not closely tied. Practically, the fact that the data are available makes it feasible to test theories about the variation of placement within networks across states. The other metrics used to identify party networks, such as shared consultants (Nyhan and Montgomery 2015), staff employment (Skinner, Masket and Dulio 2013) require ample data collection to provide a picture of the national network in a single time period. Collecting similar data across multiple states and multiple elections would be difficult if not impossible. Granted, donation patterns are not a perfect reflection of policy demander strength. Concerns can be raised that groups donate for a variety of strategic reasons that do not necessarily reflect a belief in shared ideological interests with the candidate. For example, access-oriented groups are less likely to donate based on shared interests or ideology, and more likely to donate towards candidates that they believe will win. These types of donors tend to favor incumbents over challengers because of the incumbents improved chance of success (Snyder Jr 1992; Fouirnaies and Hall 2014). A purely access-oriented group would not be expected to be in any particular party network, but it might be hard to identify this if donation data is taken at face value. The effects of this can be minimized by processing the donation data to identify donations that are unique given the general patterns of donations. This is possible a using backboning process developed by Neal(2014) which identifies only unexpected connections between donors based on the empirical pattern of donations in each

38 state. In addition, organizations have different degrees of resources. This can limit the amount that is donated because a group does not have the funds to donate more. A group could artificially appear to have fewer relationships because of their resources. The backboning process again reduces the impact of this as it controls for a groups propensity to donate (which ought to be in part a function of their resources). Finally, it is necessary for their to be relevant candidates for groups to coordinate around. If two groups have a relationship, but no candidates exist that the groups can coordinate upon then there will be no relationship found between the two groups. This limitation reflects a sparseness in the political world that cannot be easily relieved.

3.2 State Level Donation Data

I use data from the National Institute of Money in State Politics (NIMSP) to generate party donation networks. The NIMSP has aggregated donations in state elections, both executive and legislative, from 2000 to 2016. These data include information available from contribution reports that candidates file along with information appended by NIMSP. Although the data includes executive campaigns, I have excluded them from what I present here. Executive elections represent only a small subset of state elections and vary drastically across states. Variation in networks surrounding executive offices might represent only variation in the individual candidates or the type of elected position in the state. Although state legislative offices also vary in their size, scope, and professionalism, this variation is minimal in comparison to elected executive offices. For example, during this time period in Maine the Governor is the only elected position within the executive branch, while in Kentucky there are elections for Governor and Lieutenant Governor (on the same ticket), Attorney General, Auditor of Public Accounts, Commissioner of Agriculture, Secretary of

39 State and Treasurer. Identifying the party network requires decisions about the temporal nature of the network. Although elections happen regularly (2 to 4 years) the entire legislature is not elected at each cycle and there is variation in the percentage of the legislature elected at a given point. In addition, between states there is variation among the state upper house in the percentage of the legislature that is elected in a given cycle. Even within a state there can be variation, as some states require that the entire upper house is reelected at the first election after redistricting. In order to account for this variation, the data is organized by State-House-Full- Cycle. For elections where an entire chamber is elected, this approach includes all donations made that electoral cycle (since the previous election). In states where only part of a chamber is up for election, the data includes multiple cycles until every legislator in a chamber has been up for election once. Because of this, some donations are included in multiple sets of data. For example if half of a chamber is up each two years, the data will be be organized in groups such as 2001 to 2004, 2003 to 2006, 2005 to 2008, etc. Starting with all donors to state legislative candidates, I exclude donations from individuals. This limits donations to those from groups, party committees, and other candidate committees. The focus of extended party network is chiefly group based, although it can also include individuals and activists. By only identifying groups, I am able to directly test my theory through the use usie group codes created by NIMP to identify preferences of groups. NIMSP has three levels of categorization which they call groups (16 categories), industries (over 100 categories) and businesses (over 300 categories). Next, I exclude groups that only participate in a single election cycle. Intense police demanders in the party network are part of long coalitions and so must regularly participate in elections. The complete dataset consists of 699 State-House-Full-Cycle. Before identifying the relationships it is useful to examine the data. The Pennsylvania 2005-2006 lower house race has the highest number of candidates, with 396 candidates, and

40 the Florida upper house cycle from 2005 to 2008 had the most donor groups with 11,981 in total. Connecticut has some of the smallest races, a result of their public financing laws (Cha and Rapoport 2013) with only 8 donors participating in the 2015-2016 state lower house that meet the requirements above, and a similarly small number of candidates. Because donations matter so little in states with robust public financing I am forced to exclude Connecticut, Arizona and Maine for my analysis.1 The median number of candidates in each State-House-Full-Cycle is 100, and the median number of groups is 623. In total there are 801,595 group-cycles in the data, and 81,781 candidate-cycles.2 Figure 3.1 plots the number of groups and number of candidates for each State-House-Full-Cycle, there is only a weak correlation between them with a Spearman’s ρ of 0.21, meaning that as the number of candidates increase the number of groups increases as well. In general for each additional candidate there are 3.16 additional groups. Not all groups donate to all candidates. Figure 3.2 plots the percentage of group-candidate pairs where a donation was made from the group to the candidate. This ranges from 0.98% to 15.24% with an average of only 5.48%. In all states groups donate to only a minority of potential candidates, although there is a degree of variation in how small of a minority. In 3.3 I show the density and histogram of the median donation in each State-House-Full-Cycle. The minimum median donation is $100, and the maximum is $2,000 with an average of $530.51 across the datasets. The donation data provides a large amount of information about the interactions between donors and candidates. Information that, if used properly, can identify the relationships of interest between donors. There is a large amount of variance across states, in the amount of

1These states all have clean election laws. Candidates that opt in to the program must receive a certain number of very small donations from individuals. After reaching that threshold they then receive money from the state and have very strict rules on their ability to raise money directly from contributors. This makes it significantly more difficult to identify relationships between groups as most candidates opt in to the government campaign support and so opt out of raising funds from groups. 2I use group-cycles and candidate-cycles here as groups and candidates can participate over multiple election cycles.

41 Figure 3.1: Number of Candidates vs Groups in Each State-House-Full-Cycle

● ●● ●●●● ●● ● ● ● ● ●●●● ● ●●● ● ● ● ●● ●● ●●● ●● ● ● ●●●● ● ●●● ●● ● ● ● ●●●● ●●● ●● ● ●● ● ● ●●● ● ● ●● ●●●●● ● ●●● ● ● ● ●●● ●●● ●●●●● ●●● ● ● ● ●●● ● ●● ●● ●●●● ●●●●●●●●●● ● ● ● ●● 200 ●● ● ●●● ●● ● ● ● ●● ● ● ● ● ●●● ●●●●●●●●●● ● ● ●●● ●●● ●●● ●● ● ●●●●●●● ● ● ●●●● ●●●●●●●● ● ●●● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ●● ●●●●● ●●●● ● ● ●●●●●● ●●●● ●●●● ● ● ● ●● ● ● ● ● ● ●● ● ●●● ● ●●● ● ● ●●●● ●●● ● ●●●●● ● ●● ●●●●●●●● ● ●● ●● ●●● ● ● ● ● ● ● ●● ●●●●●●● ●●● ●● ●● 100 ● ● ● ● ●● ● ●● ●●● ●●●●●●●●●● ●● ●● ●●●●● ● ●● ●● ●● ●●●●●●●●●●●● ●●● ●●● ● ● ●●● ● ● ● ●● ● ● ●●●●●●●●● ●●● ●●●●● ●● ●● ●●●● ●●●●●● ●●●●●● ● ● ● ●● ●●●●● ●●●●●●●●●● ●●●● ● ● ●● ● ● ●●●● ●● ●●●● ●●● ● ● ●● ● ●● ● ● ●● ● ●●● ●● ● ● ● ● ● ● ● ●●● ● ● ● ●●● ● ●● ●● ●● ●● ● 50 ● ● ●● ● ●●●● ●● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ●●● Number of Candidates ● ●● ●

● 20

50 100 200 500 2000 5000

Number of Groups

Note: Distribution of the number of candidates in each state house cycle. There is a weak positive correlation between the number of groups and candidates in a cycle. The line is a regression line and the correlation is 0.21

Figure 3.2: Percentage of Non-Zero Donor-Candidate Pairs 0.20 0.15 0.10 Density 0.05 0.00

0 5 10 15

Percent Group−Candidate Pairs

Note: Histogram and density plot of the percentage of non-zero donor candidate pairs for each State-House- Full-Cycle. In all states groups donate to only a small percentage of the possible candidates.

42 Figure 3.3: Median Donation in Each Dataset 0.004 0.003 0.002 Density 0.001 0.000 0 500 1000 1500 2000

Median Donation

Note: Histogram and density plot of the median total donation for all donor-candidate pairs in a given State-House-Full-Cycle. This only includes donor-candidate pairs where donations exists (so excludes zeros). donors, the degree they participate, and how much they donate.

3.2.1 Inferring Relationships

In order to infer relationships between candidates I need to use methods developed for social network analysis. In the donation data, groups are connected to candidates through donations, but groups are not connected directly to each other. Within social network analysis terminology, the groups and candidates are both nodes and this is a bipartite network as there are two sets of nodes which cannot connect directly to a node of similar type (groups cannot connect to other groups and candidates cannot connect to to other candidates) (Davis and Gardner 1941; Breiger 1974). The network of donors and groups can be represented in matrix notation, where G is the number of groups, and C the number of candidates then A is a G × C matrix. Each element in the matrix, ag,c, is the amount donated from group g to candidate c.

43 It is possible to naively infer connections between groups by examining how often they contribute to the same candidates. If two groups never donate to the same candidate, then these two groups are not connected. If two groups donate to the same candidates, then they are connected and that connection can be weighted by how many shared donations they have and the amount that they donate to those shared candidates. Formally this involves taking the crossproduct of the A matrix. This produces a G × G matrix B, where each element of the matrix is the shared donations between two groups weighted by the size of the shared

PC donations, such that bi,j = c=1 ai,c · aj,c. If a group i does not donated to candidate c, then

ai,c = 0 and so the amount that group j donates to candidate c does not affect the sum. If

both groups donate to candidate c, then ai,c, aj,c > 0 and so ai,c · aj, > 0. The weights are naive measures of how close two groups are connected (Latapy, Magnien and Del Vecchio 2008). A large weight on the edge between two groups means that they are more tightly bound. Using these weights directly though is problematic as groups vary in the amount of resources they are willing or able to provide candidates. A group with lots of resources will be shown as more tightly bound to other groups because it is able to provide more resources to candidates. In addition, as discussed before, groups have different reasons for donating. Groups that donate in order to gain access will donate to many candidates independent of ideological position (Snyder Jr 1992). Candidate propensity to receive donations will vary based on the candidates position within the legislature as well as their incumbency status (Ramsden 2002; Fouirnaies and Hall 2014). The weights and edges will reflect these factors which are not necessarily part of the party network process, which is what is of interest here. Social network analysts have developed methodology to move beyond the reliance of naive edge weights. This is known as backboning and is used when a researcher wants to identify the most important connections within a network, which is called the backbone of the network. The most basic from of backboning is employing a single threshold on the weights across

44 the network. Edges between groups only exist in the backbone if the weight on the edge exceeds the threshold (Latapy, Magnien and Del Vecchio 2008; Watts and Strogatz 1998). The backbone network does not have any weights on the edges. For example a threshold of 0 can be used, which would preserve edges between all groups that share any donations. Increasing the threshold means that groups would have a larger amount of shared donations in order to have a backboned connection between them. The specific threshold value must be determined by the researcher and has a strong effect on the backbone (Butts 2009). Using a single threshold to backbone the network would hide the problem of differential resources between groups but not incorporate the differences between group propensity to donate and candidate propensity to receive donations. More advanced methods have been propose to backbone networks that start as bipartite networks. One common approach relies on modeling actor characteristics (in this case the actors are groups) through a variety of different methods (Serrano, Boguná and Vespignani 2009; Borgatti and Halgin 2011). This fails to account for variation in the candidates (which would be referred to as artifacts from this perspective). One way to account for both candidates and groups is to simulate the entire bipartite network many times (Horvát and Zweig 2013), but this is often computationally intractable (Bezáková 2008). Neal(2014) proposed a novel method that acknowledges variation across groups and candidates but is computationally tractable except in the largest of networks. His method relies on the creation of a null distribution of edge weights between groups. To estimate the null distribution a researcher models the process of a group donating to a candidate as a function of the propensity for a group to donate, the propensity of a candidate to receive a donation, and the interaction between the two. Formally this is:

C G C !  G  X X X X ai,j = α + β1 ai,j + β2 ai,j + β3 · ai,j  ai,j i=1 j=1 i=1 j=1

45 where α, β1, β2, β3 are estimated from the data. After estimating the coefficients, they are used

to draw a large number of simulations of aˆi,j. The crossproduct of each resulting simulated donation matrix is taken. This leads to a distribution of edge weights for each group dyad. The researcher then includes only edges within the backboned network when the observed weight exceeds a percentile threshold in the empirical distribution of edge weights (here I employ a 97.5% threshold). For complete details of this process see AppendixA. Through this process I remove an average of 92% of the edges from each dataset. I use this method to generate networks for each State-House-Full-Cycle, 699 donor networks in total. Figure 3.4 shows two of these networks. The one on the left is the the network from 2013-2014 lower house cycle in South Carolina and the right plot is the same cycle for Colorado. Each dot, or node, is a group (intense policy demander) and they are colored as a function of the amount they donated to either Republicans or Democratic candidates. Blue dots donated 100% to Democrats, red 100% to Republicans and the black dots donate equally to both. Note, I am not displaying any isolated nodes (which are not connected to anyone). These two networks show some of the range of variation found across party networks. The Colorado network, on the right, shows clear division between each party with very little overlap. The Democratic network is larger than the Republican network, but each tend to have more connections with groups internal to the party than external. In contrast in South Carolina there is more overlap, and the Democratic network is much less distinct. This can be seen in the fact that there are red dots (Republican donating) throughout the area where the majority of donors are blue (Democratic donating). It is possible to use this donation data to examine only those that tend to donate to one party. I do this by weighting the edge between two donors by multiplying together the percentage of donations from each donor to Democrats (or Republicans if I am examining the Republican network). This means that when two donors both donate solely to Democrats the edge between them has a weight of 1, and shrinks towards 0 as one (or both) donates

46 Figure 3.4: Examples of Full Donation Networks

(a) South Carolina House Network 2013 to 2014 (b) Colorado House Network 2013 to 2014

Note: The full donation network in South Carolina and Colorado state house elections in 2014. Each node is a donor and is colored based on the percentage it donates to Republicans or Democrats. These networks show the degree of separation between partisan donors that can exist. In Colorado there is little overlap between out-partisan donors, Democratic and Republican donors cluster with their co-partisans. In contrast in South Carolina, although there is some partisan structure, it is much less pronounced.

47 more to Republicans. If either donates to Democrats, then the edge has a weight of 0. This creates a network where ties exist only when groups donate in a similar way, and when those donations go to one party. Figure 3.5 shows the Democratic donation networks and Figure 3.6 shows the Republican donation networks from Colorado and South Carolina. Before examining who is in the network, it is worth noting that we see similar patterns to what we noted before. The Democratic network in Colorado is tightly bound together whereas in South Carolina groups do not have as many connections with their co-partisans. The Republican networks are similarly dense in each state but the South Carolina Republican network is large compared to the Colorado network. In Fig 3.5 and 3.6 we can also see the types of groups that make up the donation network. Using coding from the NIMSP I have color coded the nodes as a function of several types. Businesses, in the light green, are the most common organizational type in all the networks. This is especially true within the South Carolina Republican network where there are many business groups that are densely connected to each other. In contrast, the Colorado Democratic network has a large number of labor unions that are deeply embedded in the network. The fact that labor unions are more deeply involved in the Democratic network than in the Republican networks fits with our knowledge of labor politics in the US. The Colorado Republican network has a lot of deeply embedded formal party organizations. This is not true of the South Carolina Republican network where the formal party organizations have few connections and so are generally on the outside of the network. In this section I have used donation data to develop networks of donors across 47 states from 2000 to 2016. The relationships within the networks are captured by identifying similarity in donation patterns that are unexpected given the general propensity of groups to donate and candidates to receive donations. These party networks show a lot of variation across states and party, both in who is involved and the overall structure of the network.

48 Figure 3.5: Examples of Democratic Donation Networks

(a) South Carolina House Network 2013 to 2014 (b) Colorado House Network 2013 to 2014

Note: Example of Democratic donation networks for South Carolina and Colorado state house races in 2014. Donors had to donate to some extent to Democratic candidates. The edges between donors are weighted based on the amount that each donate to Democratic candidates, a wider edge indicate that both donate majority to Democrats, as the edge becomes thinner than either one of them or both of them donate at a lower percentage to Republicans. Nodes are colored based on modified categorizations from NIMSP. The Colorado network again shows tight clustering for Democratic donors while the South Carolina does not.

49 Figure 3.6: Examples of Republican Donation Networks

(a) South Carolina House Network 2013 to 2014 (b) Colorado House Network 2013 to 2014

Note: Example of Republican donation networks for South Carolina and Colorado state house races in 2014. Donors had to donate to some extent to Republican candidates. The edges between donors are weighted based on the amount that each donate to Republican candidates, a wider edge indicate that both donate majority to Republicans, as the edge becomes thinner than either one of them or both of them donate at a lower percentage to Republicans. Nodes are colored based on modified categorizations from NIMSP. Both networks show a fairly dense network. The South Carolina network is larger and more dominated by business interests than the Colorado network.

50 3.3 Validation

Before using the donation networks to test and develop theories it is necessary to validate that they capture the underlying construct: the party coalition. Construct validity is critical and especially difficult when trying to validate measures of relatively new constructs. In order to test that these donation networks are valid I will lay out a variety of expectations and test to see how they hold within the networks. I start first with very simple concurrent validity (Trochim and Donnelly 2001), that the Republican and Democratic networks ought to capture a different set of actors. Next I examine another set of concurrent validity concerns—the relationship between groups with known similar interests. In this case I focus on a set of donors that we would expect to have similar interests. A lack of relationship between these intense policy demanders would indicate that the donation data is not collecting shared interests.

