Policy Collaboration in the United States Congress

Home , Aisle (political term), Bipartisanship, Congressional staff, Don Cazayoux, Problem Solvers Caucus, Scott Rigell

Dissertation

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

Alison W. Craig, M.A.

Graduate Program in Political Science

The Ohio State University

2017

Dissertation Committee:

Janet M. Box-Steffensmeier, Advisor Skyler J. Cranmer Michael A. Neblo Herbert F. Weisberg ⃝c Copyright by

Alison W. Craig

2017 Abstract

Is there a benefit to working well with others in Congress? Many of the bills introduced are written not only by the single member listed as its sponsor, but by a coalition of representatives who have worked together to author mutually agreeable language. Similarly, members frequently collaborate with colleagues in writing policy letters, running caucuses, and hosting events. Yet there is very little understanding of the nature of these relationships, or how members of Congress benefit from them, as data availability has limited the ability of legislative politics scholars to estimate their impact. Using a unique dataset of Dear Colleague letters, which are an essential communication tool in the modern Congress, I identify the members who collaborate on policy initiatives in a substantive manner. I use these data to map the policy collaboration network of the House of Representatives to answer three key questions that will greatly improve our understanding of congressional behavior and the legislative process: 1) How do members of Congress choose their collaborative partners? 2)

What are the legislative benefits of collaboration? 3) What are the electoral benefits of collaboration?

The first question is addressed using a temporal exponential random graph model

(TERGM) that allows me to consider the policy collaboration network for each

Congress in its entirety and examine the endogenous and exogenous factors that lead members to working with each other. I find evidence of several distinctive patterns,

ii including a strong tendency towards bipartisan collaboration in a highly polarized

Congress, an overall inclination towards collaboration where there are shared constituencies, and a network where personal relationships and reputations are key. The second essay examines the legislative benefits of collaboration, specifically whether more collaborative members are more effective legislators. I create several new measures of propensity towards collaboration and use them in a series of temporal network autocorrelation models that examine whether the relationship between collaboration and legislative effectiveness is the result of members putting in effort to advance their agenda, working with other successful colleagues, or using collaboration to send infor- mative signals. I find that members who are strategic in their collaborative decisions find the most success, particularly those who moderate their usage of collaboration.

Finally, I consider the electoral benefits of collaboration, again using the temporal network autocorrelation model and my measures of propensity towards collaboration.

I find that for electorally vulnerable members of Congress, there is a significant benefit to collaborating with members of the other party as it allows them tobuilda reputation for bipartisanship with their constituents.

Taken together, these three essays provide us with a greater understanding of the role that policy collaboration plays in the modern Congress. Members use collaboration with their colleagues to find common ground in a polarized Congress, to advance their legislative agenda, and as a form of symbolic representation that allows them to distance themselves from the “dysfunctional” Congress.

iii For my parents, who never stopped teaching

me about politics.

iv Acknowledgments

I am indebted to the institutions who provided invaluable financial support for my research: the National Science Foundation Political Science program (grant #1627358), the National Science Foundation Graduate Research Fellowship (DGE-1343012), the

Dirksen Congressional Center (grant #00031432) and the Institute for the Study of

Democracy at the Ohio State University. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation or other sponsoring entity.

This research would not have been possible without the work of scholars who have come before me. In particular, I would like to thank those whose data and resources were essential to this project: Scott Adler and John Wilkerson with the

Congressional Bills Project, Craig Volden and Alan Wiseman with the Legislative

Effectiveness Project, and Philip Leifeld, Skyler Cranmer, and Bruce Desmarais with the xergm: Extensions of Exponential Random Graph Models package.

I would also like to thank those who have provided helpful comments and sugges- tions for portions of this project along the way including Jenna Bednar, Sarah Binder,

Matthew Green, Michael Heaney, Greg Koger, Michael Lynch, Will Massengill, Nate

Monroe, Jason Morgan, Vincent Moscardelli, Jason Roberts, Wendy Schiller, Sean

v Theriault, and Jennifer Victor, as well participants in the American Politics Workshop at Ohio State University and the Visions in Methodology Conference.

I have been incredibly fortunate to work with a truly fantastic committee and I would have been lost at several points along the way without their advice and support.