3.3.1 Republican and Democratic Networks

The first test of validity assumes that the party coalitions ought to be different from each other. This is not to argue that they should be entirely divergent, but that as a group becomes more influential in one network they ought to become less influential in another network. The partisan polarization in American politics at this time (Fiorina and Abrams 2008; Levendusky 2009), makes it unlikely that there would be a large amount of overlap across party coalitions. Put another way, we would expect that groups that donate to Republicans tend to have relationships with groups that donate to Republicans. Relationships are not assumed to be independent of partisan donation patterns. To test this I calculated the assortative coefficient in each network by the percentage of donations to the Republican Party. Assortative coefficients are similar to correlation coefficients but measure the relationship between an attribute of nodes in a network and who

51 they have relationships with. Like correlation coefficient it ranges from -1 to 1 with values close to 1 indicating that groups with the same attribute are more likely to be connected. The attribute I use here is the percent a group donates to the Republican Party and so I expect that the assortative coefficient should be close to 1. This indicates that donors that donate a high proportion to Republicans (Democrats) tend to be connected to donors that also donate a high proportion to Republicans (Democrats). I calculate this independently for each State-House-Full-Cycle network. Figure 3.7 plots the density of these coefficients. The mean across all party networks is 0.61 and the range is almost entirely limited to positive coefficients. This indicates that in most State-House- Full-Cycle, the donation networks are polarized based on party donations. Groups that donate more to Democrats tend to be connected to each other while groups that donate more to Republicans also tend to be connected to each other. The donation networks, after backboning, do tend to identify patterns that we expect. Another way to test to see if the networks capture partisan polarization is to compare how important intense policy demanders are within each party. Again I expect that as an intense policy demander becomes more deeply embedded in one party it should become less embedded in the other party. I employ two measures of how embedded a groups is. First I use the number of other donors it is tied to (the degree). Second I use a Eigenvector Centrality which measures not only how many donors a donor is connected to but also how many donors the secondary donors are connected to and so on (Newman 2010). In both cases larger values indicate that a donor is more central to the network. To check how embedded they are within a party I use the party weighted edges. Again, this is two sets of edges for each network where the edges are weighted as a function of the proportion of Republican (or Democratic) donations. For each State-House-Full-Cycle in my data I calculate the correlation between the degree of a node using the Republican edge weights and the degree of a node using the Democratic edge weights. I do the same but use

52 Figure 3.7: Assortative Mixing by Party Assortative Mixing by Party 1.5 1.0 Density 0.5 0.0

−0.2 0.0 0.2 0.4 0.6 0.8 1.0

Assortativity Coefficient

Note: Density plot of the assortative coefficients for percentage of donations to Republican candidates for all State-House-Full-Cycle in the data. A coefficient of 1 indicates that groups with similar partisan donation patterns have relationships. In almost all cases this is positive, indicating that there is partisan polarization in the networks, as is expected.

eighenvector centrality instead of degree centrality. A high correlation here indicates that a node that is central in the Republican network is also central in the Democratic network. Figure 3.8a and 3.8b plot the density of the correlations between the degrees and the correlation between eigenvector centrality. The mean correlation for the measures of degree is -0.36 and for eigenvector centrality it is -0.32. In both cases 95% of correlations are below 0. The plots show a distribution that is relatively tight around their means. The negative correlations indicate that on average there is a slight inverse relationship between how central a node is in the Democratic Party and how central it is in the Republican Party. The donors and their place within the network are a function of the partisanship of their donations. Groups with similar partisan donation strategies are more likely to have ties with each other than would be expected if they were donating independently of partisan strategies. Similarly, there is a trade off among donors. As they become more central to one

53 (a) EigenvectorCentrality across Correlations Parties (b) DegreeDegree Correlationsacross Parties 2.5 2.0 2.0 1.5 1.5 Density Density 1.0 1.0 0.5 0.5 0.0 0.0

−0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4

Correlation Correlation

Note: Density plot of the correlations of measures of centrality within each party. The left plot uses an Eigenvector measure of centrality and the right uses degrees. Negative numbers indicate that as a donor becomes more central to one party they become less central to the other. This is found to be the case using both measures, showing that partisan polarization is captured within the networks. partisan coalition they become less central to the other. Taken together this indicates that the networks developed here capture an important aspect of contemporary American politics.

3.3.2 Groups with Similar Interests

Now I examine particular relationships that exist within the donor networks. In developing these networks I assumed that donations reflect shared interests. If the process outlined above is able to identify shared interests, then groups with known shared interests should be more likely to be connected. To test this I can examine how likely a group is to be connected to another group given that they have similar interests. This test requires examining the generative process of the network. The generative process of the network can be modeled using an Exponential Random Graph (ERGM) model. I explain these models in full in Chapter 5. For now it is necessary to understand that these models allow researchers to estimate the likelihood of two nodes being tied given their

54 own covariates while also controlling for the network structure. Here I am interested in the existence of homophily, where similar nodes are connected to each other. This sort of homophily can be directly modeled in the ERGM setup and so I can test for it. With the ERGM framework I can test for homophily, but I first need to identify relevant interests. I return to the NIMSP coding used above and examine homophily between labor unions and businesses. Labor unions include both public and private unions. Labor unions will generally have agreement about preferences, working towards strengthening workers rights. Businesses make up a variety of groups within the NIMSP data and includes all for-profit organizations that are lobbying on their own behalf.3 Businesses are less likely to agree on every issues but do likely share broad interests in lowered taxes and other general pro-business policy. Chapter 5 and AppendixC contains a full description of these and the other terms I use. Figure 3.9a plots the 95 percentile confidence intervals for coefficients for homophily among labor groups. Each line is a coefficient from a different State-House-Full-Cycle and black lines are statistically significant using a 95% two-tailed test. The coefficients can be interpreted in a similar way to a logit coefficient. A positive coefficient indicates that labor donors within a network are statistically more likely to connect with each even while controlling for a variety of donor and network characteristics.4 In the case of labor groups, 83.3% of coefficients are significant. Many of these coefficients are substantially large as well. Labor groups are likely to connect with each other in these networks. Figure 3.9b plots similar estimates for businesses within each network. Unlike the labor interests these are less uniformly significant. This is unsurprising given that the category includes a more diverse set of interests than the labor category. Still 77.5% of the coefficients are significant indicating that in general there is a preference for business to tie together in

3Excluded lobbyists who are for-profit but lobby on the behalf of others 4The additional variables are discussed in detail in AppendixC

55 Figure 3.9: Measures of Homophily

(a) Coefficient onLabor Labor Homophily (b) Coefficient onBusiness Business Homophily

−4 −2 0 2 4 6 8 10 −4 −2 0 2 4 6 8

Coefficient Coefficient

Note: Plots of the 95% confidence intervals for coefficients on labor homophily and business homophily. Red lines cross, or are entirely below, the 0 line. A positive coefficient indicates that labor (business) groups are more likely to develop ties to other labor (business) groups. The majority of these are significant indicating that the relationships identified are capturing that groups are working together.

56 these networks. Examining the ERGM coefficients shows a clear pattern of groups with similar interests being more likely to be connected. The edges created through the backboning process are indicative of actual relationships across groups. Labor unions were especially likely to form connections with each other, which again would be expected as they have the closer shared interests than businesses.

3.4 Conclusion

In this chapter I have developed the party coalitions/networks that will be the focus of the analysis in future chapters. Party coalitions are informal structures that are not easily delineated. This lack of formality does not make them less important, in fact this informality makes them very robust organizations that are hard to regulate or control (Masket 2016). The informality does though pose a challenge for identifying them directly. To overcome this challenge I used donation data as well as a new method for identifying relationships among individuals within a network. Doing this I can identify connections between groups based on their donation patterns while controlling for a group’s propensity to donate, and a candidate’s propensity to receive donations. This process is critical. In order to make inferences about the party coalition it is necessary to first identify the party coalition and connections within it. Without properly identifying these connections any inferences from the party coalition structure would be suspect. Validating these relationships is difficult though as there are no clear analogous measures. In order to validate them I focus on what we know about party coalitions. That they should be different across parties and that groups with similar preferences should be close within the coalition. I found that the party networks are distinct from each other, with very little overlap between them. In addition, labor unions are found to often have relationships with each other,

57 even while controlling for a variety of reasons for them to connect. This provides evidence that the party coalitions identified here reflect the expectations of what party coalitions are. Next I explore how party coalitions impact the legislative process. In the next chapter I take up their role in the developing state legislative ideology, showing that money alone does not buy power within the party. Legislative ideology also reflects the placement of groups within the party, not just the amount of money they are willing to put into the party. In the following chapter I show how party coalitions bleed into the legislature by altering the ability of legislators to agree with each other and work together to pass legislation.

58 Chapter 4 | Party Ideology

4.1 Introduction

Variation across state legislative parties is well documented in both pop culture and political science. In 2011 to 2012, Republican caucuses in both Wisconsin and Michigan launched broadsides against unions through right-to-work legislation while their co-partisans in Penn- sylvania sat idle. In the 2000s Democratic controlled legislatures passed laws protecting transgender individuals from employment discrimination, yet some states like Maryland, with a Democratic controlled legislature and governor for much of this time, did not implement similar protections. What leads legislators within the same party to be more or less extreme than their co-partisans in other states? Understanding the variation in state parties has been a major research field in American politics. In attempting to explain these differences scholars have examined the role of state electorates (Brown 1995), party cultures and development (Mayhew 1986; Gimpel 1996), party organizations (Roscoe and Jenkins 2015; Cotter et al. 1989) and partisan competition (Key 1949). Missing from this research are the coalitions outside the legislature that support and mobilize the party (Bawn et al. 2012). As argued earlier, these coalitions work to elect legislators with views that are congruent to the structure of the party coalition. This process

59 leads to the creation of party ideologies which eases future interactions among the party coalition. Party ideology though is not permanent nor is it dictated from philosophical understandings of human nature, instead it is a reflection of the party coalition as it currently stands and will change as the party coalition changes. Understanding the sources of party ideology is especially important given that parties have been shown to be more responsive to the interests of the wealth (Rigby and Wright 2013; Schlozman, Verba and Brady 2012). The effects of this unequal representation will only be heightened if current trends towards increasing inequality continue Piketty(2014). If party ideology is reflective of relationships, as I argue, then policies that target only the role of money in politics will not necessarily lead to more equal responsiveness. Instead, reformers that are concerned about the outside impact of the wealthy will have to find more imaginative and potentially more radical solutions. In this chapter I use the party coalitions developed in the previous chapter to test my theory of party ideologies. I find that the Democratic and Republican Party ideology is a reflection of the structure of the party coalition. becoming more conservative as business groups become more central to the party coalition while becoming more liberal as labor issue groups become more central. This is true even when controlling for the amount of resources provided by these groups. I start by outlining briefly the relationship between party coalitions and ideology, as well as discussing why their might be asymmetries across the two parties. I use the party networks developed in the previous chapter and coding from the NIMSP to measure how central different sets of intense policy demanders are within each party coalition. These measures are my main independent variables that I use to model party caucus ideology.

60 4.2 Party Coalitions and Ideology

Party coalitions exist to elect candidates to office. Before electing candidates to office though party coalitions need to first select candidates to support. As discussed earlier, candidate selection is important as it defines the party ideology. Party ideology comes about as coalitions identify the issues that are important to them, issues that are critical and so are ‘litmus test’ for support, and issues that they are not concerned about. This set of preferences is reflected in the candidates that are selected by the party networks. For example, in the 2010s discourse among Democratic Party elites revolved around if being pro-choice was a necessary qualification for a Democratic candidate (Kamisar and Wilson 2017). The type of candidates selected and their positions on abortion is how the Democratic Party network decides their aggregate position on abortion. Not every coalition member though is as influential in the process of selecting candidates. This inequality has an important effect on the party ideology as the most influential intense policy demanders will, by definition, be able to pull the party ideology closer towards them. Returning to the example of Democrats and abortion, the more influence pro-choice groups have, at the local, state, or national level, the more they will be able to enforce abortion rights as a litmus test for candidates. Candidates will not be supported by the party network, and so the party will move towards a stronger pro-choice position. Understanding what makes an intense policy demander influential is critical for under- standing variation in party ideology. Often influence is assumed to be related to a groups ability to provide resources. As discussed earlier, candidates need resources to win elections. Money, as the most fungible resource, is the most important, but candidates also need volunteers, strong staff, media access, and information on voters. A group with the most of these resources ought to have influence on the ideology of political parties. This fits with a broad set of academic and popular literature that has argued that moneyed interests have

61 undue influence in the political process (Gilens 2012; Schlozman, Verba and Brady 2012). Political parties are not just a bundle of resources where the group or individual with the most amount of money is able to dictate the process. Parties are complex webs of relationships among intense policy demanders with similar interests in taking political power but sometimes divergent interests in what to do with that power. I argue that this has clear implications for what it means for a group to be influential. Since parties are networks of interests, the relationships that exist between intense policy demanders are important indicators of influence. Groups can use those relationships to bring others closer to their position. These relationships are useful because they can be used to pass information, coordinate privately on candidates, or keep each other aware of what issues are important to each other. In sum, relationships ought to be able to be leveraged to gain more influence over the party. Returning to the example at the beginning of this chapter, the Republican Party’s position on right-to-work legislation we can see how these relationships can be important. Although Wisconsin and Michigan, both states with long union history, passed right-to-work legislation after the 2010 election, Pennsylvania did not. Labor unions remain an important part of the Pennsylvania Republican Party coalition. This is obvious by examining the party networks created in the previous chapter. Teamster locals, the Pennsylvania State Troopers Association, and the International Union of Painters and Allied Trades have hundreds of connections in the Republican coalition. As a validation of this we can look for other demonstrations of influence within the Republican Party. When Republican State Senator John Rafferty declared his intent to run for Pennsylvania attorney general he did so from the Pennsylvania State Troopers Association headquarters (Murphy 2015). In the 2010 election, the Pennsylvania AFL-CIO endorsed 5 Republican State Senate Candidates and 16 Republican State House candidates. In contrast, the Michigan AFL-CIO, although endorsing candidates in almost every state legislative race, did not endorse a single Republican candidate. Although some Michigan unions had relationships within the Republican Party, there were very few, and so when

62 right-to-work legislation was proposed there were no Republicans to speak against it. Based on the coalitional or network structure of the party, I hypothesize a different source of power within the party. Instead of money buying influence, centrality to the party network buys influences. Groups that are more central to the party network will be able to persuade more groups to prioritize their issues over the issues of other groups. Labor unions within the Republican Party in Pennsylvania have connections to others within that party and so can advocate within the party against right-to-work legislation. This is not the case in Michigan where labor unions are almost entirely contained to the Democratic Party. Viewing these actions directly is difficult as these sorts of conversations are not often public. What is public though are the network structures left by groups as they navigate one part of the political process, candidate donations. In summary, I have two hypothesis about how groups will influence the political process. One based on resources and devoid of the relationships that are supposed to be at the root of political parties, and the second one that embraces the network structure. These hypotheses are:

Hypothesis 1 (Resources): Political ideology will be responsive to the intense policy de- manders that provide the most resources to a party.

Hypothesis 2 (Relationships): Political ideology will be responsive to the intense policy demanders that have the strongest set of relationships within a party.

4.2.1 Differences Between Republicans and Democrats

The theory proposed here assumes that there is potential room for variation within the party coalitions understanding of their political party. That not all Democrats agree on what it means to be a Democrat and that not all Republicans agree on what it means to be a Republican. Given that American political parties are embedded in a federal system with

63 variation across states and that the positions of both parties on a range of issues has varied significantly, this seems to be a relatively weak assumption. Recent work by Grossmann and Hopkins(2016) has challenged this assumption though. In particular Grossmann and Hopkins(2016) argue that the Republican Party is internally coherent, reflective of an ideological conservative movement. The Democratic Party, in contrast, is a party of groups and so reflects the interests of those groups. This can be seen most obviously in how candidates discuss themselves in campaign ads. Republicans consistently use the language of ideology to describe themselves as conservative and their opponents as liberals, whereas Democrats almost never refer to themselves as liberal (Neiheisel and Niebler 2013). Democrats instead rely on connections to groups, and specific policies that speak to those groups. Democratic supporters view themselves as being part of some group that supports Democrats, whereas Republicans view themselves as being a conservative and so therefore a Republican (Grossmann and Hopkins 2016). Asymmetry between the two parties is not in conflict with the idea of party coalitions or networks. This sort of asymmetry, where one party is much more unified around an ideological label, would exist if the party coalition is it self very strongly united. The Republican network has calcified around an idea of Republican ideology and all influential members of the network agree on this. The Democratic network continues to feature disagreement over what it means to be a Democrat. The differences between the two parties leads to a third hypothesis:

Hypothesis 3 (Party Differences): The Democratic Party will be more responsive to variation within the party coalition.

64 4.3 Testing the Effects of Network Position

I argue that as policy demanders become more central to a network they will be able to better dictate the positions of the party. To test this I need to quantify the strength of policy demanders, their ideological positions and the ideology of the party. As discussed, I examine two types of strength, one based on relationships the other based on resources. In order to measure the relationship strength I return to the networks identified in the previous chapter, whereas resource strength is measured using the initial donation data. To test the theory of policy demander strength on party ideology, I use a cross-sectional approach with the state-year as the unit of analysis. I categorize policy demanders into coherent sets with shared ideological preferences that are assumed based on their interests. I use the strength of these sets within each network to explain variation in party ideology, while controlling for a variety of potential confounders. I detail this after explaining how I measure policy demanders strength.

4.3.1 Policy Demander Relationships and Resources

It is necessary to categorize or place policy demanders in ideological positions. The predomi- nant measure of donor ideology comes from Bonica(2014). Bonica(2014) uses donation data to estimate a latent ideology for candidates and donors. Because this measure is generated by using donation data, which contains the data that is used here to estimate the networks it does not provide an independent assessment of ideology. The assumptions made by Bonica (2014) and I about donations are similar, but the fundamental goal is different. Bonica(2014) focuses on developing individual level information about candidates, donors, and elected officials. Here I focus on the macro-level, in particular about the relationships between donors within a state. Therefore, I begin by creating four categories of intense policy demanders, which I call

65 industries, based on the data provided by the National Institute of Money in State Politics. NIMSP provides three levels of coding for each donor, from least specific to most specific they are: Group, Industry, and Business. The four policy demander industries I create are business interests, labor interests, liberal ideological interests and conservative ideological interests. For the business interests I include any groups that are coded as Chamber of Commerce or Pro-Business organizations in the Business code, this captures groups that lobby over business interests but are not themselves businesses. Labor interests are groups coded as Labor in the Group code. Liberal interests include those labeled as Drug Legalization Advocates, Pro-gun control, Public school advocates, Animal rights, Minority and Ethnic groups, Gay and lesbian rights and issues, Abortion policy, pro-choice, Environmental policy and Democratic- based groups (but not official party committees) and generic liberal/progressive ones. Finally, conservative interests include Labor, anti-union, Christian Coalition, religious right, School choice Advocates, Abortion Policy, pro-life, Anti-gun control and Republican-based groups (but not official party committees) and generic conservative ones. To measure the relationship strength of each of these industries within a party I need to first calculate the strength of the relationships of the constitute policy demanders that make up the industry. This measure should account for not only how many relationships a group has but also who it has relationships with. A relationship with another group that itself has a lot of relationships is a more useful relationship than a relationship with a group that has few relationships. This type of measure can be calculated using Eigenvector Centrality which is a way to weight the number of relationships a group has by how many relationships each group it is connected to also has. Intense policy demanders with a high Eigenvector Centrality score will have a lot of relationships with groups that have a lot of relationships. If I calculated the Eigenvector Centrality scores on the whole donation network I would have a measure of how central they are to all donors, not just within a party. To measure just centrality within a party, I use the party edge weights described in the previous chapter.