My chair, Jan Box-Steffensmeier, is the best mentor a person could ask for, inspiring me with her brilliance, gently steering me back on course whenever I was distracted by shiny objects, and encouraging me with her endless faith in me. Skyler Cranmer provided not only methodological guidance, but also pushed me to work harder, think bigger, and stop giving boring presentations. Michael Neblo’s enthusiasm for me and my project was a constant source of inspiration, and Herb Weisberg could always be counted on to talk out even my most half-baked ideas.

I am also indebted to the colleagues from my past life as a congressional staffer who supported me in the decision to turn my life upside down to become an academic and have continued to serve as a resource in my research. And I would be remiss if

I did not mention the influence of a certain John McCain campaign volunteer who sparked my political involvement at the age of six when he stole my sign.

Finally, I owe thanks to all of my cheerleaders who provided endless support and encouragement along the way, but in particular Kati, who put up with my whining and yelled at me when I needed it, Bri, who kept me going with a healthy balance of coffee and wine, and Greg, who never shut up about how proud he was ofme.And my parents, who helped me be fearless, with the knowledge that they would catch me if I fell.

vi Vita

2001 ...... B.S. Political Science, University of Oregon 2001-2008 ...... Congressional Staff, Congresswoman Darlene Hooley 2009-2012 ...... Congressional Staff, Congressman Kurt Schrader 2014 ...... M.A. Political Science, Ohio State University 2014-2017 ...... Graduate Research Fellow, National Science Foundation

Fields of Study

Major Field: Political Science

vii Table of Contents

Page

Abstract ...... ii

Dedication ...... iv

Acknowledgments ...... v

Vita...... vii

List of Tables ...... x

List of Figures ...... xii

1. Introduction ...... 1

2. The Room Where it Happens: Collaborative Strategies in the U.S. House of Representatives ...... 13

3. Lone Wolves and Team Players: Policy Collaboration Networks and Leg- islative Effectiveness in the House of Representatives ...... 52

4. Running from Washington: Policy Collaboration as Symbolic Representation 92

5. Conclusion ...... 126

Bibliography ...... 133

viii Appendices ...... 147

A. Sample Dear Colleague Letters ...... 149

B. Alternate ERGM Specifications ...... 153

ix List of Tables

Table Page

2.1 Probability of Policy Collaboration ...... 43

3.1 Policy Collaboration Network Summary Statistics ...... 71

3.2 Most and Least Collaborative Members of the 110th Congress . . . . 76

3.3 Relationship Between Collaboration and Legislative Effectiveness . . 83

3.4 Relationship Between Bipartisanship and Legislative Effectiveness . . 88

4.1 Summary Statistics for Dependent and Key Independent Variables . . 113

4.2 Relationship Between Past Electoral Performance and Collaboration . 118

4.3 Relationship Between Collaboration and Electoral Performance . . . . 120

x B.1 Probability of Policy Collaboration with Party Mixing Covariate . . . 154

xi List of Figures

Figure Page

2.1 Promoting Bipartisan Collaboration via Twitter ...... 18

2.2 Distribution of Letters by Purpose in Sample Population ...... 25

2.3 Signers per Letter ...... 29

2.4 One-Mode Projections of the Member Networks ...... 31

2.5 Partisanship of Ties by Congress ...... 34

2.6 GWESP and GWDSP statistics ...... 37

2.7 Probability of Tie Formation Coefficients ...... 45

2.8 Goodness-of-fit Diagnostics ...... 47

xii 3.1 Distribution of Dear Colleague Letters by Purpose ...... 64

3.2 Sample Dear Colleague Letter from 110th Congress ...... 66

3.3 Distribution of Signers by Source ...... 68

3.4 Policy Collaboration Network for the 110th Congress ...... 70

4.1 Promotion of Policy Collaboration via Twitter ...... 99

4.2 Distribution of Dear Colleague Letters by Purpose ...... 102

4.3 Sample Dear Colleague Letter from 111th Congress ...... 104

4.4 Ego Networks for Congressman Henry Cuellar (D-TX) and Congress-

man Glenn Thompson (R-PA) in the 111th Congress (2009-2010) . . 109

A.1 Dear Colleague Example: Community Pharmacies ...... 150

A.2 Dear Colleague Example: U.S.-Qatar Relations ...... 151

A.3 Dear Colleague Example: Health Care ...... 152

xiii Chapter 1: Introduction

Collaboration on policy is a widespread phenomenon in the United States Congress and yet it remains largely underestimated and misunderstood. Public perception is of a legislature so dysfunctional that members barely speak to each other, let alone collaborate with each other. This view is supported by regular stories in the media decrying the lack of civility in Washington, DC, many of which quote former members of Congress who mourn the days when members brought their families to