66 These edge weights reflect the percentage of donations made to the party of interest by the two groups that are connected. Centrality using the edges weights identifies who is central to the set of donors that donate chiefly to a single party. After calculating Eigenvector Centrality scores for all individual intense policy demanders I can identify how central each one is to either political party. Finally, I aggregate the individual scores into an industry level score by calculating the percentage of the network that the industry makes up, weighted by the Eigenvector Centrality of each intense policy demander. Formally for each of the four industries I calculate:

  Pn 1 if group i is in category k 1 (EVi · δi)  Sk = Pn δi = 1 (EVi)  0 otherwise

where EVi is the Eigenvector centrality of group i. This measures the percentage of relation- ships that an industry has out of the total percentage of relationships that exist. Finally, because the ideological industries exist on a similar continuum I take the difference between conservative ideological industry and the liberal ideological industry strength. Negative scores indicate that liberal groups have more relationship strength and a positive score indicate that conservative groups have more relationship strength. This weighted percentage counts groups more heavily if they are deeply embedded in the network. For example, as the weighted percentage of the labor industry increase it means that there are either more labor intense policy demanders in the network, or that they are more deeply embedded in the network (connected to more groups), or that both of these things are true. As these labor groups increase how embedded they are the policy position of the party should become more liberal. Figure 4.1 shows box-plots of how embedded these industries are within each party network in the data. What is initially clear is how important the labor industry is within Democratic

67 Figure 4.1: Network Relationship Strength Scores

(a) Business Strength (b) Labor Strength

● ● ●●● ●● ● ● ● ● ● ● ● ●●●●●● ●● ●●●●●● ●● ● ● House Rep House Rep

●●● ● ● ● ● ●● ● ● ● ●●●●●●●●●●●●● ●● ● ● Senate Rep Senate Rep

●●●●●●● ●●● ●●● ●● ● ● ● ● ● ● ● ● House Dem House Dem

●●●● ●●● ●● ●●● ● ●●●● ● ● ● ● Senate Dem Senate Dem

0.00 0.02 0.04 0.06 0.08 0.10 0.0 0.2 0.4 0.6 0.8 1.0

(c) Ideology Strength

● ●● ●●●●●●●●● ● ● ●● ● ● ● ● ● House Rep

● ●● ● ●●●●●●●●●●● ●● ●● Senate Rep

● ● ● ● ● ●●●● ●● ● ●●●●●●●●●●●● ●● ● ●●● ●● ● House Dem

●● ● ● ●●●● ●●●●●●●●●●●●● ● ●● ● ● Senate Dem

−0.2 −0.1 0.0 0.1 0.2

Note: Boxplots showing the distribution of network strength across the different industries investigated here. The business industry is much more important in the Republican network than the Democratic network whereas the opposite is true for the labor industry. Ideological scores move in the expected direction with Democrats having a net negative (liberal) score and68 Republicans having a net positive (conservative) score. See text for details on how the industry types are defined. Party networks. The labor industry is involved in the Republican network, but at a much smaller degree than in the Democratic network. In contrast, the business industry is more important to the Republican network than it is to the Democratic network. In total the ideological industry makes up a small proportion of the network, but there are differences across the two parties. The Democratic network tend to have more powerful liberal ideological groups (and so have a negative score for the ideological measure) whereas the opposite is true for the Republican network. To calculate a resource measure I ignore the network structure, and calculate the total amount donated by each industry to each party network. This is normalized by dividing it by the total amount of resources donated by all groups to the party. This provides a measure of the percentage of donations that come from a particular industry, and so show the raw power of an industry. Donations do not capture all types of resources but do capture an especially important type of resource.

4.3.2 Party Position

Party position can be narrowly or broadly defined. Work by Poole and Rosenthal(1991, 2000) has shown that there is a pattern to how Members of Congress vote. In particular, it is possible to place Members of Congress along a single ideological dimension that captures the majority of variation in how they vote. A similar line of research has been carried out at the state level. Shor and McCarty(2011) show that state legislative ideology also often fits well on a single dimension. They are also able to estimate ideology measures that can be compared across state lines and over time. This makes it possible to compare the ideological mean of the Wisconsin legislature in 2006 to the ideological mean of the Arizona legislature in 2010. These broad measures of ideology capture the predominant purpose of the party network,

69 Figure 4.2: Distribution of Party Ideology

(a) State Lower House Ideology (b) State Upper House Ideology

Democrat Democrat 2.0 2.0 Republican Republican 1.5 1.5 1.0 1.0 0.5 0.5 0.0 0.0

−2 −1 0 1 2 −2 −1 0 1 2 Note: Density plots of the state party ideology for the state-years in the dataset. The red is the Republican Party mean ideology, and the blue is the Democratic Party ideology.

to create cohesive ideologies that reflect the interests of the policy demanders within the party. Groups within the party network will pull these ideologies closer to their own position. As such, I expect that party networks with stronger conservative policy demanders will, all else equal, have more conservative policy positions. Conversely, party networks with stronger liberal policy demanders will, all else equal, have more liberal policy positions. To examine my expectations I use the state party mean ideology as calculated by Shor and McCarty(2011) as the dependent variable. This ranges from -1.73 to 1.70 with negative numbers indicating more liberal positions and positive numbers more conservative policy positions. Figure 4.2 plots the distribution of the ideology for each party divided by the two chambers. There is minimal overlap between the two parties, and the standard deviations are relatively consistent with the smallest standard deviation for Senate Republicans (0.34) and the highest for Senate Democrats (0.38). Given this dependent variable I expect that the coefficient on union relationship strength will be negative for both Democratic and Republican networks, meaning that unions strength

70 push the parties to be more liberal. The coefficient on business network strength is expected to be positive for both Democratic and Republican networks. The ideological relationship strength has the same polarity as the dependent variable and so the coefficient is expected to be positive. My expectations for the effects of resource strength are parallel to the expectations for relationship strength.

4.3.3 Modeling Relationship Between Party Networks and Ideol-

ogy

I use a linear random effects model with state random effects. The state random effects account for unobserved heterogeneity between states. In total I have 46 states in the dataset. Nebraska is dropped because of its non-partisan elections, and Connecticut, Arizona and Maine are dropped because of their strong public financing laws which make it hard to identify the party network. The data is all from 2000 to 2016, although certain years are not available from Shor and McCarty(2011). In order to test for differences between the political parties I include a dummy variable for the Republican Party and interact it with all other variables. In line with the hypothesis of party asymmetry, I expect that the interaction will attenuate the relationships between party network and party ideology. In addition I include a variety of additional controls as well: Citizen Ideology. First and foremost, there is wide variation in the preferences of citizens between states, with Californians expected to be on average more liberal than Alabamans. Such differences in citizen ideology should lead to differences in party ideology. To account for this, I include Berry et al’s (1998) measure of citizen ideology. This is estimated using interest group scores of Members of Congress and is available from 1960 to 2013. Berry et al. (1998) orient it so that higher values indicate more liberal state ideology, with the maximum being 93.2 and the minimum being 13.4. I reverse the polarity here so that lower values

71 indicate more liberal states and divide it by 100 to put it on a scale similar to that of my dependent variable. I expect the coefficient to be positive in all cases. Again, the Republican Party is expected to have a more attenuated relationship between citizen preferences and party position. Party competition. Initial arguments around party competition viewed competition as a moderating force on political parties, pushing them towards the median voter (Downs 1957). Recent work though has found that competitive pressures lead to more partisan politics and more polarized parties (Carroll and Eichorst 2013; Hinchliffe and Lee 2016). Hinchliffe and Lee(2016) employ five different measures of competition, finding each of them associated with increased polarization. The most consistent predictor they find is an index of state competition which includes a folded average of the gubernatorial vote, and seats in each state house. I include this measure in my model. Given its impact on polarization I expect that it will push both parties towards ideological extremes, so the coefficient will be negative for the Democratic Party ideology and positive when modeling the Republican Party ideology. Chamber party strength. Related to party competition is the strength of the party in the chamber, measured by the percentage of the chamber controlled by the party being analyzed. In uncompetitive states it is possible for a party to either be far into the majority or far into the minority. After controlling for competition, I expect that party strength should be inversely related to their particular ideology. Larger parties will likely be more moderate as their coalition will be more diverse and so will tend towards moderation. The coefficient on party strength for Republicans should be negative, indicating that a larger Republican coalition will tend to be more moderate. The opposite is expected for the Democratic Party. Size of Network. Following the same logic discussed by party strength, it is expected that larger party networks will, all else equal, push towards a more moderate party. This moderation is the result of larger networks leading to a more diverse set of actors and so move towards moderation. Again the coefficient on Party Network size will be expected to

72 be negative for Republicans and positive for Democrats. I log this measure as the data is heavily skewed. Proportion Urban. States with a large urban population have found to be more polarized (Hinchliffe and Lee 2016), potentially as a result of increased two party competition. Democrats and more left leaning voters also tend to be densely packed in urban areas (Chen, Rodden et al. 2013). Taken together this implies that states with more urban population will have a more liberal Democratic Party. There is theoretical reason to expect that Republicans could be pushed more to the left because of large percentage of urban population or that there could be a backlash leading the Republican Party to be more conservative. Data on percentage urban is from Hinchliffe and Lee(2016). Finally in AppendixB I include controls on the limits on donations in each state. Recent work has shown that these controls can have polarizing effects (Barber 2016). The main results presented here are substantively the same and I find no significant effect for donation limits on the mean party ideology.

4.4 Results

Table 4.1 displays the results of modeling party ideology. Because I interact all other variables with the Republican dummy variable, there are a large number of interaction terms. In the first column I display the main effects while the second column shows the coefficient on the interaction terms. Substantively the first column shows the effect for Democratic state caucuses while the second column shows the difference in effect between the Democratic and Republican Party, and so a significant term here indicates that the relationship is different depending on the party that is analyzed. In addition in Table 4.2 I show the effects conditioned on each party. The first column shows the effects conditioned on the Democratic Party and so is the same as the first column in Table 4.2. The second column shows the effect conditioned

73 on modeling the Republican Party. Clear patterns emerges from the estimated coefficients that supports the role of policy demander strength within both parties. In both the Republican and the Democratic Party business and labor groups are able to use their relationships within the party network to influence the position of the legislative caucuses. The coefficient for business relationship strength is 0.84 for the Democratic Party and 0.98 for the Republican Party indicating that as more business groups become embedded within the party network the caucus is pushed to be more conservative. The effect is not significantly different across the two parties either, indicating that both seem to be responsive to business interests in their network. The relationship is similar for labor union relationship strength in each party. The coefficient for Democratic Party is -0.17 and -0.59 for the Republican Party. This indicates that as labor unions become more central to each party network they are able to push the related party to take more liberal stances on issues. Interestingly there is a difference between the coefficients for Republicans and Democrats with the Republican Party ideology changing to a larger degree than the Democratic ideology. This is not expected given the arguments of Grossmann and Hopkins(2016) as it indicates that the Republican caucus preferences are more responsive to the makeup of their coalition than the Democratic caucus. The coefficient on the relationship strength for ideological policy demanders are not significant for either party. In contrast, the coefficients for single issue groups are significant for their donation strength. The coefficient for the Democratic Party is 1.77 and for the Republican Party it is 1.05. This variable is scaled so that positive coefficients indicate that the the caucus is responsive to the balance between the amount of donations from liberal and conservative issue groups. In this case as liberal groups donate more than conservative groups to the party the party will become more liberal, and the reverse is true if conservative groups donate more. The fact that donation strength is significant while relationship strength is not indicates that these ideologically aligned groups are able to influence policy through

74 money more than they are through relationships. The coefficients on labor donation strength tells a similar story as the relationship strength. Labor is able to garner influence both through the donations they make as well as the relationships they build across party networks. And again the effect seems to be more extreme in the context of the Republican Party than in the Democratic Party. In contrast business donations do not have a significant effect on the Democratic Party but do on the Republican Party. The effect for the Republican Party is in the opposite direction than expected with more donations leading to a more liberal Republican Party. The opposite effect might have two potential causes. First, business interests might at times act as a moderating influence on the Republican Party, depending on the state. The business industry might need to be better categorized to capture differences within it which would lead to the different signs of the coefficients on donation strength versus relationship strength. Second, this might demonstrate a different causal stories. When business groups are not deeply embedded in a party network the party will be more liberal. In response business groups attempt to buy back the party through donations but are unable to influence it which leads to a negative coefficient as business interests only rely on donations when a party is relatively liberal. Moving beyond the party network variables there are interesting results for several of the controls. Competition has a moderating effect on both parties. The coefficient on the Democratic caucus is 0.80, and for the Republican caucus it is -0.66 meaning that as competition increases the Democratic caucus becomes more conservative while the Republican caucus becomes more liberal. The effects of party networks is contra to what was expected. The coefficient on the log number of groups for the Democrats is -0.06 and for Republicans it is 0.04 and significantly different from 0. The Democratic caucus is more liberal in states with a larger party network while the Republican caucus is more conservative in states with a larger party network. There is no dilution of ideology as the party network grows, instead a large party network reflects a

75 Table 4.1: Analysis Party Caucus Ideology

Main Effect Interaction Term Business 0.84* 0.13 (0.38) (0.42) Labor -0.17* -0.42* (0.05) (0.18) Ideology 0.28 -0.10 Relationship (0.22) (0.33) Business 0.80 -3.91* (0.54) (0.61) Labor -0.80* -1.95* (0.10) (0.28)

Donations Ideology 1.77* -0.72 (0.30) (0.45) Senate -0.02 0.01 (0.01) (0.02) Proportion of Memb 0.41* -0.50* in State Leg (0.09) (0.15) Log Number of -0.06* 0.11* Groups (0.02) (0.02) Competition 0.80* -1.46* (0.21) (0.15) Proportion Urban -0.94* 0.66* (0.22) (0.11) Citizen Liberalism 0.21* 0.17 (0.10) (0.10) Republican 0.48* (0.10) Intercept 0.24 (0.18) State σ2 0.04 (0.21) Residual σ2 0.05 (0.21) N 1730 Note: Model of party ideology. The unit of analysis is the state year. State random effects are included. Asterisks indicate a p-value less than 0.05. See text for more details on variables and specification.

more extreme party. The Democratic caucus becomes more moderate as it take a larger proportion of the

76 legislature while there is not significant effect for the Republican caucus. This indicates an important asymmetry between the two parties, as the Republican Party appears to stay conservative no matter the size of their majority (or minority) while the Democratic caucus shifts. There is similar asymmetric responsiveness to the proportion of the population that is urban. The coefficient on the Democratic caucus is -0.94 indicating that the party becomes more liberal as the proportion of the urban population grows. The Republican caucus does not significantly change. Finally, as expected, both parties respond to the preferences of voters within the state. The coefficient for Democratic caucus is 0.21 and for the Republican caucus it is 0.38. The interaction term is not significant at standard levels of statistical significance and so there does not appear to be evidence for a different level of responsiveness across the two parties. Both of them become more or less conservative as a function of the citizen preferences within the state. To examine the substantive effect, in Figure 4.3 I plot the predicted ideology of the Democratic and Republican caucus across the observed range of business and labor group relationship strength. All other variables are held at mean observed variables for the associated Minnesota caucuses. The effect on the Republican Party are especially pronounced. The labor industry, although not readily embedded in the Republican Party, have a strong moderating effect, pushing the party towards the center, while the business industry pushes them to the extreme. The effects are less substantive for the Democratic caucus but there is still clear movement across the observed range of values.

4.4.1 Predictive Accuracy

I use 10-fold cross validation as an additional check of my hypotheses. If network relationship strength matters then inclusion of these variables should improve the ability of a model to

77 Table 4.2: Effects Conditioned on Party

Democratic Republican Significant Effect Effect Difference Business 0.84* 0.98* (0.38) (0.21) Labor -0.17* -0.59* X (0.05) (0.17) Ideology 0.28 0.18 Relationship (0.22) (0.25) Business 0.80 -3.11* X (0.54) (0.46) Labor -0.80* -2.75* X (0.10) (0.28)

Donations Ideology 1.77* 1.05* (0.30) (0.32) Senate -0.02 0.01 (0.01) (0.02) Proportion of Memb 0.41* -0.09 X in State Leg (0.09) (0.09) Log Number of -0.06* 0.04* X Groups (0.02) (0.02) Competition 0.80* -0.66* X (0.21) (0.22) Proportion Urban -0.94* -0.29 X (0.22) (0.21) Citizen Liberalism 0.21* 0.38* (0.10) (0.10) Note: The coefficients conditioned on the party from the model in Table 4.1. The first column shows the effects for Democrats, the second the effects for Republicans. The third column shows if there is a significant difference based on parties at 0.05 p-value or less (the same information can be obtained by examining the significance of interaction terms in Table 4.1) accurately predict the ideology of the party caucus. Traditional measures of model fit might not accurately capture differences in models as a result of over-fitting the model. 10-fold cross validation is an attempt to limit over-fitting by testing model fit on subset of data that is held out from the estimation. To do this I subset the data into 10 parts, train the model on 9 of the parts, and then calculate the root mean square error (RMSE) for predictions

78 Figure 4.3: Predicted Ideology

(a) Labor Relationship Strength (b) Business Relationship Strength 1.0 1.0 0.5 0.5 0.0 0.0 Predicted Ideology Predicted Ideology −0.5 −0.5 −1.0 −1.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.1 0.2 0.3 0.4

Labor Relationship Strength Business Relationship Strength Note: Predicted ideology for Democratic state house caucus across the observed range of business relationship strength and labor relationship strength. Everything else is held constant at the mean observed values for the associated Minnesota caucus across this time period.

onto the held out tenth of the data. I repeat this process 20 times with different random partitions of the data (Kohavi 1995; Rodriguez, Perez and Lozano 2010). Figure 4.4 shows the distribution of the average RMSE for four different models: a base model with neither donation nor relationship strength variables, a relationship model which is the base model plus the relationship variables, a donation model which is the base model plus the donations variables, and a full model which includes all variables. What I expect is that the full model should perform the best, the base model should perform worse, and the other two should be in between. The mean RMSE for the base model is 0.058, meaning that its out-of-sample prediction for ideology deviates on average by 0.058. The inclusion of the relationships and donation variables separately both improve model fit, moving it to 0.056 and 0.051. The donation variables improve fit more, but both are useful when predicting state party ideology. Most importantly though both appear to provide independent information as the full model continues to out perform both other models and has a mean RMSE of 0.049.

79 Figure 4.4: Repeated 10-Fold Predicted RMSE 0.065 0.060 Base Model

Relationship

0.055 Model

Donation Root Mean Sqared Error 0.050 Model Full Model 0.045

Model Note: Boxplots of the average root mean squared error for the held out subset of the data using 20 different partitions of the data. The lower RMSE indicate that the model is performing better. This shows that the full model preforms well and so that the relationship data does provide useful information when predicting ideology.

This shows that the inclusion of the relationship variables improves the fit of the model, better predicting party ideology. The use of 10-fold cross validation limits the likelihood that this is the result of overfitting the data. The relationship variables do provide useful information when attempting to predict a party’s preference.

4.5 Discussion and Conclusion

The results here provide support for the hypothesis that policy demander relationships are important. Strong policy demanders within both the Republican and the Democratic Party are able to alter the party ideology, shifting it closer towards their own preferences. This

80 shift means that as business oriented interests become more important, both parties become more conservative while labor groups are able to push them both to the left. This adds to the burgeoning evidence of the importance of the party network in shaping our political parties. It also highlights the importance of informal networks that exist. Parties in the United States have often been derided for being weak and ineffectual. This conclusion has been formed by first focusing on the actions and limits of formal party organizations. But, as I show here, informal networks exert influence on parties. Parties are not necessarily weak, but they are much more amorphous than the formal organizations would imply. I show here how the role of informal party organizations is especially important in understanding actions within the legislature. The constellation of interests outside the legislature has an impact of the ideology of those working inside the legislature. They are able to pull the interests of legislators more in line with their own interests and so can have a large impact on the legislation produced at the state level. Explaining the variance in state legislative ideology has become increasingly important as more federal programs are devolved to the state level (Thompson 2008). My results also have important implications for those concerned over equal representation in politics. The preferences of political parties are not simply a function of who provides resources to them, but also reflect the network of relationships that underlie the party. Regulations that target only the source of resources of parties will not necessary succeed in opening the party to unheard interests. Instead it is necessary for much broader changes to the political system which will alter the way that interests work together. This work is only a first cut at establishing the relationship between coalition composition and party ideology. Although I have presented evidence for this relationship, I am unable to firmly establish the causal order. A variety of other work has explored the power of the party network in other cases (Masket 2011; Masket and Shor 2015), but it is still necessary to directly test causal claims here as well. Demonstrating the causal nature of party coalitions

81 is a difficult challenge that will require specific attention to how party coalitions form outside the legislature. This leads into a secondary strain of future research, the mechanisms that lead to the formation of party coalitions and their ability to come into agreement over candidates. I have assumed that they do, and that the agreement is affected by power imbalances, but it remains to be explored what sort of variation might exist. It is plausible that some of the lack of results here are because of variation in party cohesion. If parties do not agree then the strength of different groups within the party might be more important, but if parties broadly agree then strength is unlikely to have an effect. Testing this requires analysis of the party networks that takes into account the network generation to identify network level statistics.

82 Chapter 5 | Party Cohesion

5.1 Introduction

Now that I have explored how party networks shape the preferences of party caucuses I move on to examining how party networks are able to unify party caucuses. The role of party networks is to implement or keep policy that is preferred by those within the party network. They do this by electing a team of legislators that will work on legislation directed at these goals. Legislation though requires a majority (or more) of legislators to support it and so the team that is elected needs to be able to work together. As discussed in the previous section though, there can be disagreement within the party network. If the party network is unable to work together to elect a team that can themselves work together then it becomes more difficult for legislation to be passed. For a party network to work together they must be cohesive, and this cohesion sets the stage for the legislative process. In order to understand the relationship between cohesion outside and inside the legislature I first explore what it means for a party to be cohesive. I do this by discussing examples where cohesion has broke down within a party, something that is most obvious during primary elections. Next I turn to how cohesion outside the legislature ought to be connected to the work of legislators. I do this by discussing broader theories of party within the legislature.

83 5.2 Division in Kansas

To understand why cohesion is important and where it originates from it is useful to look at the case of Kansas and the passing of their landmark tax reform in 2012 and the falling out afterward. At the time this legislation was passed Republicans controlled almost 74% of the seats in the Kansas State House of Representatives, 78% in the State Senate and the Governorship. The tax plan initially proposed by Governor Brownback included both deep cuts to income taxes as well as increase to sales taxes and the removal of income tax deductions to pay for them (Sullivan 2012). Party leadership across the two chambers could not keep together a coalition that supported the end of popular tax deductions as well as increases to sale tax and so these pieces were jettisoned from the final legislation. The legislation was only passed after Republican leadership in the State House abandoned an agreed plan with Republican leadership in the State Senate to continue negotiations and instead passed legislation that the State Senate had only passed as a placeholder until more negotiations took place (Wistrom 2012). Calls for Governor Brownback to veto the legislation came not only from traditional liberal organizations like labor unions and education advocates (Marso 2012) but also from Republicans. Former Republican state legislators, as part of a newly formed a group called the Traditional Republicans for Common Sense, joined efforts to veto the legislation (Wistrom 2012; Hunter 2013). Disagreement within the Republican Party did not stop after the initial passages of the tax cutting legislation. In 2013 Governor Brownback again proposed legislation to cut income taxes to be mainly offset by raising sales tax and was met with resistance from his own party (Hanna 2013a). The legislature, still with large Republican majorities, again had hard time passing the tax cuts and the end deal stopped short of increasing the sales tax as much as Governor Brownback asked for while retaining, and in some cases strengthening, the cuts to income tax (Hanna 2013b).

84 In the years that followed Kansas saw large increases in debt even after steep cuts to government spending (Carpenter 2017; Mazerov 2018). This eventually lead to a revolt among moderate Republicans and Democrats within the Kansas legislature. Starting as early as 2015 Republican leaders in the state legislature began sponsoring bills to increase the income tax in order to end the budget deficits and spending cuts (McLean 2015). The efforts culminated in 2017 when the Kansas legislature voted to override Governor Brownback’s veto of their new tax proposal which increased income tax and ended the new exemptions (Hanna 2017). The Kansas State House voted 88 to 31 for the override with 58% of Republicans supporting it, the Kansas State Senate voted 27-13 for it, again with 58% of the Republican caucus supporting it. Taxes are not the only instance where Republicans within the legislature were divided and ended up reversing their own position. In 2014, as part of an education funding package, the legislature voted to end tenure for Kansas public school teachers (Hanna 2014). The initial vote passed with a bare majority in both chambers and with no Democratic support. In 2017 and 2018 the State House approved legislation to re-implement tenure. In both cases a substantial minority of Republicans voted with Democrats to pass the legislation (Carpenter 2018). In 2018 34 Republicans were joined by all 39 Democrats to vote for it while 48 Republicans voted against it. Although this legislation has not passed the Senate the Republican Party is still clearly divided on this issue. In the case of Kansas we can see what happens when cohesion within a party legislature breaks down. Although Republicans controlled large majorities within each chamber, they had difficulty in creating and sustaining majority coalitions to pass policy. In the case of Governor Brownback’s tax cuts, changes made to the policy in order to create a majority coalition ended up creating a policy that was not sustainable in the long run. This eventually led to the rolling back of the tax cuts with the support of Democrats. Although the case of tenure is not as extreme a similar dynamic was in place, internally the party was not able to

85 keep all legislative members together in order to pass policy. Contrast this briefly with the example of Congressional Democrats when they passed the Affordable Care Act in 2009 and in the years the followed. Although Democrats had majorities within the House of Representatives and the Senate, Senate Republicans threatened to filibuster any efforts. In order to override the filibuster and pass the legislation Democrats in the Senate had to keep every member in support of the bill. The Democrats caucus was coherent enough that they were able to stop any member from defecting and successfully passed the legislation (Kessler 2017). Democrats were able to protect the Affordable Care Act in the following years by keeping their members together. The closest the Republicans came to repealing it was 2017 when the Senate used budget rules to prevent a filibuster. Still the legislation failed because of the defection of multiple Republican Senators. The legislation was only saved by the inability of Republican Senators to be as coherent as the Democrats were when they passed it (Pear and Kaplan 2017).

5.3 Where Does Cohesion Come From?

Where does the differences in these parties come from? Why is the Kansas Republican caucuses so divided, while Democrats in Kansas were able to stay closely united. Here I outline what leads to a cohesive, or as in the example of Kansas, a divided party. In most theories of political parties cohesion is a result of the electorate. When electorates are more diverse the legislators elected by them will find it more difficult to work with each other even when they are of the same political party (Aldrich and Rohde 2001; Weber and Parsons 2016). Political parties might be able to work around division within the party and the electorate (Cox and McCubbins 2005) but they are predominantly passive in determining the division within legislatures. I argue that political parties are not passive in this process, but that cohesion of elected officials is in part the result of cohesion within the political party network.

86 In order to understand how political parties can create or deter cohesion we must remember that political parties are primarily networks of intense policy demanders. In general we assume that these intense policy demanders work together in order to gain policy that is agreeable to everyone within the party network (Cohen et al. 2008; Bawn et al. 2012). But this does not necessarily have to be the case. As groups enter the political system they have limited choices in the party networks that they participate in. The American electoral rules make third parties unlikely to succeed (Riker 1982) and so intense policy demanders that want to participate in a party are likely to participate in one of the major party networks. This leads to parties representing a very diverse set of interests. The process of selecting candidates forces party networks to choose a set of individuals that best represent the party network (as described in the previous chapter this will reflect positions within the network). Conflict will appear through this process when there is disagreement within the party network. Primary elections are likely to always to feature some degree of conflict within the party, but the degree of conflict will vary depending on the degree of conflict within the party network. When conflict within the network is minor there should be only minor conflict during primary elections and it should be quickly resolved after the primary election. Conflict will be resolved because either the differences between candidates are minor in the eyes of most of the party network or because one group within the party network is dominant. In both cases this is unlikely to create a legislative caucus that is divided. When one group is dominant they are likely to be able to nominate similar candidates in other primary elections and when the differences are minor this is likely to be the same across all primaries. In contrast when there are major differences and when there is not a clearly dominant group within the party network conflict will not be limited to the primary but will continue into the legislative caucus. Without a clearly dominant party faction, primary elections will not lead to a single cohesive team and the caucus will have similar factions.

87 This is what took place in Kansas. The Kansas Republican network is far from cohesive. As discussed above, recent division within the party led to the creation of groups such as Traditional Republicans for Commonsense (Hunter 2013). But, division also predates Governor Brownback’s tax cuts. In 2005 Republicans formed the Kansas Traditional Republican Majority which included Republican Party officials as well as a former State Senate Majority Leader in their leadership (Painter 2005). They positioned themselves in opposition to social conservatives and extremists within the Republican Party. Around the same time there was division within the conservative wing of the Kansas Republican Party. The Kansas Republican Assembly, a voice of conservatives within the party, broke apart and the Kansas Republican Action Assembly was created as a more conservative alternative. The two groups endorsed different Republican governor candidates in the 2006 primary (Painter 2006). This led to tough competition within Republican primary races for state legislative office. In 2010 for example, 86 out of 125 State House districts had competitive Republican Primaries (with more than one candidate) while only 74 districts had competitive general elections. In the primary, 10% of the races were won by only 5 percentage points. The division within the Republican Party is despite the fact that Kansas is a relatively homogeneous state. Disagreement within the party network is important to understand the disagreement within the Republican caucus in the legislature in Kansas. As groups within the party network come into disagreement they were unable to nominate a cohesive set of candidates. This divided set of candidates had difficulty working together leading to the confusing and conflictual legislative process described above. The reason that the Kansas GOP had trouble passing conservative tax cuts is because the Kansas GOP cannot necessarily agree on what it means to be a conservative and this disagreement starts before anyone sets foot in a legislative chamber.

88 5.3.1 How Cohesion Matters

The importance of party networks in creating cohesion is dependent on how political parties are understood within the legislature. As discussed there are different theories of how parties operate within the legislature, although disagreement within a party is important to all of them. In order to examine this I take up three main theories of parties in government: responsible party government, conditional party government, and cartel theory. In each case I briefly outline the theory before discussing what it means that party networks are the source the coherence within the party. Responsible Party Government Responsible party government is not a descriptive theory of what parties are but is a normative argument for what parties should be. The oldest of the three theories discussed here, the theory of responsible party government has created a unique research agenda focused on dissecting why parties in the United States diverge from this standard, and more recently how they might be approaching it (David 1992; Rae 2007). According to the responsible party theory, parties should be consistent entities that outline clear platforms for voters and that once elected implement that platform (Committee on Political Parties 1950). These changes would make for a more robust democracy that could better respond to conflict within the electorate. Internal cohesion was deemed not just important, but necessary: “an effective system party requires. . . that these parties possess sufficient internal cohesion to carry out these programs” (Committee on Political Parties 1950, 1). In discussing cures for the lack of internal party unity or cohesion the Committee focused on changes to party organizations and leadership, for example outlining a party council chiefly composed of elected officials and officials nominated through party systems (Committee on Political Parties 1950, 43). The lack of party cohesion is seen chiefly as a problem of internal structures within parties and

89 not necessarily a problem of American institutions. In contrast, party cohesion as a result of party networks implies that unity cannot be solved by changes to the structures of political parties. Party organizations reflect dissension within the party network, although changes to organizational rules might better contain internal disagreement it cannot change the amount of internal disagreement. Unity within party networks is the only way to create unity within party organizations. Creating unity within a party network is significantly more difficult than changing the structure of party organizations. This might be naturally happening though. At the end of the 20th century researchers began to note the increasing polarization of political parties (Hetherington 2001; Levendusky 2009; Abramowitz and Saunders 2008). As a corollary of this polarization parties have become to be viewed as more responsible (Rae 2007). This is unexpected as polarization between parties has also be related to more cohesive ideologies within party elites and members (Carsey and Layman 2006). Identifying the source of party polarization remains an open question but the importance of party activists in this process lends to support to the role of party networks (Carsey and Layman 2006). Conditional Party Government In contrast to responsible party theories, the theory of conditional party government (CPG) does not argue for cohesive parties but argues that the actions of legislators is dependent on how cohesive parties are. CPG is a theory to explain the impact of political parties in Congress. Parties in Congress are at their strongest when political parties are internally homogeneous and clearly delineated from each other (Rohde 1991; Aldrich 1995; Aldrich and Battista 2002). When these conditions hold legislators are willing to give up their own individualized power to party leadership which uses that power to implement policy. In this case legislators believe that they gain more from working as a party than they do by working individually. Party leaders can use the power ceded to them to increase party discipline. The logical next question, what leads to party cohesion, has been under theorized

90 (Ladewig 2005; Fiorina 2002). There is general agreement that cohesion is the result of electoral incentives (Aldrich and Rohde 2001) but there is little to this beyond arguing that elections set the stage for what happens within Congress or other legislative assemblies. Parties are not constrained to what happens within legislative assemblies though. Party networks provide a way to bridge the gap between parties outside and parties inside the legislature while retaining a similar focus on party cohesion. The cohesion of party caucus is not dependent on the idiosyncrasies of elections but are the result of cohesion within the party network outside the legislature. This framework also provides a mechanism for feedback into the party outside the legislature. As coalitions within the legislature operate those outside the legislature respond, building or destroying relationships with other groups in the party network. The new network selects candidates in upcoming elections and the process continues. It also highlights the complicated role of legislators. In alignment with extended party networks, party networks are not tools of candidates but are important movers. Once within the legislature though the power dynamics change. Party networks main tool in influencing legislature is who they place within the legislature but once there, candidates are relatively unencumbered as individuals. This leads to a reversal of the relationship where parties within the legislature are a tool of elected officials. The varying place of political parties, both as leaders and followers, has made political parties hard to understand. Depending on the perspective political parties can be viewed as powerful or weak. Cartel Theory The final theory, parties as cartels, places cohesion in a very different position. Cartel theory argues that political parties in Congress can be viewed as ‘legislative cartels’, similar to economic cartels where members benefit by working together although there are incentives for individuals to defect. By working together the majority party is able to gain more legislative utility by only focusing on legislation that they agree on. This cartel is possible because

91 legislative leaders can use negative agenda control to prevent legislation that would split the majority party from coming to the floor (Cox and Mccubbins 1993; Cox and McCubbins 2005; Anzia and Jackman 2013). Not all members gain equally from the cartel, as more moderate members have more incentive to defect and vote with the minority party. To keep these moderate members in the cartel additional resources or other benefits might need to be funneled to them (Jenkins and Monroe 2012). In the context of cartel models, party cohesion is something to accentuate through legislative rules. Although as parties become more divided it might become more difficult to prevent this division from appearing, party leaders still have tools to try to limit it. Party cohesion outside the legislature will have less of an apparent effect on party cohesion inside the legislature (at least in the context of roll call votes), even if there is a real causal relationship, because party leaders work to make their party appear cohesive. Cohesive party networks might have a different type of effect in the cartel model. As mentioned above, moderate legislators often have to be given extra incentives to keep them within the party cartel (Jenkins and Monroe 2012). If a party network is cohesive, then the legislative team ought to be cohesive as well. This makes it less necessary to provide additional benefits to moderate members. In a more divided legislature, moderates can be rewarded by being placed in powerful positions. When the party is cohesive these positions will instead be taken by individuals that better reflect the preferences of the party as a whole. In addition, in a more cohesive legislature, leadership will not need to rely on negative agenda control as much as there will be fewer issues that can potentially divide the party.

***

In each of these three theories the fact that legislative cohesion or unity starts outside of the legislature alters our understanding of parties. For responsible party government it means that if a policymaker is concerned with creating more responsible parties they cannot

92 focus solely on organizational issues. Responsible parties require responsible party networks; creating responsible party networks is a much more complicated task. The role of party networks though does help to explain how polarization has led to more responsible parties within the legislature. Conditional party government is built around variation in legislative caucus cohesion. Using extended party networks to expand our understanding of parties beyond legislature actions helps fill in gaps in how parties operate. Party cohesion is important, not just within the legislature but outside the legislature. A cohesive party network will send a cohesive legislative caucus which will cede power to party leadership creating even stronger parties within the legislature. If those within the party network are happy with the actions of the legislature this will provide feedback to those outside the legislature which can lead to an even more cohesive network. In contrast if legislative caucuses do not do what members within the party network want this can also break up the party network and so change future party cohesion in the legislature. Finally, cartel theory leads to a different path. The logic of cartel theory is that legislative parties work to hide disagreement within the party. This will make it more difficult to find direct effects of party networks as the relationship should be weaker. Party caucuses come to power with a certain degree of true heterogeneity, which I argue is a function of the party network outside the legislature. Party leadership in the legislature work to hide that heterogeneity. All party leaders will do this, but those with more naturally homogeneous caucuses will have to exert less influence in the legislature to accomplish this. Therefore it will be harder to identify the relationship between outside cohesion and inside cohesion because party leaders work to hide the appearance of inside cohesion. For now I examine the first step, how party network cohesion can create homogeneity within the party caucus. This allows me to directly test the relationship between party networks and party caucuses. Given the assumptions of cartel theory though, this relationship

93 might be difficult to identify.

5.4 Testing the Effects of Cohesion

To test the relationship between parties outside and inside the legislature I need to first find good measures of each. In the following section I discuss how I approach measuring cohesion within party networks by using Exponential Random Graph Models. I then turn to measuring cohesion within the legislature and return to the data that I use in the previous chapter to measure ideology. Finally I discuss my modeling assumptions and control variables.

5.4.1 Measuring Cohesion Outside the Legislature

Measuring how cohesive party networks are is difficult as we can not directly observe why intense policy demanders in the network are tied together. In an ideal world we would be able to directly measure how groups perceive their interests are aligned with all other members in the party network, and a cohesive party would be one where interests are broadly aligned across all members of the party network. This is not feasible, but instead we can examine how groups are able to work together. In particular we can look at groups relationships with each other and how these relationships transfer to other members within the party network. These relationships have already been identified in Chapter 3 and we can use them again here to examine how relationships transfer across intense policy demanders. The idea of relationships transferring is referred to as transitivity in social networks. Transitivity is an important phenomenon in social networks as it has been found to be common across a wide range of applications (Wellman 1983). In a structural sense, transitivity focuses on how triads of nodes connect to each other (Cartwright and Harary 1956; Granovetter 1977). In addition a lack of transitivity can be the results of individual attributes that make it less likely for nodes to form widespread relationships (Hallinan and Kubitschek 1988).