DC, built social relationships with each other, and “would play tennis atop the Hart

Office Building” (e.g. Merica, 2013; Thomas, 2017). Partisanship in Congress has undoubtedly been on the rise when measured by the roll call votes that members take, but that is a trend that began in the 1970s, two decades before Newt Gingrich urged Republican members of the House to leave their families at home. And despite substantial disagreement on the major policy issues of the day such as health care and tax reform, members of Congress do still try to work together when possible.

As described by Congressman Erik Paulsen (R-MN), “Working with members of the other party, on legislation that matters, is the way I keep my sanity. With Karen

Bass (D-CA) and Louise Slaughter (D-NY), we worked on sex-trafficking legislation that just went to the president’s desk” (Warren, 2014).

1 The importance of relationships in Congress is routinely emphasized in congres-

sional scholarship, from members who take cues from each other when making voting

decisions (Matthews and Stimson, 1975; Kingdon, 1989) to members collaborating

on press events in the Senate (Desmarais et al., 2015). The interest in these studies

is not in the social relationships between members of Congress, but rather the policy relationships. In fact, there is very little evidence to support the assertion that discord in Congress would be lessened if members spent more time socializing with each other. Some of the policy relationships may also be social relationships, but others form between members sitting next to each other in their committee, or seeking each other out due to a shared interest in an issue. When it comes to the effect of relationships between members on the functioning of Congress as an institution,

I assert that members who collaborate on policy but never see each other outside of the House office buildings have a more meaningful relationship than members who have dinner together but avoid discussing politics. The benefit of the relationship is what it represents: two members of Congress coming together and finding common ground.

Despite the importance of policy collaboration in Congress, we have very little understanding of what these relationships look like. Political science acknowledges the existence of collaboration and it is frequently mentioned in qualitative interview data, however empirical studies largely dismiss coauthored legislation as inconsequen- tial or use measures that fail to fully capture the collaborative relationship between members. Much of the legislative politics literature treats members of Congress as independent observations, if not theoretically, then methodologically. There has been a much-needed rise in the study of Congress as a network, driven by Fowler (2006a,b),

2 whose study of the cosponsorship network has provided important insights into the connectedness of members and the relationship between cosponsorship and legislative productivity. However, the data needed to distinguish substantive policy collaboration from policy agreement frequently reflected in cosponsorship and voting similarity have not been available until now as members of Congress are difficult to survey and reluctant to discuss the specifics of their relationships.

I introduce a new dataset of congressional communications that allow us to observe the core policy collaboration network of the U.S. House of Representatives for the first time. When members of Congress introduce legislation they frequently send a letter, known as a Dear Colleague letter, to all House member and committee offices to announce their work and ask their colleagues to sign on as cosponsors. These letters have a long history in Congress, as early House rules required members to build support for legislation before it could be introduced, resulting in letters sent between offices to describe and promote policy initiatives, although they did not cometobe known as “Dear Colleagues” until the early 20th century (Peterson, 2005). The letters existed solely on paper and were not archived in any systematic way until 1998 when

Congress created an electronic listserv and the distribution of Dear Colleague letters gradually shifted from paper to electronic letters.

While many of these letters are signed by a single member of Congress, nearly half are signed by two or more. Members who sign Dear Colleague letters together are jointly claiming credit for the policy in question and sending a clear signal to their colleagues that it is a collaborative effort. When Congressman Paulsen worked with

Congresswomen Bass and Slaughter on sex-trafficking legislation, they would have also sent out a Dear Colleague letter together, although the legislative record only

3 allows for a single member to be listed as a bill’s sponsor. Dear Colleagues provide a means to uncover these relationships which would otherwise go largely unnoticed.

With over 80,000 of these letters sent over a period of eight years (2003-2011), I identify the members who collaborate with each other on policy in a substantive manner. I then create a network of policy collaboration for each Congress in which members are connected to each other in the network if they signed a Dear Colleague letter together.