94 In order to understand how relationships can identify agreement within the network it is useful to return to the example of Kansas. As discussed above, in response to the Kansas Republican Party moving to the right several moderate Republican groups have been formed, including the Traditional Republicans for Common Sense. They joined the Mainstream Coalition, and continued to support more moderate positions, including a moderate position on gun rights. Given this we would expect that the Mainstream Coalition would be connected to other Republican oriented organizations, but not connected to organizations that take a conservative stance on gun rights. In Figure 5.1 I show a subset of the 2013-2014 Kansas donation network including only groups connected to the Mainstream Coalition and the National Rifle Association. The National Rife Association is on the left side and is deeply embedded with other donors while the Mainstream Coalition is on the right. They are both connected to 20 other groups, mainly businesses and business organizations. They are not directly connected to each other though. The fact that they are working with over 20 groups but not working directly together indicates that there is disagreement between them over what it means to be a Republican Party. This is an example of a single open triad, where two groups are connected through other groups but not directly with each other. As more open triads appear within a network it indicates that there is a lot of disagreement within the party network. Given that transitivity indicates this idea of party cohesion the next question is how to measure it. The simplest is to count the number of closed triads. This is problematic though as closed triads could be the result of a tendency for triads to close or a tendency for groups with similar backgrounds to cluster (Goodreau, Kitts and Morris 2009). It is necessary to account for the other factors that might lead to the appearance of transitivity in order to prevent incorrect inferences by improperly measuring the amount of cohesion. To do this I use Exponential Random Graph Models (ERGMs) to estimate the generative structure of the model. ERGMs are statistical models of networks that loosen assumptions about the

95 Figure 5.1: Subset of Kansas Lower House Donation Network (2013-2014)

Note: Subset of the Kansas donation network including only the National Rifle Association and the Majority Coalition and those groups that are directly connected to them.

96 independence of nodes. In AppendixC I elaborate on the specifics of ERGMs and the terms that I use to control for other factors leading to the observed networks. Geometrically Weighted Edgewise Shared Partners (GWESP) is a commonly used term to capture transitivity within ERGM analysis (Hunter 2007; Goodreau, Kitts and Morris 2009; Grund and Densley 2015). GWESP is a measure of the amount of shared partners that two connected nodes have, down-weighting the marginal effect of additional shared partners. In AppendixC I expand upon this term to account for the fact that I am interested in triad closure just among co-partisans. To do this, instead of counting each shared partner as 1, I weight it based on the minimum percentage of contributions a party’s candidates. This allows me to create two terms one that is weighed by donations to Republican candidates, another that is weighed by donations to Democratic candidates (the donations are lagged). In Figure 5.2 I show the observed distribution of the estimated coefficients from all the networks in my data while accounting for uncertainty around the coefficient itself. To do this I took 1,000 draws from the distribution of the cohesion measure and estimated the density across all such draws for a party. The pointwise estimate ranges from -2.41 to 12.12. The larger values tend to have a wider confidence interval which leads to the long tails in the distribution.

5.4.2 Measuring Cohesion Inside the Legislature

To measure cohesion within the legislature I return to the legislative ideology data from Shor and McCarty(2011). Along with estimating the mean party preference they estimate party heterogeneity by calculating the standard deviation of legislator ideology within a party. This measure captures the idea of variation within the legislature caucus around preferences for policy solution. As the standard deviation within the party caucus increases there is increased heterogeneity within the party caucus and it is more likely for members of the party

97 Figure 5.2: Density of Party Network Cohesion

Republican Party 0.30 0.20 Density 0.10 0.00 −10 −5 0 5 10 15 20

Party Network Cohesion

Democratic Party 0.30 0.20 Density 0.10 0.00 −5 0 5 10 15

Party Network Cohesion

Note: Density of cohesion within the party networks. Since this is measured with error 1,000 draws were taken from the distribution of each coefficient and the density plots were estimated from this. to disagree over policy. As the standard deviation shrinks members of the caucus are more likely to have the same preferences. Figure 5.3 plots the density of the observed party standard deviation within each party. It ranges from 0.04 to 0.76. The distributions are relatively similar although the Republican Party has more outliers than the Democratic Party. The mean for the Democrats is 0.31 and for Republicans it is 0.27. For context, the range of the mean party ideology is -1.73 to 1.70.

5.4.3 Controls and Modeling Assumptions

I include a variety of measures that I expect to also be related to internal party cohesion Logged Number of Members in Caucus: As there are more members within a caucus

98 Figure 5.3: Density of Legislative Cohesion

Republican Party 4 3 2 Density 1 0

0.0 0.2 0.4 0.6 0.8

Caucus Ideology Standard Deviation

Democratic Party 4 3 2 Density 1 0

0.0 0.2 0.4 0.6 0.8

Caucus Ideology Standard Deviation

Note: Party heterogeneity as measured by the standard deviation within each chamber caucus. Data from Shor and McCarty(2011).

it is likely that there would be more heterogeneity within a caucus. Because of this I expect that the log of the number of caucus members will have have a positive coefficient. Increased caucus membership can be the result of two sources, the size of the legislature itself and the proportion of the legislature controlled by the party. For example the New Hampshire General Court (the lower house) has 400 members while the California State Senate has 40. Even if a party controls only 10% of the membership of the New Hampshire lower house they will have the same number of caucus members as the entire California State Senate. Proportion of Legislature: Along with the total membership the proportion I expect that the proportion of the legislature made up by the party will have a positive effect. When a party controls a substantial majority within the legislature there will be less pressure to

99 elect a unified legislative party as they do not need a unified party in order to pass legislation. In contrast minority parties will have more pressure to be ideologically homogeneous in order to stop any of their membership from defecting on close votes. Party Donation Strength: Party elites will work to exert some uniformity across candidates as well. To control for this I include the amount of donations that come from formal party organizations. These are similar to the measures using in the previous chapter but instead capture the resource strength of the formal party organizations. I expect the coefficient to be negative indicating that as more resources come from formal party organizations the less heterogeneity there is among the party caucus. Party Relationship Strength: I also include the relationship strength of formal party organizations. Similar to the resources measure I expect the coefficient here to be negative indicating that as formal party organizations are more embedded within the party they are more able to constrain the degree of heterogeneity within the party caucus. Electoral Competition: I expect that electoral competition will lead to a more hetero- geneous party. As candidates for office face more competition they will be pressured more to represent particular interests within their legislative districts. Unless there is uniformity across legislative districts this pressure will lead to increased variance within the party caucus. I use the same measure from Hinchliffe and Lee(2016) as I employed in the previous chapter. Party Dummies and Interactions: Finally I also include a dummy variable for the party that is 1 if Republican and 0 if Democrat. According to the logic of Grossmann and Hopkins(2016) I expect that the Republican Party will be more homogeneous and so the coefficient to be negative. In addition, I interaction the Republican dummy with all of the variables. This allows me to test for asymmetries in how the parties respond to these different variables. Again, given the work of Grossman and Dominguez(2009) I expect that the Republican Party heterogeneity to be less responsive in general to all of these variables. There are two more modeling assumptions to address. First, to account for unobserved

100 heterogeneity across states I include state random effects. Second, I do not use the point estimates of the coefficient of cohesion in the party network. These coefficients are estimated with error, and so that error needs to be incorporated when using them to model internal cohesion. I take 500 draws from the distribution of the coefficients, estimate the model on each of these sets, and use the methods suggested by Rubin(1987); King et al.(2001) to aggregate the results.

5.5 Results and Discussion

Table 5.1 displays the results. Again, the column on the left shows the coefficient for the main effects and the column on the right shows the coefficients for the interaction terms, interacting the variable with a dummy for Republican Party. To ease interpretation I have the coefficients, conditioned on party, in Table 5.2. Starting first with the coefficient on party network cohesion, given my theory I expect that as a party network becomes more cohesive that the caucus will become less heterogeneous. This implies a negative a coefficient for both parties. What I find is a significant coefficient of -0.006 for the Democratic Party, indicating that as the Democratic Party coalitions becomes more cohesive the party caucus becomes more homogeneous. The interaction term is significant, indicating that there is differential degrees of responsiveness between the Democratic and Republican Party. The coefficient for the Republican Party (in Table 5.2) is 0.0004 and is not significantly different from 0. The Republican caucus therefore does not have this similar relationship with the party coalition. This is surprising and goes against my expectations. The lack of relationship with the Republican Party provides some support to the asymmetry ideas proposed by Grossmann and Hopkins(2016). Before looking at the substantive effects of cohesion it is worth discussing the effects of the

101 controls. The only variable that had a significant effect across both parties was party resource strength. For Democrats the coefficient is -0.13, and for Republicans it is -0.09. This means that in both parties, as more resources that are provided by formal party organizations, the more homogeneous the party caucus is. Formal party organizations are important actors within the party network, they are a stabilizing force that can ensure coherence across the elected party caucus. The same relationship is not found for relationship strength. Formal party organizations, because of their privileged place, do need to rely on relationships with others in the party network in order to influence the behavior of legislators.

Table 5.1: Analysis of Party Cohesion

Main Effect Interaction Party Network -0.0006* 0.007* Cohesion (0.002 ) (0.003) Proportion of -0.02 0.07 Legislature (0.06 ) (0.06) Total Caucus 0.04* -0.08* Membership (Logged) ( 0.02) (0.01) Electoral -0.01 0.18* Competition (0.10) (0.07) Party Resource -0.13* 0.04 Strength (0.04 ) (0.04) Party Relationship -0.10 0.32* Strength (0.21 ) (0.27) State Senate 0.05* -0.07* (0.02) (0.02 ) Republican 0.17* (0.05) Intercept 0.19* (0.05) State Variance 0.08 N 1141 AIC -2336 Note: Model of party cohesion. The unit of analysis is the state year. State random effects are included. Asterisks indicate a p-value less than 0.05. See text for more details on variables and specification.

In contrast, there are not consistent effects for either the size or the proportion of the

102 caucus. The homogeneity of the party caucus is not related to having a larger proportion within a legislative chamber. There is a relationship found in for the size of the caucus, but only for Democrats and it is opposite than the hypothesized relationship. The Democratic caucus becomes more homogeneous the more members it has, indicating that there does not seem to be a difficulty of scale. Finally, there is no effect of party competition on the homogeneity of the legislative caucus.

Table 5.2: Model for Party Cohesion Conditioned on Party

Democrats Republicans Difference Party Network -0.006* 0.0004 X Cohesion (0.002 ) (0.002) Proportion of -0.02 0.05 Legislature (0.06) (0.05) Total Caucus 0.04* -0.03 X Membership (Logged) ( 0.02) (0.02) Electoral -0.01 0.17 X Competition (0.10) (0.09) Party Resource -0.13* -0.09* Strength (0.04 ) (0.03) Party Relationship -0.10 0.23 X Strength (0.23 ) (0.18) State Senate 0.05* -0.02 (0.02) (0.02) Note: Shows the coefficients conditioned on the party from the model in Table 5.2. The first column shows the effects for Democrats, the second the effects for Republicans. The third column shows if there is a significant difference based on parties at 0.05 p-value or less (the same information can be obtained by examining the significance of interaction terms in Table 5.2)

To examine the substantive effects I plot the predicted degree of legislative cohesion across the observed range of party network cohesion. All other covariates were held at the mean observed value in the Kansas Democratic legislative caucuses during this time period. It is clear that as the amount of network cohesion increases, the legislative cohesion also increases as well. At a network cohesion of -1.5, the legislative caucus has a predicted ideological standard deviation of 0.32, when network cohesion is 12 the standard deviation is 0.22.

103 Below the plot of the predicted values I show what the donation network could look like at that level of network cohesion. To do this I simulate the Kansas Lower House (2013-2014) network using the ERGM coefficients obtained from modeling it and then adjust the coefficient for Democratic MWESP (the amount of party cohesion). On the far left the cohesion is set to 0. In this case there is no clear cluster of Democratic donors, they exist on the edges of the extensive Republican network. This leads to a low degree of cohesion within the legislature. On the far right the cohesion is set to 10 and the Democratic network is distinct from the Republican network. This leads to a similarly bound Democratic caucus. The middle network is simulated using the observed level of network cohesion of 6.10. In this case there is some clustering of the Democratic donors. For the Democratic Party cohesion outside the legislature impacts the amount of party cohesion inside the legislature. As intense policy demanders within the Democratic Party coalition come to a better understanding of what it means to be a Democratic the network becomes more tightly bound and they are more successful at electing a similarly tightly bound set of legislators. Legislative cohesion can also be affected by other traits, including the strength of formal party organizations.

5.6 Conclusion

In this chapter I tested another theory of how party networks will shape the party in the legislature. I argued that party cohesion outside the legislature will create party cohesion within the legislature. As networks become more divided they will be more likely to send a similarly divided set of legislators into office. This should have important implications for the ability of the party to control legislatures as individual legislators will be unwilling to cede power unless it benefits them. I tested this by using Exponential Random Graph Models (ERGMs) to measure how

104 Figure 5.4: Predicted Legislative Cohesion 0.40 0.35 0.30 Legislative Cohesion Legislative 0.25 0.20

−2 0 2 4 6 8 10 12

Party Network Cohesion

(a) Low Network Cohe- (c) High Network Cohe- (b) Observed Cohesion sion sion

Note: Top plot shows the predicted degree of legislative cohesion across a range of values of party network cohesion. All other variables are held at the median observed in the Kansas legislature. The bottom three networks show simulated networks across the range of party network cohesion. On the left the the coefficient on the MWESP term was set to 0, the right it was set to 10, and in the middle it was at the observed level (6.10) for the Kansas State Lower House Democratic network in 2013-2014. cohesive party networks are outside the legislature. With ERGMs I was able to model the formation process of the network to see how important ‘friend-of-friend’ dynamics were in

105 their creation. Using the measure of cohesion from this I modeled the internal heterogeneity of the party caucus in the legislature. I found that the Democratic caucus responded in the expected way but that the Republican caucus did not. The differential results could be support for asymmetry across parties, but also could be the result of modeling decisions. In particular, the Republican coalition might be more reliant on individual level donors than group level donors and by including only group level donors I may have missed important information about dissension within the Republican caucus. Future work on this might be able to better test this theory though small N case studies and examining how party networks have changed over time within a state or other political entity. Though this it might be possible to get better leverage on the problem while holding many confounders constant. It also might be possible to better measure cohesion within a party network by examining a smaller subset of cases. Work on large cross-sectional data like the tests presented here might require more attention to measuring party networks in a way that ensures comparability across states. This is a difficult task given the different institutional and electoral rules that might lead to different network structures. This was a first cut at attempting to do this and hopefully future work can build off of it.

106 Chapter 6 | Conclusion

6.1 Summary

In this dissertation I explored how groups outside of legislatures affect what happens within legislatures. I did this through the theoretical perspective of parties as networks, which conceptualizes parties as being diverse coalitions of interests that work together to elect candidates. Starting from this position I extended this theoretical framework to explain how party coalitions translate into party caucuses. At its simplest my dissertation argues that if parties are networks, then their network structure should have implications for political parties. I briefly summarize my theory and results here before discussing other aspects of the dissertation. In Chapter 4 I argued that because parties are networks, relationships within the party network is important for explaining the positions that a party takes. As a group has more relationships within a network they will be able to pull the preferences of that party closer to their own preferences. This is true even when controlling for the resources that an intense policy demander provides to partisan candidates. To test this I developed measures of different industries within the donation networks: labor unions, business interests and ideological interests. I measured how central each of these industries was to each party network and

107 used this to predict the mean ideology of the party’s legislative caucus. I found that business and labor union relationships pushed each party to be more conservative (for business) or more liberal (for unions). This effect held even when controlling for the amount of donations provided by these industries and for the preferences of citizens. In Chapter 5 I examined what happens when party networks are unable to find agreement. There are a lot of reasons for party networks to be contentious places, and I argued that this contention within the party network ought to leak into the legislature. That as a party network becomes more cohesive outside the legislature the party caucus will become more cohesive inside the legislature. Testing this was more difficult as it required accurate measures of cohesion within the party network. To do this I used Exponential Random Graph Models (ERGMs) which allowed me to measure network cohesion while controlling for other factors. To measure cohesion in the caucus I used the standard deviation of the legislator’s ideology score within the caucus. Using these I found that Democratic caucuses do respond in the expected way but that there is no strong relationship for the Republican caucus. My tests of network effects on party cohesion and party ideology show how party networks impact the political process. This demonstrates that when researchers talk about political parties as networks we must take this seriously and examine the ramifications of network structures. In the remaining of this chapter I discuss the broader implications of my dissertation, limitations to it and finally future work. I argue that this dissertation informs how we understand inequalities within democracy and what theories of parties need to be able to grapple with. In the limitations section I discuss what my focus on donations might miss and remaining questions of causations. Finally, for future work I discuss ways to examine more closely the creation of ties between intense policy demanders as well as the relationship with social movement activist networks.

108 6.2 Broader Implications

This dissertation focuses on an important aspect of the American political process: political parties. The findings, that relationships within parties are critical for understanding the actions of parties , have important ramifications for how we understand democracy. Rela- tionships cannot be easily regulated yet will still often be the function of the resources of those who have are involved. This indicates that there is another barrier against the equal participation of all groups. Those without relationships will have trouble participating in the political process, and to the extent that those relationships reflect privilege, those without relationships will be the already underprivileged and unrepresented. In addition, my research also illuminates some of the more complicated questions about political parties in the United States. Political parties touch on many aspects of political life yet scholarship has often studied one aspect of political parties independent of others. My research shows directly the relationships that exist between aspects of parties and how we must study parties holistically.

6.2.1 Representation and Democracy

One of the questions at the heart of political science is how well does the policy implemented by government reflect the preferences of citizens. This is a fundamental issue as a core assumption of democratic governance is that the government ought to reflect the will of the people. Yet within the American context there are many instances where policies have been out of line with the preferences of the majority. Recently, attention has been focused on how policy does not represent all equally, but tends to represent the most affluent the best (Gilens 2012). Political parties have been shown to follow similar dynamics, where the poor are often underrepresented within the party (Rigby and Wright 2013). These questions are especially important given growing income inequality (Piketty 2014).

109 While many researchers have argued that there is a gulf between how policy responds to the rich versus the poor, it has often been difficult to find direct evidence of money being a corrupting influence on the political process (Baumgartner et al. 2009). The explanation for differential responsiveness is instead placed on the fact that affluence correlates with political activity, that the rich participate more and have more means to amplify their voice (Schlozman, Verba and Brady 2012). The findings here support this, showing again how the current design of the political process excludes un- or under-organized interests. Political parties are networks of intense policy demanders, that build relationships between each other and work with each other to implement policies. As shown, the relationships within groups are critical and influence the aggregate positions of political parties. Unorganized interests though cannot take part in this process. They cannot build relationships and so will not be able to influence policy in the same way that organized interests can. When unorganized interests do organize they might still have difficulty participating in political parties. According to a more traditional understanding of political parties, if a new set of interests appears and is unrepresented then candidates will move to represent them in order to gain their votes. From this view, candidates are relatively free agents that can change preferences when new interests appear. But, political parties are not just a loose set of preferences. They are structured organizations. For a new interest to gain something in the political process they must build relationships within a political party. If the intense policy demanders within the political party perceive the new entrant as incompatible with their own preferences then they will work to prevent them from entering the party, perhaps pushing candidates to take stances against this new entrant. If these groups are powerful within the party they can successfully freeze out this new interest and so keep them as a minor interest in the political system as a whole. This is of course not the only potential reaction to a new entrant into the political party. If there is disagreement within the party over these new preferences then the new entrant

110 may split the party. As I show in Chapter 5, when political party networks are divided the party caucus in the legislature is also divided. This threatens to hamper the parties ability to legislate. If party elites are unsure how others within the party will respond to the new entrant then they will likely tend towards a conservative approach to prevent the party from splitting. The final option for political parties facing new entrants is for them to accept them quickly and easily into the party network. This requires the current participants within the party to view the new entrant as being acceptable to themselves and other important policy demanders within the party network. This will enhance the upper class bias within the political process. The current party reflects a history of mobilization and relationships and so is likely to already have an upper class bias. They are more likely to accept new interests that are not going to upset this position. Parties being unwilling to respond to new, especially controversial interests, explains why social movements and contentious politics continue to be a characteristic of established democracies. The formal institutions within the United States are open to new interests, yet the way that politics unfolds is much more resistant to change. Groups within the party network will be hesitant to build relationships with new interests as it risks upsetting the apple cart. New entrants will have to therefore find ways outside of the party process to bring about change and so will instead participate in contentious politics. In sum, political parties, although coming out of group conflict, do not fit the idea of pluralism as expressed by Dahl(1961) but instead are situated in the neopluralist view (Lowery and Gray 2004; McFarland 2007). Although these groups can clash at times they have a preference for maintaining the status quo of the party network as is. Those intense policy demanders that are central to the party network will be able to use this preference for the status quo to ensure that outsiders have limited access. This has the result of amplifying power imbalances that might already exist.

111 6.2.2 Connecting Parties

Along with the role of political parties in a democracy, this dissertation has implications for how we should study political parties. Scholarship on political parties in the United States has often focused on parties in one domain or another. In particular it has followed the classic tripartite division from Key(1949): party in the electorate, party in government, and party as organization. This has led to attempts to examine political parties in one place that are independent of how political parties act in other areas. Parties within the legislature are often analyzed independently of political parties outside the legislature. The evidence presented in this dissertation though shows how the structure of the party network outside the legislature is important for understanding what political parties do within the legislature. Political parties are complex and so researchers have a tendency to want to simplify how they study them to make the problem tractable, but this leads to theories of political parties that do not necessarily explain the world well. Aldrich(1995) argued that political parties are teams of politicians bent on gaining power. This is perhaps the most complete view of parties, connecting elections to what happens within Congress. But outside of elected bodies his theories did not explain much. His basic argument is that politicians are atomized individuals interested in power who create the structure of a party to help them in their quest. Left unanswered is how these individualistic actors come to agreement about what to do when they are not in the legislature. If, as Aldrich argues, parties are gatekeepers, who exactly operates the gate? What candidates are deemed worthy of getting the support of the party. The answers to these questions are important. As shown in this dissertation, it is not possible to disconnect the actions of the party outside the legislature from the party inside the legislature, they both inform each other. Even the network approach to parties is not perfect. In this dissertation I have explained

112 how the party network affects the party caucus. In particular that the structure of the party network leaves an imprint on the party caucus and so the same sort of divisions and power concentrations within the network are present in the legislature. There is likely though a feedback mechanism. Elected party leaders also participate in the network and so they will work to shape the network in a way that supports their interests as well. This remains unspecified as of now, and will be important in a fuller understanding of political parties. The electorate is also passive in the current discussion, and it remains to specify how much they should be expected to shape the party network or if they only respond to it. In this dissertation I have shown how parties outside the legislature play a direct role in the legislature. This is an important point as it highlights why scholarship cannot separate aspects of the party when theorizing about them. Although I did not answer all the necessary questions of political parties, I did highlight the need for a more complete understanding of political parties.

6.3 Limitations

As with all research this dissertation has limitations. Here I discuss two of the most important ones and what they mean for the findings. Contributions, although very useful, do not capture all possible relationships that exist between organizations. This means that some intense policy demanders might be excluded from this analysis that are part of the party network and some relationships might also be missing. In addition, although I have formulated my theory as one where networks cause changes in the party my tests are not causally identified. Causal identification is difficult in this context and so I discuss the importance of causal identification and what the lack of it means for my dissertation.

113 6.3.1 What Contributions Miss

The networks that I developed were based around campaign donation data. This is a dominant way to examine party networks but is not the only plausible way. Previous work has examined the exchange of mailing lists (Koger, Masket and Noel 2009), the movement of staff (Skinner, Masket and Dulio 2013) and consultants (Nyhan and Montgomery 2015). All of these sources of relationships, including donation data, are biased in a similar way. They bias the examination of party networks towards more professionalized and wealthy interests. To be able to donate, employ staff or consultants and buy mailing lists an organization needs a certain degree of professionalization. Donations require perhaps the least amount of professionalization as it generally only necessitates an interest being a formalized group and having some pooled money to donate. This excludes barely organized groups, and those groups that opt to not donate for candidates because of ideological reasons. Groups that are barely organized are unlikely to be deeply involved in the party network. To some extent they are excluded from participating and so measuring their lack of involvement is not problematic. Exclusion of groups that have decided to not participate through donations, but do participate in other ways, is potentially more concerning. Legal frameworks can make it hard for groups to directly participate. Groups have developed common ways to navigate legal hurdles, but the way they navigate them makes it harder to identify the groups and often requires a group to be more professionalized (La Raja 2013). Future work on party networks needs to approach new ways to document the network, and embrace multiplex networks that incorporate edges from different types of interactions. One possible new and so far unused source of network information could be gathered though examining relationships between groups on social media. Social media connections can provide a way of examining how relationships transfer beyond donation or other campaign related actions. In addition social media data lowers the bar for entry for groups, making it

114 even easier for them to be identified as participants in the network. This of course comes at a cost. Social media is a relatively costless activity and so the extent that these relationships are meaningful might be questioned. In addition, the focus on groups might miss the role of important and influential individuals. This might be especially true within the Republican Party where individuals like Charles and David Koch have built their own parallel network of political influence (Mayer 2016). Their importance might be identified through the groups that they participate in, but they are not directly involved in the data analysis. This might explain the lack of findings for the Republican Party in Chapter 5. The donation data employed in this analysis does not capture all of the relationships between groups, and might exclude some groups from the analysis that are active participants. Although it is likely that it only excludes a small percentage of groups, future work might be able to overcome this problem by using multiple data methods to identify network structure and group participation.

6.3.2 Party Networks: Cause or Effect?

The theory of party networks supposes that party networks are the main causal force in party politics. The party coalition comes together and works to elect candidates, and so those candidates reflect the structure of the party network. It is possible though that the party responds to the actions of candidates. Candidates have a set of positions and the party structure reflects the dominant positions held by the party’s candidates. Similarly, it is possible that as candidates and elected officials become more contentious with each other the party network also starts to break down. Delineating between these two causal directions is outside the scope of this dissertation. Other works have demonstrated the power of the party network to push candidates into

115 positions that they would not normally want to take (Masket 2007; Masket and Shor 2015; Masket 2016). For example, Masket(2007) looks at how the shock of cross-filing in California changed the ability of the party networks to structure partisanship within the state legislature. During cross-filing legislative politics became significantly less polarized and was harder to explain through the lens of parties. The end of cross-filing meant a return to very partisan division within the California legislature. The effect of the sudden switch to cross-filing provides compelling evidence of a causal relationship between party networks and partisanship in the legislature. Now that party networks have been developed and validated for a wider range of time across all states it might be possible to use other similar shocks to the system to get leverage over causal identification. Changes to state laws during this time period might be one source of leverage. In 2008 Connecticut implemented voluntary public financing campaigns and by 2012, 77% of candidates were participating in it. With the implementation of public financing a different set of candidates began running for office and those candidates that remained perceived a decreasing influence of special interests (Cha and Rapoport 2013). This sort of change should lead the party network to having significantly less influence on the legislature. Future work can investigate this and other cases of variation to better test the question of causality.

6.4 Future Work

I have already discussed a few ways that future work can continue examining party networks: looking at variation in state laws or examining other types of relationships. There are important theoretical extensions that I hope to work on in the future years. First, I want to examine how party networks form and threats to the party network. Importantly I want to look at how social movements might destabilize party networks. Second, I am interested

116 in examining how networks of movement activists, both involved and not involved in party politics, overlap with party networks

6.4.1 The Creation and Destruction of Party Networks

As of now I have taken party networks to be fixed, something that exists and causes variation in party caucuses. Although I have speculated on why there might be variation in party networks I have not specified clearly how these party networks evolve and grow over time. In particular, I am interested in the centrifugal forces that might lead a party network to be strained. We have seen many examples in US history of heightened division within party networks. For example the 1960s and 70s saw the anti-Vietnam War movement and the Civil Rights Movement lead to division within the Democratic Party. In the 2010s the Tea Party both energized and divided parts of the Republican Party coalition. Explaining why these movements caused coalitions to shift and sometimes fight will help better our understanding the outcomes of social movements. Some recent scholarship has highlighted how social movements can participate in American political parties, acting as an anchor to the party (Schlozman 2015). I look to take this further than examining just how a movement can become central to the party, but also examine how movements can change the entire structure of the party network, shifting relationships and alliances. I expect that social movements, even when they do not fully enter a party, can lead a party to divide as elites in the party decide how to potentially respond to the movement. This can also illuminate how the structure of party networks evolve overtime. A movement entering a party network is an extreme of what should be a regular occurrence: new interests entering the political system and deciding to participate in a party. How new interests are welcomed to a party network is critical for questions of representation and democracy. It is necessary to theorize when a party network is willing to accept these new interests: if it is a

117 function of potential resources, division among voters, or something else.

6.4.2 Activist Networks Versus Party Networks

Along with studying how new movements enter into the party network I am interested in studying how social movement and activist networks overlap with party networks. Heaney and Rojas(2015) shows the important overlap among anti-war activists and the Democratic Party, arguing that dual affiliation between participants in both was important for understanding the trajectory of the anti-war movement after the ascendancy of the Democratic Party. Left unanswered is how activists that are not necessarily aligned with a party build coalitions that often intersect with party networks. This is especially important given the increased presence of activists that look to build networks that are outside of the two major political parties (Heyward 2017). The question is first how activist networks and party networks compare. Are they similar in structure, and just vary in the groups that participate in each. Following this is how do they overlap with each other and how well do activist networks do at staying distinct from party networks. How these networks relate is important to understand how movement activists and other non-partisan organizations relate to the party network.

***

Political parties have deep impacts on American democracy, and so understanding them is critical for how we can understand democracy. Like many social phenomenon though, parties are complex and hard to define concepts. In this dissertation I have grappled with how the complicated structure of party coalitions impacts the very real actions of party caucuses in the legislature. I provide what I hope is compelling evidence that parties matter and how they matter.

118 Future work will hopefully be able to continue this line of research answering some of the questions posed here, as well as new ones.

119 Appendix

120 A Technical Details

In this appendix I describe in more detail the process of backboning used in this chapter. I start first by explaining the technical procedure, before moving on to showing several randomly selected examples. The general backboning algorithm employed here is a five step process and can be found detailed in Algorithm 1. In brief, I start with the data on group donations to candidates, model the amount donated from each group to each candidate, and then use that model to make a large number of simulated datasets. The modeling process, Step 2, requires additional detail. The data used here, donation data, is strictly 0 or greater requiring a GLM that fits count data. In addition the large number of zeros (where no donation took place) necessitates either a truncated or zero-inflated model. After testing multiple functional forms and data transformation I selected a hurdle model with the count process modeled using a Poisson distribution. In addition, the donations were transformed into $100 increments (rounded to the nearest hundred) and the covariates were logged (after being shifted by 1/2 the smallest non-zero value). This was found to be robust across all datasets and also fit the data well.

121 Algorithm 1: Backboning Process

1. Let G be the number of donors, C the number of candidates, and δcg the donation from PC donor d to candidate c. Then c=1 δcg is the total number of donations made by donor PG g and g=1 δcg the total received by candidate c. Finally ∆ is a C × G matrix made up of all Dcg. 2. Estimate a model of each candidate-donor relationship:

C G C G X X  X  X  δcg = β1 δcg + β2 δcg + β3 δcg δcg +  c=1 g=1 c=1 g=1

3. For each candidate-donor pair take a 1,000 draws from the estimate model to generate ˆ 1,000 matrices of estimate donations—Ai

ˆ ˆt ˆ 4. Take the crossproduct of all simulated matrices and of the real matrix. So ∆i · ∆i = Ai t and ∆i · ∆i = Ai 5. The backbone B is 1 for each candidate-donor where the observed donation is greater  than the α-centile of the simulated edge weights. Formally ∀bcg ∈ B, bg1g2 = I ag1g2 >  ˆ quantileα(ˆag1g2 ) where ag1g2 ∈ A and aˆg1g2i ∈ Ai

To demonstrate this process I have selected 9 State-House-Full-Cycle at random. Figure A.1 plots the real donations from a sample of states. Each observation is a group-candidate pair. As discussed above, in the majority of cases this is a 0 as there was no donation made from the group to the candidate. To model the data, I used the data transforms discussed above along with a hurdle model with a Poisson count process. Figure A.2 displays the estimated coefficients from the 9 models for each state along with their 95% confidence intervals. Although the coefficients vary across the units, there are broad similarities. The top row shows the coefficient from the count proportion of the model, the bottom row shows the estimated coefficient from the logit model predicting if a donation takes place. The intercepts in the logit model are all negative, reflecting the fact that any given group

122 Figure A.1: Real Donations from a Sample of States

Note: Donation data from 9 sample states. Each observation is group-candidate pair, showing the amount donated from a group to a candidate. This shows how the majority of donations are zero and that there is a large amount of variation in the amount donated. donating to any given candidate is relatively unlikely. The interaction terms in the count model are also all significant, indicating that the coefficients for the propensity of a group to donate, and the propensity of a candidate to receive, depend on each other. Figure A.3 plots a set of predictions for each State-House-Full-Cycle against the real donations. Although

123 many of the dots fall on the 45◦ line, there are also a substantial proportion where either no donation was predicted and one was made or where no donation was made and one was predicted.

Figure A.2: Estimated Coefficients for Sample Models

Intercept Log(Group Donations) Log(Candidate Donations) Interaction

NV 2001 2002 House ● ● ● ● NY 2009 2010 House ● ● ● ● WI 2001 2002 House ● ● ● ● VA 2014 2015 House ● ● ● ● GA 2009 2010 Senate ● ● ● ● VT 2009 2010 House ● ● ● ● Count Model NH 2003 2004 Senate ● ● ● ● MA 2003 2004 House ● ● ● ● ID 2009 2010 House ● ● ● ●

−2 0 2 4 6 8 −2.0 −1.0 0.0 1.0 −2.0 −1.0 0.0 1.0 −1.0 0.0 0.5 1.0

Intercept Log(Group Donations) Log(Candidate Donations) Interaction

NV 2001 2002 House ● ● ● ● NY 2009 2010 House ● ● ● ● WI 2001 2002 House ● ● ● ● VA 2014 2015 House ● ● ● ● GA 2009 2010 Senate ● ● ● ● VT 2009 2010 House ● ● ● ● Logit Model NH 2003 2004 Senate ● ● ● ● MA 2003 2004 House ● ● ● ● ID 2009 2010 House ● ● ● ●

−11 −9 −7 −5 −1.0 0.0 0.5 1.0 −1.0 0.0 0.5 1.0 −1.0 0.0 0.5 1.0

Note: This shows the estimated coefficients across the different datasets. The top row are the coefficients in the count model, the bottom the coefficients in the logit model. The lines are 95% confidence intervals around the coefficient.

The next step in the process is to draw 1,000 simulated values from each of these models. For each of these simulated sets of donations the crossproduct is taken. This creates an empirical distribution of the expected weight placed on a tie between two groups. Using this empirical distribution I include only ties between groups where the real weight is at or above the 97.5% percentile of the empirical distribution. This significantly reduces the total number of edges in any given network. Figure A.4 plots the total number of edges in each network versus the number of edges identified in the

124 Figure A.3: Predicted versus Real Donations from a Sample of States

Note: Plots of the predicted versus real donation amounts. backbone methodology. Note that the scale of the Y axis, the number of initial edges, is orders of magnitude larger than the scale of the X axis, the number of edges identified by the backbone.

125 Figure A.4: Number of Edges Pre and Post Backboning

● ●● ● ●● ●●● ●●●● ●●●● ● ● ●●● ●● ● ●●●●●●●●● ● ● ●●●● ●●●●●●●● ● ● ●●●●●●●● ● ●●●●●●●●●● ●●● ●●●●●●●●●●●● 1e+06 ● ● ●●● ●●●●● ●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●● ● ●●● ● ●●●●●●●●●●●●●● ●●● ● ●●●●●●●●●●●● ●●●● ● ●●●●●●●●●●●●●●●●●●● ●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ● ●●●●●●●●●●●● ● ●●●●●●●●●●● ●●●●●●●●●●● ● ●●●●●●●● All Edges ●● ● ● ●●●●●●●● ●●●●● ●●●●●● ● ●● ●●●●● 1e+04 ●● ● ● ● ●●● ● ●● ● ●● ●●●● ● ● ● ●●●●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ● 1e+02 1e+01 1e+03 1e+05

Backboned Edges

Note: The number of edges without bacboning compared to the number of edges backboned. Each observation is a dataset. This shows the reduction of edges caused by the backboning process and that the reduction is proportional across the datasets.

126 B Controlling for Donation Limits

Because the networks that I devise are based on donation data there might be concern that variation in state donation laws act as a confounding variable. This is plausible given the finding of Barber(2016) that donation laws do have an impact of the positions that legislators take. To test the role of donation rules I include four more variables in the data: if there is unlimited individual donations, the limit for individual donations, if there is limit for group donations, and the limit for group donations. These variables are from Barber(2016) and follow their specification. In particular I interact the log of the donation limit with a dummy variable indicating if there are unlimited donations. The estimated coefficients are in B.1. None of the donation limit variables are significant predictors of ideology. The majority of the remaining coefficients are similar to those presented in the main body of the text. The business coefficient on relationship strength is no longer significant, because of the change in sample size, but the remaining network variables are significant and in the same direction. I also replicate the process of checking predictive RMSE as I did in the main text. The plot is displayed in Figure B.1, and shows the same pattern as found with the full model. The model including both the donation and relationship information outperforms all other models providing support for my argument that relationships have an impact of the positions parties take.