I introduce the policy collaboration network as a measure of substantive and purposive collaboration between members, which allows for the first empirical examination of the determinants and benefits of collaboration in the U.S. House of

Representatives. There is a cost to collaboration in that it requires members to compromise their own policy position to reach agreement with a colleague and it is more labor intensive than introducing a bill without any input from others. Some constituencies may prefer ideological purity to compromise and collaboration (Harbridge and Malhotra, 2011). So why do members choose to collaborate as frequently as they do? To gain a better understanding of both the nature of collaboration in Congress and its benefits I use Dear Colleague letters and the new insights into the legislative process that they provide to address three key questions: 1) How do members of

Congress choose colleagues to collaborate with? 2) What are the legislative benefits of collaboration? 3) What are the electoral benefits of collaboration? The answers to these questions not only shed light on the role of collaboration in Congress, but also provide new insights into a number of core issues in legislative politics including polarization in Congress, what constitutes legislative effectiveness, and the costs and benefits of bipartisanship.

4 What follows are three distinct yet thematically connected research papers that provide the first insights into the policy collaboration network of Congress andthe role it plays in both the legislative and electoral arenas of congressional politics.

This study is the starting point for a broader research agenda discussed further in chapter five that will provide a more comprehensive understanding of the benefits and drawbacks of collaboration for individual members, the policies they introduce, and Congress as a whole. However, the the first step to understanding collaboration in Congress is determining how and when those collaborative relationships form.

The first paper presented here, “The Room Where it Happens: Collaborative

Strategies in the U.S. House of Representatives,” addresses the question of who collaborates with whom in the policy collaboration network. How do members decide whom to work with in Congress? Considering the costs of collaboration, the decision is one members must carefully consider, as they are best served by working with a colleague who will help them achieve their goals, whether that entails seeing a bill passed and signed into law, or positioning themselves as an expert on a particular issue. For a moderate member of the minority party, working with a more ideologically extreme co-partisan could harm the already poor chances that their bill will be considered, while a member seeking to establish credibility in an issue area is likely best served by working with a more established colleague rather than a junior member of an irrelevant committee. Members may also be conscious of their reputation, as collaborating with a colleague means their names will be associated with each other in the minds of other members, and possibly in the public at large if the bill is one of the few that becomes known primarily by the names of its authors, such as

“Dodd-Frank,” or “McCain-Feingold.”

5 I argue that as a result, the decision to collaborate with a colleague is based on three key considerations: strategic advantages, personal relationships, and shared policy goals. Members who are aggressively working to advance their legislation or policy initiative through the House of Representatives are best served by prioritizing strategic considerations. These are the members who deliberately seek out bipartisan collaborators to work with. However, they also seek to minimize the ideological distance between themselves and their collaborators so as to lessen the amount of compromise that must be done to reach a mutually agreeable policy. So while members frequently collaborate with colleagues in the other party, they are more likely to prefer moderates, and the result of these collaborations is legislation or policy initiatives that members of both parties are more likely to feel they can support.

Reputation-minded legislators and those who wish to build strong relationships within Congress are more likely to be motivated by personal relationships. These are the collaborations that form because two members were introduced to each other by a common collaborator, or met at a dinner party. Although members of Congress may not be willing to disclose the details of their personal relationships in any way systematic enough to map the social relations of Congress, considering policy collaboration as a network at least allows us to account for these relationships through endogenous network statistics. Personal relationships also account for some members simply being more or less popular than others within the House. This was reinforced several times in interviews with congressional staff in which it was reported that members sought out colleagues who were viewed seriously, and avoided those with a reputation for being difficult (Interviews, 2016).

6 Finally, members who wish to establish credibility for themselves as being knowl- edgable and responsive on a particular issue are best served by collaborating with other members who have shared policy goals. This can encompass a number of different collaborative relationships, from two members from the same state working together on a local project of importance to their mutual constituents, to two members on the same committee working on an issue that a prominent interest group has prioritized. Collaborating with a colleague on the committee of jurisdiction for an issue sends a clear signal both to other members and to outside interests that the members are serious about the issue in question and collaborating with colleagues from the same state can have a similar effect for constituents at home.

To examine the role that strategic considerations, personal relationships, and shared policy goals play in the decision to collaborate with a colleague, I use a temporal exponential random graph model (TERGM) which models the probability of observing the network as it is observed compared to all of the networks that could have been observed with the same endogenous and exogenous characteristics (Des- marais and Cranmer, 2012). Rather than a dyad-level analysis that treats each pair of members independently, the TERGM allows me to consider the policy collaboration network for each Congress in its entirety.