127 Table B.1: Analysis Party Caucus Ideology with Donation Limits

Main Effect Interaction Term Business 0.73 0.34 (0.40) (0.43) Labor -0.14* -0.36 (0.05) (0.19) Ideology 0.01 -0.10 Relationship (0.24) (0.37) Business 0.93 -4.19* (0.56) (0.65) Labor -0.80* -1.92* (0.11) (0.30)

Donations Ideology 1.85* -0.63 (0.33) (0.49) Senate -0.02 0.01 (0.02) (0.02) Proportion of Memb 0.48* -0.54* in State Leg (0.10) (0.16) Log Number of -0.08* 0.12* Groups (0.02) (0.02) Competition 0.83* -0.54* (0.24) (0.16) Proportion Urban -0.96* 0.58* (0.23) (0.11) Citizen Liberalism 0.18 0.19 (0.11) (0.10) Unlimited Individual 0.32 0.01 (0.28) (0.17) Limited Individual 0.03 -0.01 × Limit Logged (0.03) (0.17) Unlimited PAC -0.25 -0.32 (0.22) (0.16) Limited PAC -0.02 -0.02 × Limit Logged (0.02) (0.02) Republican 0.76* (0.16) Intercept 0.27 (0.24) State σ2 0.04 (0.21) Residual σ2 0.05 (0.21) N 1562 Note: Model of party ideology. The unit of analysis128 is the state year. State random effects are included. Asterisks indicate a p-value less than 0.05. The inclusion of donation data does not have any substantial effects. Figure B.1: Repeated 10-Fold Predicted RMSE with Donation Limits 0.065

● 0.060

Base Model

0.055 Relationship Model Root Mean Sqared Error 0.050 Donation Model Full Model 0.045

Model

Note: The distribution of the average residual mean squared error from repeated 10-fold cross validations. The full model continues to out preform the limited models even with the inclusion of the donation limit data.

129 C Exponential Random Graph Models

The main difficulty in modeling networks is that we cannot assume independence among units within the network. Although in many social science cases the assumption of independence among units is tentative, the entire purpose of network analysis is to explore the complex dependencies between units and so there cannot be broad assumptions of independences. Exponential Random Graph Models (ERGMs) accommodate complex dependencies by modeling the network as a single realization from a multidimensional distribution. It is possible to include explanatory variables that reflect both individual (nodal) characteristics as well as characteristics about the network as a whole (which will be the focus of the idea of cohesion). The ability to model network level characteristics is important and unique to ERGM methods (Cranmer et al. 2017). ERGMs have a similar parametric framework as traditional regression models, with a set

of terms that are hypothesized to explain the network formation or structure. If Ym is the network, then Pk exp(− j=1 Γjmθj) P (Ym) = PM Pk m=1 exp(− j=1 Γjmθj)

Γmj is equivalent to an independent variable that measures some characteristic of the network. For example this can be an attribute of nodes that might give rise to edge formation, like if a node is a labor union or not. It can also be a network level characteristic that reflects the structure of the network as a whole, such as if nodes often form triangles (where three

nodes are all connected to each other). θj is akin to a coefficient showing how important a term is. The denominator acts as a normalizing constant across the range of potential networks. The estimation of ERGMs is relatively complicated given the possible number of networks. For more detail on ERGMs and their estimation see Cranmer and Desmarais (2010). I now briefly describe the nodal terms that I include to control for other reasons groups

130 might be connected. After discussing the control variables, I discuss Maximally Weighted Edgewise Shared Partners (MWESP) and how I will use it to measure cohesion within a party network.

C.1 Nodal Terms

• Group Type: This is a factor variable using a modification of the group type from the NIMSP data. I recode this data so that there are at most 10 categories: Businesses, Candidates, Democratic Party, Republican Party, Third Party (rare) Ideology/Single Issue, Labor Unions, Lawyers and Lobbyists, and Other. The Business category includes multiple categories coded by NIMSP. The group data is included as a factor variables to account for some group types being more or less likely to have connections than other groups.

• Group Homophily: Because groups of the same type are likely to have shared interests they are also potentially more likely to connect with each other than with groups outside of their type. This phenomenon is known as homophily (McPherson, Smith-Lovin and Cook 2001). To account for this I included a differential homophily term that allows for the degree of homophily to be unique for each type. I do this as there are also differences across types. Although labor unions might be likely to be tied together, lawyers and lobbyists are a much more diverse set of groups and so are not necessarily likely to be tied together.

• Logged Total Donations (lagged): This is a measure of how much a group has donated in the previous electoral cycle. It accounts for the fact that groups that were very active in previous election might be more likely to have connections with groups in the current election.

131 • Logged Absolute Difference in Total Donations (lagged): Edge formation might not only be responsive to the amount of donations a group makes but also to how similar two groups donation patterns are. In particular groups that donate at similar levels might be more likely to be connected. This is accounted for by this term which is the log of the absolute difference in donations between two nodes.

• Percentage of Donations to Democratic Candidates (lagged): Calculated:

P Dem Donations P Dem Donations + P Rep Donations + P Other Donations

This term controls for difference in Democratic versus Republican leaning groups. In some states where one party dominates, nodes that participate in that party might be more likely to have ties. A positive value for this indicates that as a group donates more to Democratic candidates the more likely they are to have edges, negative values indicate the opposite.

• Absolute Difference between Democratic Donation Percentage (lagged): This is similar to the logged absolute difference in total donations but instead ac- counts for the similarity in partisan leaning of donations. A significant negative coefficient indicate that groups that donate at the same level tend to be tied together, so Democratic leaning groups are more liked to be tied to other Democratic leaning groups

C.2 Geometrically Edgewise Shared Partners and Extensions

ERGM analysis has developed several set of terms that account for network dependencies. One particular term, GWESP (Geometrically Weighted Edgewise Shared Partners) account for the tendency of nodes to form if they have shared partners. When there is a single shared

132 partner they create a triangle. Triangles are often expected in networks if friends of friends (edgewise shared partners) are likely to also be friends. It is possible though for two friends to have multiple shared partners (forming multiple triangles). Figure C.1 shows an example of what happens when two edgewise shared partners exist. Here 1 and 2 are both connected to 3 and 4 and so we might expect this to increase the probability of an edge between 1 and 2 forming. Theoretically we might expect that the additional edgewise shared partner matters, but not as much as the first edgewise shared partner (that the adding more edgewise shared partner has a decreasing marginal effect). The extent that additional edgewise shared partners matter is modeled, which is where GWESP receives the name geometrically weighed. The formula for GWESP is:

n−2  i θt X  −θt  v(y, θt) = e 1 − 1 − e EPi(y) i=1

θt controls the weighting and EPi(y) is a count of the number of instances of i edgewise

1 shared partners. For the network in Figure C.1 EP1(y) = 4 and EP2(y) = 1. If θt = 0 then

Pn−2 only the first edgewise partner is counted and the formula becomes simply i=1 EPi(y). In

contrast at θt → ∞ then each additional edgewise partner counts as 1 and this is equivalent of adding up all the edgewise shared partners within the network. GWESP accounts for a common feature in social networks, that friends of friends are also likely to be friends. This might also apply to donor networks, where a donor that works with another donor is also likely to work with the donors associated with that donor. For this to be the case though it is necessary for the all the donors in that set to have a shared idea of what they are working for. This is exactly that idea of cohesion that I am interested in here, where groups within the network generally cohere in how they see the team they are trying to get elected. A large coefficient on a GWESP term indicates that this sort of

1Note that 1-4, 1-3, 2-3, and 2-4 all have a single edgewise shared partner

133 Figure C.1: Examples of Edgewise Shared Partners

1 2

3

4 Note: Shows an example of edgewise shared partners between 1 and 2. 1 and 2 are connected and also both connected to 3 and 4 so they have 2 edgewise shared partners. If 1 and 2 were not connected then they would have 2 dyad shared partners.

friend-of-friend relationship is very important in explaining how the network formed, and so that cohesion likely exists within the network. This is why I use an extension of GWESP to capture cohesion within the party network.

C.3 Maximally Weighted Edgewise Shared Partner

The GWESP term is limited in that it assumes that edgewise shared partners across all nodes within a network are equivalent, and all that matters is the number of edgewise shared partners. The networks examined here though contain nodes that are not necessarily similar. Take for example four groups, start by assuming that all four donate only to Democratic candidates, this is captured in Figure C.2a. In this case we would expect that the shared partners between 1 and 2 would increase the likelihood of a connection between them. Now assume that node 2 is replaced with a group that donates only to Democrats (Figure C.2b). The probability of an edge forming between 1 and 2 given a similar configuration of edgewise shared partners seems unlikely to be the same as in the previous example. Put differently it is unlikely that a friend of a friend is as meaningful in this context.

134 Figure C.2: MWESP Examples

(a) All Democratic Donors (b) One Republican Donor

1 2 1 2

3 3

4 4 Note: Figures show two examples of nodes with different MWESP value. The set of nodes on the left all donate to Democratic candidates and so the MWESP value is the same as the GWESP value. The set of nodes on the right has one that donates solely to Republicans and so has a MWESP value of 0.

GWESP terms, as described above, assume uniformity across all nodes within the network and so cannot account for these differences in nodes. To account for the way that groups donate to parties I extend the GWESP term so that the similarity of partisan donation strategies can be included. I do this in the case where the weighting term (θ) is equal to 0 and so only the initial edgewise shared partner counts. It is possible to extend this to allow for other values of θ but for simplicity I start here.

To extend GWESP start by redefining (sp)lj the shared partner statistics between two nodes (l and j). Normally this counts the number of shared partners between two nodes and so is equal to an integer value greater than or equal to 0 (Goodreau 2007). This can be modified to a maximally weighted shared partners where

mwsplj = max [((ylj)(ylk) · min(nj, nl, nk))] k∈V

where nj, ni, and nk are the value of nodes and the maximum is taken across all nodes in the network. In the networks presented here the values of node range from 0 to 1 (no donations

135 to a party and all donations to a party). The formula for MWESP is:

W X 0 r(y) = wEPi(y) w=0

0 . Where EPw is similar to EPi(Y ) in the GWESP equation but now counts the instances of maximally weighted edgewise shared partners with weight w instead of the instances with i

edgewise shared partners. This is summed across the set W which contains all the mwsp values observed.

C.4 Estimation

As discussed above, the estimation of ERGMs can be tricky. Unlike traditional regression models, misspecification can lead to degeneracy where the underlying estimation schema tends towards networks that are either compete or empty (Schweinberger 2011; Hunter, Krivitsky and Schweinberger 2012). Estimating ERGMs across a range of networks requires specification that is robust across these networks. In addition, the large size of some of the networks can make traditional MC-MLE process computationally infeasible. The latter case is more easily accommodated. In cases where there are more than 1,000 nodes I use MPLE parametric bootstrap as described in Schmid and Desmarais(2017). This is only necessary in 38 of the networks and in these cases I take 100 draws in order to estimate the distribution of the coefficients of interest. In addition a subset of estimations did not converge if the percentage of Democratic donations were included either as the difference in the percentage of donations or the summation of the percentage of donations. These networks were estimated with those variables dropped. The problem from convergence was likely a result of their being collinearity between those variables and the MWESP variables. After estimation, models with an absolute

136 value of MWESP greater than 15 were checked for additional convergence issues and were often re-estimated after dropping the other party variables.

137 Bibliography

Abramowitz, Alan I and Kyle L Saunders. 2008. “Is polarization a myth?” The Journal of Politics 70(2):542–555.

Abramowitz, Alan I. and Steven Webster. 2016. “The Rise Of Negative Partisanship And The Nationalization Of Us Elections In The 21st Century.” Electoral Studies 41:12–22.

Aldrich, John. 1995. Why Parties? Chicago University Press.

Aldrich, John H and David W Rohde. 2001. The logic of conditional party government: Re- visiting the electoral connection. In Congress Reconsidered, ed. L Dodd and B Oppenhimer. CQ Press.

Aldrich, John H and James S Coleman Battista. 2002. “Conditional party government in the states.” American Journal of Political Science pp. 164–172.

Anzia, Sarah F. and Molly C. Jackman. 2013. “Legislative Organizaiton and the Second Face of Power: Evidence from U.S. State Legislatures.” Journal of Politics 75(1):210–224.

Bafumi, Joseph and Michael C Herron. 2010. “Leapfrog representation and extremism: A study of American voters and their members in Congress.” American Political Science Review 104(3):519–542.

Balz, Dan. 2006. “McCain Reconnects with Liberty University.” Washington Post May 14, 2006.

Barber, Michael J. 2016. “Ideological Donors, Contribution Limits, and the Polarization of American Legislatures.” The Journal of Politics 78(1):296–310.

Bartels, Larry M. 2000. “Partisanship and voting behavior, 1952-1996.” American Journal of Political Science pp. 35–50.

Baumgartner, Frank R, Jeffrey M Berry, Marie Hojnacki, Beth L Leech and David C Kimball. 2009. Lobbying and policy change: Who wins, who loses, and why. University of Chicago Press.

138 Bawn, Kathleen, Martin Cohen, David Karol, Seth Masket, Hans Noel and John Zaller. 2012. “A Theory of Political Parties: Groups, Policy Demands and Nomination in American Politics.” Perspective on Politics 10(3):571–597.

Baylor, Christopher A. 2013. “First to the Party: The Group Origins of the Partisan Transformation on Civil Rights, 1940-1960.” Studies in American Political Development 27:111–141.

Berry, William D, Evan J Ringquist, Richard C Fording and Russell L Hanson. 1998. “Measuring citizen and government ideology in the American states, 1960-93.” American Journal of Political Science pp. 327–348.

Bezáková, Ivona. 2008. “Sampling binary contingency tables.” Computing in Science & Engineering 10(2).

Bhaskar, V. 2000. “Egalitarianism and efficiency in repeated symmetric games.” Games and Economic Behavior 32(2):247–262.

Bonica, Adam. 2014. “Mapping the ideological marketplace.” American Journal of Political Science 58(2):367–386.

Borgatti, Stephen P and Daniel S Halgin. 2011. “Analyzing affiliation networks.” The Sage handbook of social network analysis 1:417–433.

Breiger, Ronald L. 1974. “The Duality of Persons and Groups.” Social Forces 53(2):181–190.

Brown, Robert D. 1995. “Party Cleavages and Welfare Effort in the American States.” American Political Science Review 89(1):23–33.

Burke, Edmund. 1770. “Thoughts on the cause of the present discontents.”.

Butts, Carter T. 2009. “Revisiting the foundations of network analysis.” science 325(5939):414– 416.

Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The American Voter. John Wiley and Sons, Inc.

Canes-Wrone, Brandice, David W Brady and John F Cogan. 2002. “Out of step, out of office: Electoral accountability and House members’ voting.” American Political Science Review 96(1):127–140.

Carpenter, Tim. 2018. “House moves to restore teacehr due process.” The Topeka Capital- Journal March 8, 2018.

Carpenter, Time. 2017. “Kansas state government bond debt surges $2 billion since 2010.” The Topeka Capital-Journal January 12, 2017.

139 Carroll, Royce and Jason Eichorst. 2013. “The role of party: The legislative consequences of Partisan electoral competition.” Legislative Studies Quarterly 38(1):83–109.

Carsey, Thomas M and Geoffrey C Layman. 2006. “Changing sides or changing minds? Party identification and policy preferences in the American electorate.” American Journal of Political Science 50(2):464–477.

Carson, Jamie L. 2005. “Strategy, selection, and candidate competition in US House and Senate elections.” Journal of Politics 67(1):1–28.

Cartwright, Dorwin and Frank Harary. 1956. “Structural balance: a generalization of Heider’s theory.” Psychological review 63(5):277.

Caswell, Bruce E. 2009. “The presidency, the vote, and the formation of new coalitions.” Polity 41(3):388–407.

Caughey, Devin, James Dunham and Chris Warshaw. 2016. Polarization and Partisan Divergence in the American Public, 1946-2012. Midwest Political Science Association Conference. Presented at the Midwest Political Science Association Conference, 2016.

Cha, J. Mijin and Miles Rapoport. 2013. Fresh Start: The Impact of Public Campaign Financing in Connecticut. Demos.

Chen, Jowei, Jonathan Rodden et al. 2013. “Unintentional gerrymandering: Political geography and electoral bias in legislatures.” Quarterly Journal of Political Science 8(3):239– 269.

Clifton, Brett M. 2004. “Romancing the GOP: Assessing the strategies used by the Christian Coalition to influence the Republican Party.” Party Politics 10(5):475–498.

Cohen, Marty, David Karol, Hans Noel and John Zaller. 2008. The Party Decides. University of Chicago Press.

Committee on Political Parties, American Political Science Association. 1950. Toward a More Responsible Two-Party System. Rinehart.

Cooper, Russell, Douglas V DeJong, Robert Forsythe and Thomas W Ross. 1989. “Commu- nication in the battle of the sexes game: some experimental results.” The RAND Journal of Economics pp. 568–587.

Cotter, Cornelius C., James L. Gibson, John F. Bibby and Robert J. Huckshorn. 1989. Party Organizations in American Politics. University of Pittsburgh Press.

Cox, Gary W. 1997. Making votes count: strategic coordination in the world’s electoral systems. Cambridge University Press.

140 Cox, Gary W. and Mathew D. Mccubbins. 1993. Legislative Leviathan: Party Government in the House. University of California Press.

Cox, Gary W. and Mathew D. McCubbins. 2005. Setting the Agenda. Cambridge University Press.

Cranmer, Skyler J and Bruce A Desmarais. 2010. “Inferential network analysis with expo- nential random graph models.” Political Analysis 19(1):66–86.

Cranmer, Skyler J, Philip Leifeld, Scott D McClurg and Meredith Rolfe. 2017. “Navigating the range of statistical tools for inferential network analysis.” American Journal of Political Science 61(1):237–251.

Crawford, Vincent P and Hans Haller. 1990. “Learning how to cooperate: Optimal play in repeated coordination games.” Econometrica: Journal of the Econometric Society pp. 571– 595.

Dahl, Robert. 1961. Who Governs? Democracy and Power in an American City. Yale University Press.

David, Paul T. 1992. “The APSA committee on political parties: Some reconsiderations of its work and significance.” Perspectives on Political Science 21(2):70–79.

Davis, Alison and Burleigh Gardner. 1941. Deep South: A Social Anthropolgical Study of Caste and Class. University of Chicago Press.

Desmarais, Bruce A., Raymond J. La Raja and Michael S. Kowal. 2015. “The Fates of Challengers in U.S. House Elections: The Role of Extended Party Networks in Supporting Candidates and Shaping Electoral Outcomes.” American Journal of Political Science 59(1):194–211.

Downs, Anthony. 1957. An economic theory of political action in a democracy. Harper and Row.

Drutman, Lee. 2017. Political Divisions in 2016 and Beyond: Tensions Between and Within the Two Parties. Technical report Voter Study Group. URL: https://www.voterstudygroup.org/reports/2016-elections/political-divisions-in-2016- and-beyond

Duverger, Maurice. 1959. Political parties: Their organization and activity in the modern state. Methuen.

Fiorina, Morris P. 2002. “Parties and partisanship: A 40-year retrospective.” Political Behavior 24(2):93–115.