The results of the model show evidence of all three collaborative dynamics at play in the modern Congress. Members frequently collaborate with colleagues from the other party, despite the polarization in Congress, and they seek to minimize the ideological distance between themselves and their collaborators. Existing relationships and reputations play a part in shaping other relationships in the policy collaboration network. And shared constituencies, whether voters or interest groups, are frequently

7 associated with increased collaboration as members from the same state, on the same committee, or co-chairing caucuses work together.

The second paper, “Lone Wolves and Team Players: Policy Collaboration Net- works and Legislative Effectiveness in the House of Representatives,” begins to tackle the question of how members of Congress benefit from collaboration, starting with the legislative benefits of forming these relationships. In interviews, congressional staff repeatedly emphasized the belief that collaborative legislation was more likely to be successful, particularly if it was the result of bipartisan collaboration (Inter- views, 2016). Members use cues from their colleagues when deciding what legislation to support and seeing that multiple members have worked together to find agreement on a bill or policy letter is a positive signal. This is particularly true when the collaborating members are from both parties, which signals that the bill is ideologically moderate or non-controversial. It stands to reason that members who collaborate more frequently with colleagues, particularly in the other party, should therefore be more successful as well. While a bill-level analysis will be the subject of future work, in this paper I examine whether members who form more relationships in Congress are rewarded for being a team player with increased legislative effectiveness.

Members may build strategic connections with colleagues who are in a position to help advance a bill because they are on a key committee, or have the ear of the leadership. The larger a member’s personal network is, the more likely they will have someone they can reach out to when they need help getting a bill through the legislative process. Members who collaborate more may also be rewarded for their effort, as they are more likely to advance a bill than the member who introduces it and does nothing to move it through the process. At the same time, I expect

8 diminishing and eventually even negative returns from collaborating too frequently as those members may not be taken as seriously as colleagues who are more judicious in their choices. They may develop a reputation for being someone who will add their name to anything, thereby reducing its value. To examine this relationship between collaboration and legislative effectiveness, I create a measure of propensity for collaboration that reflects the strategic balance I expect members are best served by and rewards those whose collaborative behavior is in proportion to the bills they introduce and close to the chamber norm.

I also expect that members who collaborate with more successful colleagues will be more successful themselves, whether that is the result of learning from those more successful colleagues, making use of the relationships those colleagues have already built, or homophily between effective members. The more successful a member’s collaborators, the more successful that member will be as well. Collaboration may also have a signaling effect. As Dear Colleague letters are used to provide quick, easily digestible summaries of legislation and other policy initiatives, the signatures on the letters can serve as a quick cue for busy members and their staff. If bipartisanship is as valued in the legislative process as staff believe that it is, members who work more frequently with colleagues in the other party should benefit from advertising those relationships within the House. This should be particularly true in the earliest stages of the legislative process when the bipartisan Dear Colleague letter serves as a signal that the policy in question is “safe” to support regardless of party.

For this paper, I use a temporal network autocorrelation model (TNAM) that allows me to examine the relationship between my measures of propensity for policy collaboration and a series of legislative effectiveness measures. I examine the degree to

9 which members over or under perform their expected legislative effectiveness score, as provided by Volden and Wiseman (2014), as well as their ability to attract cosponsors to their legislation, pass bills through the House, and see them signed into law. The analysis reveals that the members who put effort into forming relationships with their colleagues and are moderate in their decision to collaborate see the greatest benefit, although there is also support for the hypothesis that successful members work with other successful members, and limited support for the theory of a signaling effect. The legislative reward for bipartisanship for an individual member of Congress is limited to an increase in the number of cosponsors across all of that member’s legislation, but only if they are in the majority party.

The third paper, “Running from Washington: Policy Collaboration as Symbolic

Representation,” then turns to the electoral benefits of collaboration, considering it as a form of symbolic representation that members use to project a desired image of themselves to their constituents. Specifically, considering the poor approval of

Congress as an institution and the public perception that members are unable to work together to get anything done, members may advertise their collaborative relationships when they are in the district to distinguish themselves from a gridlocked and unpopular legislature. Considering collaboration as a form of symbolic representation also sheds new light into the tension between public preferences for bipartisanship in the aggregate and partisan policy victories at the individual level by considering the electoral fortunes of members who cultivate an image of themselves as being a particularly bipartisan member.