Fiorina, Morris P and Samuel J Abrams. 2008. “Political polarization in the American public.” Annu. Rev. Polit. Sci. 11:563–588.

141 Fleischer, Doris Zames, Fleischer Doris Zames and Frieda Zames. 2012. The disability rights movement: From charity to confrontation. Temple University Press. Fouirnaies, Alexander and Andrew B Hall. 2014. “The Financial Incumbency Advantage: Causes and Consequences.” The Journal of Politics 76(3):711–724. Frymer, Paul. 2010. Uneasy alliances: Race and party competition in America. Princeton University Press. Gibson, James L, Cornelius P Cotter, John F Bibby and Robert J Huckshorn. 1983. “Assessing party organizational strength.” American Journal of Political Science pp. 193–222. Gilens, Martin. 2012. Affluence and influence: Economic inequality and political power in America. Princeton University Press. Gimpel, James. 1996. National Elections and the Autonomy of American State Party Sytems. University of Pittsburgh Press. Gimpel, James Graydon and Jason E Schuknecht. 2009. Patchwork nation: Sectionalism and political change in American politics. University of Michigan Press. Goodliffe, Jay. 2001. “The effect of war chests on challenger entry in US House elections.” American Journal of Political Science pp. 830–844. Goodliffe, Jay. 2005. “When Do War Chests Deter?” Journal of Theoretical Politics 17(2):249– 277. Goodliffe, Jay. 2007. “Campaign war chests and challenger quality in Senate elections.” Legislative Studies Quarterly 32(1):135–156. Goodreau, Steven M. 2007. “Advances in exponential random graph (p*) models applied to a large social network.” Social networks 29(2):231–248. Goodreau, Steven M, James A Kitts and Martina Morris. 2009. “Birds of a feather, or friend of a friend? Using exponential random graph models to investigate adolescent social networks.” Demography 46(1):103–125. Granovetter, Mark S. 1977. “The strength of weak ties.” Social networks pp. 347–367. Grossman, Matt and Casey B. K. Dominguez. 2009. “Party Coalitions and Interest Group Networks.” American Politics Research 37(5):767–800. Grossmann, Matt and David A Hopkins. 2016. Asymmetric politics: Ideological Republicans and group interest Democrats. Oxford University Press. Grund, Thomas U and James A Densley. 2015. “Ethnic homophily and triad closure: Mapping internal gang structure using exponential random graph models.” Journal of Contemporary Criminal Justice 31(3):354–370.

142 Hallinan, Maureen T and Warren N Kubitschek. 1988. “The effects of individual and structural characteristics on intransitivity in social networks.” Social Psychology Quarterly pp. 81–92.

Hanna, John. 2013a. “Some GOP lawmakers resisting gov’s tax plan.” The Hutchinson News February 8, 2013.

Hanna, John. 2013b. “Tepid OK marks tax plan - Some Republicans vote for it reluctantly; measure adds to income tax cuts.” The Hutchinson News June 3, 2013.

Hanna, John. 2014. “Kansas lawmakers back school funding, tenure plan.” The Hutchinson News April 7, 2014.

Hanna, John. 2017. “Kansas legislators repudiate governor by overriding tax veto.” Associated Press Statewire: Kansas June 7, 2017.

Hassell, Hans. 2016. “Party Control of Party Primaries: Party Influence in Nominations for the US Senate.” Journal of Politics 78(1):75–87.

Heaney, Michael T and Fabio Rojas. 2015. Party in the Street: The Antiwar Movement and the Democratic Party after 9/11. Cambridge University Press.

Hetherington, Marc J. 2001. “Resurgent mass partisanship: The role of elite polarization.” American Political Science Review 95(3):619–631.

Heyward, Anna. 2017. “Since Trump’s Victory, Democratic Socialists of America Has Become a Budding Political Force.” The Nation December.

Hinchliffe, Kelsey L and Frances E Lee. 2016. “Party Competition and Conflict in State Legislatures.” State Politics & Policy Quarterly 16(2):172–197.

Hirano, Shigeo, Gabriel S Lenz, Maksim Pinkovskiy and James M Snyder. 2015. “Voter Learning in State Primary Elections.” American Journal of Political Science 59(1):91–108.

Holden, Zach. 2016. 2013 and 2014: Monetary Competitiveness in State Legislative Races. Technical report National Institute on Money in State Politics. URL: http://followthemoney.org/research/institute-reports/2013-and-2014-monetary- competitiveness-in-state-legislative-races/

Horvát, Emőke-Ágnes and Katharina Anna Zweig. 2013. “A fixed degree sequence model for the one-mode projection of multiplex bipartite graphs.” Social Network Analysis and Mining 3(4):1209–1224.

Hunter, Chris. 2013. “Kansas GOP group says party ‘hijacked’ by extremists.” Kansas City Star September 18, 2013.

Hunter, David R. 2007. “Curved exponential family models for social networks.” Social networks 29(2):216–230.

143 Hunter, David R, Pavel N Krivitsky and Michael Schweinberger. 2012. “Computational statistical methods for social network models.” Journal of Computational and Graphical Statistics 21(4):856–882.

Jacoby, William. 2004. Ideology in the 2000 Elections: A Study in Ambivalence. In Models of Voting in Presidential Elections, ed. Hervert F. Weisberg and Clyde Wilcox. Stanford University Press pp. 103–222.

Jenkins, Jeffery A and Nathan W Monroe. 2012. “Buying negative agenda control in the us house.” American Journal of Political Science 56(4):897–912.

Jessee, Stephen A. 2010. “Partisan bias, political information and spatial voting in the 2008 presidential election.” The Journal of Politics 72(2):327–340.

Kamisar, Ben and Reid Wilson. 2017. “Dem campaign chief vows no litmus test on abortion.” The Hill July 31, 2017.

Kessler, Glenn. 2017. “History Lesson: How the Democrats pushed Obamacare through the Senate.” The Washington Post June 22, 2017.

Key, V. O. 1949. Southern Politics in State and Nation. A. A. Knopf.

King, Gary, James Honaker, Anne Joseph and Kenneth Scheve. 2001. “Analyzing incomplete political science data: An alternative algorithm for multiple imputation.” American political science review 95(1):49–69.

Knez, Marc and Colin Camerer. 2000. “Increasing cooperation in prisoner’s dilemmas by establishing a precedent of efficiency in coordination games.” Organizational Behavior and Human Decision Processes 82(2):194–216.

Koger, Gregory, Seth Masket and Hans Noel. 2009. “Partisan Webs: Information Exchange and Party Networks.” British Journal of Political Science 16(3):57–85.

Kohavi, Ron. 1995. A study of cross-validation and bootstrap for accuracy estimation and model selection. In International Joint Conference on Artificial Intelligence. Vol. 2 Montreal, Canada pp. 1137–1145.

La Raja, Raymond J. 2013. “Richer Parties, Better Politics? Party-Centered Campaign Finance Laws and American Democracy.” The Forum 11(3):313–338.

Ladewig, Jeffrey W. 2005. “Conditional party government and the homogeneity of constituent interests.” Journal of Politics 67(4):1006–1029.

Laitin, David D and Aaron Wildavsky. 1988. “Political culture and political preferences.” American Political Science Review 82(02):589–597.

144 Latapy, Matthieu, Clémence Magnien and Nathalie Del Vecchio. 2008. “Basic notions for the analysis of large two-mode networks.” Social networks 30(1):31–48.

Lau, Richard R. 2013. “Correct voting in the 2008 US presidential nominating elections.” Political Behavior 35(2):331–355.

Layman, Geoffrey C and Thomas M Carsey. 2002. “Party polarization and" conflict extension" in the American electorate.” American Journal of Political Science pp. 786–802.

Layman, Geoffrey C, Thomas M Carsey, John C Green, Richard Herrera and Rosalyn Cooperman. 2010. “Activists and conflict extension in American party politics.” American Political Science Review 104(2):324–346.

Layman, Geoffrey C, Thomas M Carsey and Juliana Menasce Horowitz. 2006. “Party polarization in American politics: Characteristics, causes, and consequences.” Annu. Rev. Polit. Sci. 9:83–110.

Lazarus, Jeffrey. 2008. “Incumbent vulnerability and challenger entry in statewide elections.” American Politics Research 36(1):108–129.

Levendusky, Matthew. 2009. The partisan sort: How liberals became Democrats and conser- vatives became Republicans. University of Chicago Press.

Lowery, David and Virginia Gray. 2004. “A neopluralist perspective on research on organized interests.” Political Research Quarterly 57(1):164–175.

Madison, James. 1787. “The federalist no. 10.” November 22(1787):1787–88.

Marso, Andy. 2012. “Governor Gives Tax Plan Pep Talk.” Topeka Capital-Journal May 19, 2012.

Masket, Seth. 2011. No Middle Ground: How Informal Party Organizations Control Nomina- tions and Polarize Legislatures. University of Michigan Press.

Masket, Seth. 2016. The Inevitable Party: Why Attempts to Kill the Party System Fail and How They Weaken Democracy. Oxford University Press.

Masket, Seth and Boris Shor. 2015. “Polarization without parties: Term limits and legislative partisanship in Nebraska’s Unicameral Legislature.” State Politics & Policy Quarterly 15(1):67–90.

Masket, Seth E. 2007. “It takes an outsider: Extralegislative organization and partisanship in the California Assembly, 1849–2006.” American Journal of Political Science 51(3):482–497.

Mayer, Jane. 2016. Dark Money: The Hidden History of the Billionaires Behind the Rise of the Radical Right. Anchor.

145 Mayhew, David R. 1986. Placing Parties in American Politics: Organization, Electoral Settings, and Government Activity in the Twentieth Century. Princeton University Press.

Mazerov, Michael. 2018. Kansas Provides Compelling Evidence of Failure of “Supply-Side” Tax Cuts. Center on Budget and Policy Priorities.

McCarthy, John and Mayer N Zald. 2003. Social movement organizations. In The social movements reader: Cases and concepts, ed. Jeff Goodwin and James M. Jasper. Blackwell Publishing: Malden, MA pp. 169–186.

McCarty, Nolan, Keith T. Poole and Howard Rosenthal. 2016. Polarized America: The dance of ideology and unequal riches. MIT Press.

McFarland, Andrew S. 2007. “Neopluralism.” Annu. Rev. Polit. Sci. 10:45–66.

McGhee, Eric, Seth Masket, Boris Shor, Steven Rogers and Nolan McCarty. 2014. “A primary cause of partisanship? Nomination systems and legislator ideology.” American Journal of Political Science 58(2):337–351.

McLean, Jim. 2015. “Wichita Republican leading effort to roll back Brownback tax cuts.” Kansas Health Institute News Service June 5, 2015.

McPherson, Miller, Lynn Smith-Lovin and James M Cook. 2001. “Birds of a feather: Homophily in social networks.” Annual review of sociology 27(1):415–444.

Morrow, James D. 1994. Game theory for political scientists. Princeton University Press Princeton, NJ.

Murphy, Jan. 2015. “Gop Senator to Launch Campaign for Pa. Attorney General.” Penn Live June 15, 2015.

Neal, Zachary. 2014. “The Backbone of Bipartite Projections: Inferring Relationships from Co-Authorship, Co-Sponsorship, Co-Attendance and other Co-Behaviors.” Social Networks 39:84–97.

Neiheisel, Jacob R and Sarah Niebler. 2013. “The use of party brand labels in congressional election campaigns.” Legislative Studies Quarterly 38(3):377–403.

Newman, Mark. 2010. Networks: An Introduction. Oxford University Press.

Nyhan, Brendan and Jacob M. Montgomery. 2015. “Connecting the Candidates: Consultant Networks and the Diffusion of Campaign Strategy in American Congressional Elections.” American Journal of Political Science 59(2):292–308.

Oldfield, Duane Murray. 1996. The right and the righteous: The Christian Right confronts the Republican Party. Rowman & Littlefield.

146 Olson, Mancur. 1965. logic of collective action; public goods and the theory of groups. Harvard University Press.

Painter, Steve. 2005. “Move Splinters Kansas Republicans.” The Wichita Eagle December 8, 2005.

Painter, Steve. 2006. “Division Undermines GOP.” The Wichita Eagle May 30, 2006.

Pear, Robert and Thomas Kaplan. 2017. “Senate Rejects Slimmed-Down Obamacare Repeal as McCain Votes No.” The New York Times July 27, 2017.

Piketty, Thomas. 2014. Capital in the 21st Century. Harvard University Press.

Polsby, Nelson W. 1968. “The institutionalization of the US House of Representatives.” American Political Science Review 62(01):144–168.

Poole, Keith T and Howard Rosenthal. 1991. “Patterns of congressional voting.” American journal of political science pp. 228–278.

Poole, Keith T and Howard Rosenthal. 2000. Congress: A political-economic history of roll call voting. Oxford University Press on Demand.

Public Citizen. 2017. “2016: Advancing a Progressive Agenda.” Public Citizen News 2017(Jan/Feb):5–13.

Rae, Nicol C. 2007. “Be careful what you wish for: The rise of responsible parties in American national politics.” Annu. Rev. Polit. Sci. 10:169–191.

Ramsden, Graham P. 2002. “State Legislative Campaign Finance Research: A Review Essay.” State Politics and Policy Quarterly 2(2):176–198.

Rigby, Elizabeth and Gerald C. Wright. 2013. “Political Parties and Representation of th ePoor in the American States.” American Journal of Political Science 57(3):552–565.

Riker, William H. 1982. “The two-party system and Duverger’s law: an essay on the history of political science.” American Political Science Review 76(04):753–766.

Rodriguez, Juan D, Aritz Perez and Jose A Lozano. 2010. “Sensitivity analysis of k-fold cross validation in prediction error estimation.” IEEE transactions on pattern analysis and machine intelligence 32(3):569–575.

Rohde, David W. 1991. Parties and Leaders in the Postreform House. University of Chicago Press.

Rohde, David W. 2013. “Reflections on Practices of Theorizing: Conditional Party Govern- ment in the Twenty-First Century.” Journal of Politics 75:849–864.

147 Roscoe, Douglas D. and Shannon Jenkins. 2015. Local Party Organizations in the Twenty-First Century. State University of New York Press. Rubin, Donald B. 1987. Multiple imputation for nonresponse in surveys. John Wiley & Sons. Schattschneider, E. E. 1942. Party Government. Rinehart and Company. Schlozman, Daniel. 2015. When movements anchor parties: Electoral alignments in American history. Princeton University Press. Schlozman, Kay Lehman, Sidney Verba and Henry E Brady. 2012. The unheavenly chorus: Unequal political voice and the broken promise of American democracy. Princeton University Press. Schmid, Christian S and Bruce A Desmarais. 2017. “Exponential Random Graph Models with Big Networks: Maximum Pseudolikelihood Estimation and the Parametric Bootstrap.” arXiv preprint arXiv:1708.02598 . Schumpeter, Joseph A. 1942. From Capitalism, Socialism and Democracy. Harper and Brothers Publishing. Schweinberger, Michael. 2011. “Instability, sensitivity, and degeneracy of discrete exponential families.” Journal of the American Statistical Association 106(496):1361–1370. Serrano, M Ángeles, Marián Boguná and Alessandro Vespignani. 2009. “Extracting the multiscale backbone of complex weighted networks.” Proceedings of the national academy of sciences 106(16):6483–6488. Shor, Boris and Nolan McCarty. 2011. “The Ideological Mapping of American Legislatures.” American Political Science Review 105(3):530–551. Skinner, Richard, Seth Masket and David Dulio. 2013. “527 Committees, Formal Parties, and Party Adaptation.” The Forum 11(2):137–156. Snyder Jr, James M. 1992. “Long-term investing in politicians; or, give early, give often.” The Journal of Law and Economics 35(1):15–43. Squire, Peverill. 1992. “The theory of legislative institutionalization and the California assembly.” Journal of Politics 54(4):1026–54. Strolovitch, Dara Z. 2008. Affirmative advocacy: Race, class, and gender in interest group politics. University of Chicago Press. Sullivan, Martin A. 2012. “Brownback’s Derailed Tax Plan Becomes a Fiscal Train Wreck.” State Tax Notes June 4, 2012. The Council of State Governments. 2017. Book of the States. The Council of State Govern- ments.

148 Thompson, Frank J. 2008. “State and local governance fifteen years later: enduring and new challenges.” Public Administration Review 68(s1).

Trochim, William M. K. and James P. Donnelly. 2001. Research Methods Knowledge Base. Atomic Dog Pub.

Voss, Thomas. 2001. Game-theoretical perspectives on the emergence of social norms. In Social Norms, ed. Michael Hechter and Karl-Dieter Opp. Russel Sage.

Washington, George. 1796. “Farewell address.”. September 19, 1796.

Watts, Duncan J and Steven H Strogatz. 1998. “Collective dynamics of’small-world’networks.” nature 393(6684):440.

Weber, Till and Craig Parsons. 2016. “Dynamic party unity: the US Congress in comparative perspective.” European Political Science Review 8(4):637–662.

Weinschenk, Aaron C. 2013. “Polls and elections: partisanship and voting behavior: an update.” Presidential Studies Quarterly 43(3):607–617.

Wellman, Barry. 1983. “Network analysis: Some basic principles.” Sociological theory pp. 155– 200.

Wildavsky, Aaron. 1987. “Choosing preferences by constructing institutions: A cultural theory of preference formation.” American Political Science Review 81(01):3–21.

Wistrom, Brent. 2012. “Brownback looks forward to signing massive tax-cutting bill.” Witchita Eagle May 18, 2012.

Yang, Song, Scott Limbocker, Andrew Dowdle, Patrick A. Stewart and Karen Sebold. 2015. “Party Cohesion in Presidential Races: Applying Social Network Theory to the Preprimary Multiple Donor Networks of 2004 and 2008.” Party Politics 21(4):638–648.

149 Vita Kevin Reuning

210/209 Pond Laboratory Email: [email protected] State College, PA 16803 Web: www.kevinreuning.com Phone: 612-598-6118

Education

The Pennsylvania State University Ph.D. Political Science, August 2018 Fields: American Politics and Methodology Dissertation: Party Coalitions, Party Ideology, and Party Action: Extended Party Networks in the United States Committee: Lee Ann Banaszak (chair), Michael Berkman, Bruce Desmarais, John McCarthy (Penn State sociology), and Michael Nelson

The Pennsylvania State University M.A. Political Science, 2015 Fields: American Politics and Methodology Case Western Reserve University B.A. Political Science and Mathematics (minor in Economics), 2010

Academic Positions

Assistant Professor, Miami University 2018 -

Peer Reviewed Articles

Reuning, Kevin and Nick Dietrich. Forthcoming. “Media, Public Interest, and Support in the Invisible Primary”. Perspectives on Politics 17:2.

Working Papers

“Exploring the Dynamics of Latent Variable Models” with Michael R. Kenwick and Christopher J. Fariss Conditional accept at Political Anaylsis “Movement Success, Political Fracturing, and Demobilization: Explaining the Tea Party’s Short Romance with Protest Demonstrations” with John McCarthy, Patrick Rafail and Hyun Woo Kim. Under review