I expect that the electoral vulnerability of a member of Congress will lead to clear differences in both their collaborative behavior and the benefits they reapfrom

10 collaboration. Research by Harbridge and Malhotra (2011) has shown that strong partisans prefer partisanship while weak partisans and independents favor moderation and bipartisanship. Members who represent safe districts therefore have little need to engage in bipartisanship in order to be reelected as they can rely on the votes of strong co-partisans to secure reelection. Safe members may even be harmed by bipartisanship if they are seen as too compromising or if the reputation of their party is poor and collaboration leads voters to see little distinction between the candidates.

However, for electorally vulnerable members, bipartisan collaboration can play an important role in their reelection strategy, helping them differentiate themselves from their colleagues in Washington, DC by presenting an image of a collaborative member who is willing to put aside partisan differences and work for the good of their district and the country.

Using a temporal network autocorrelation model (TNAM) again, this time examining the relationship between a member’s propensity for policy collaboration and their electoral margin, I find that members who were elected by larger margins areless likely to collaborate in general, and particularly less likely to collaborate with members of the other party. For incumbents elected with 60% or more of the vote, this is a safe decision, as they are neither helped nor harmed by collaborating more with members of either party. However, there is a clear benefit for electorally vulnerable members as those who collaborate frequently with colleagues in the other party are able to build a reputation for bipartisanship among their constituents, and therefore see higher electoral returns. The results of this article have concerning implications for the the future of bipartisanship as the redistricting process has resulted in more

11 “safe” districts in which the representatives receive little benefit from collaborating with the other party.

The results of the three essays presented here provide a greater understanding of the role that policy collaboration plays in the modern Congress. I demonstrate that collaboration is more widespread than expected as members routinely seek out other members to coauthor legislation and other policy initiatives. Collaboration allows members to find common ground on policy, even in a highly polarized Congress, as they work with colleagues from the other party and those with whom they share relationships and policy goals. The incentives for collaboration lie in both the legislative and electoral arenas, as members use collaboration to advance their agenda, and as a form of symbolic representation that allows them to create distance between themselves and the “dysfunctional” Congress.

12 Chapter 2: The Room Where it Happens: Collaborative Strategies in the U.S. House of Representatives

In February 2009, Rep. Carolyn Maloney (D-NY) introduced H.R. 847, the James

Zadroga 9/11 Health and Compensation Act of 2010, to provide medical treatment and compensation to first responders who became sick as a result of their exposure to toxins at Ground Zero. Representative Maloney was listed as the sole bill sponsor, but the legislation was the product of several years of work on the part of not only

Representative Maloney, but also Representative Jerrold Nadler (D-NY) and Repre- sentative Peter King (R-NY). These three members described each other as the bill’s

“coauthors” in their floor statements, and they sent out joint press releases and Dear

Colleague letters to promote their legislation, which was signed into law on January

2, 2011 (Maloney, 2010; King, 2010).

How do members of Congress choose whom to collaborate with? In the case of the

James Zadroga Act, the three coauthors were united by their New York constituents, with Representatives Maloney and Nadler representing Manhattan and Representa- tive King representing Long Island. The need to provide care to New York City’s first responders united these three members in a bipartisan collaboration andthey were able to tout their work together to voters, colleagues, and the media. In other

13 cases, the unifying thread in a collaborative relationship may not be so obvious. Rep- resentatives Adam Putnam (R-FL) and Joe Courtney (D-CT) collaborated on H.R.

897, the Long-term Care Retirement and Security Act of 2009, despite being from opposing parties, representing different states, and sitting on different committees.

Their shared interest in promoting long term care may have resulted from a personal interest in the issue, the influence of an outside organization, or similarly affected constituencies.

Members of Congress frequently engage in these collaborative relationships with their colleagues. Collaboration is a widespread practice in the House of Representa- tives as members work together to draft legislation, circulate policy letters, and host events. These relationships go beyond cosponsorship to members actively working together to craft policy and promote their agenda both within and outside of the house. Although these relationships are not easily observed, they are important to our understanding of Congressional behavior and the legislative process. Particularly as Congress becomes increasingly polarized, and it is more difficult for the average member to pass legislation, collaboration has become a key tool for members seeking to advance their agenda. Legislation should not be treated as sole-authored or a reflection of a single member’s preferences when it is the result of unobserved compromises and collaboration intended to create a bill with the broadest appeal. While we may not be able to gain insight into the specific negotiations on collaborative bills that occur behind closed doors, knowing who is in “the room where it happens” is an important first step to understanding why a bill looks the way it does andhow collaboration shapes the legislative process as a whole.

14 Despite the important role that collaboration plays in Congress, from fostering relationships between members to sending signals about both the member and policy in question, we have very little understanding of what the core collaborative process in the House of Representatives look like. In this paper, I provide the first real insights into the collaboration network of the United States Congress using a measure of substantive and purposive policy collaboration. With a unique dataset of Dear

Colleague letters, which are sent by members to promote legislation and other policy initiatives within the House, I identify the members who, like Representatives Mal- oney, Nadler, and King, coauthor bills and policy letters together. I demonstrate that collaboration is a widespread practice in the House of Representatives with members frequently coauthoring legislation and policy letters together. I then examine who collaborates with whom and argue that members of Congress choose their collaborative partners based on strategic considerations, personal relationships, and shared policy goals. Through these collaborations members are able to show a broad base of support for their policy initiatives, strengthen relationships with their colleagues, and cultivate a strong reputation among the audiences who are key to a legislator’s success, from voters to interest groups. As a result, collaborative relationships help members be successful in the contemporary Congress where the traditional avenues of legislative success have been curtailed.

A Collaborative Congress

Members of Congress are undeniably strategic in their decision to collaborate.

Members can collaborate as much or as little as they choose and with an assortment of different colleagues. However, because of the signals that these relationships send

15 there is both a strategy to choosing collaborative partners with the greatest potential pay off and a trade off between different types of collaborative relationships. The

“right” collaborative partner can and does vary depending on the substance of the legislation in question, but the ultimate goal is the same - to find a collaborator who will best help a member advance their agenda. I posit that the decision to collaborate with a colleague will be the result of three considerations: shared policy goals, existing relationships, and strategic advantages.

For members pushing a bill they hope to see signed into law, the ideal collaborator is one from the opposing party. In interviews with congressional staff, senior aides from both parties described their bosses as actively seeking bipartisan collaborators on the legislation they introduced because “[b]ipartisan legislation has a better chance of moving and becoming law” (Interviews, 2016). For members of the minority party, finding a majority party colleague to work with is almost a necessity if they wishto see their legislation considered in the House, to the point that several staff members reported that they had drafted legislation and the necessary supporting materials and then handed it over to a majority party colleague to introduce as the lead sponsor.

As described by the legislative director to a Democratic member, “If [a policy] is legitimately achievable, we’re more likely to give that to a Republican to carry”

(Interviews, 2016). Members of the majority party also find benefits in bipartisan collaboration as the vast majority of bills signed into law are bipartisan. Of the

293 bills signed into law between 2013 and 2014, over half passed on voice vote and every bill received at least some support from Democratic members of the House. For members of both parties the ultimate goal is the same: to attract a broad base of support for a bill that will improve its chances of passage.

16 Bipartisan collaboration also sends a signal to audiences outside of Congress, particularly voters. Voters in the aggregate want their elected representatives to represent the median of their district (Black, 1948; Downs, 1957) and have been shown to pun- ish members of Congress who are too closely aligned with their party rather than the district (Canes-Wrone, Brady and Cogan, 2002). Policy collaboration provides a tool for members to give the appearance of being willing to work across the aisle and moderate than their voting record might reflect. Although it is unlikely that the member’s constituents are aware of many of the bills the member either sponsored or collaborated on, press releases frequently highlight legislation as being bipartisan such as the one issued by Representative Betty McCollum (D-MN) on June 12, 2014 entitled "McCollum Introduces Bipartisan Legislation to Improve Access to Health

Care for Native Americans” which touted the bill she collaborated on with Represen- tative Tom Cole (R-OK) (McCollum, 2014). Social media is also used to highlight collaboration, such as in figure 2.1, which shows a 2014 public Twitter exchange between Representative Kurt Schrader (D-OR) and Representative Reid Ribble (R-WI) highlighting their collaboration on the Biennial Budgeting and Enhanced Oversight

Act. A commitment to bipartisanship is also a common talking point in members’ reelection campaigns, as it was for Representative Ann Kuster (D-NH) who includes

“Working Across the Aisle” as one of eight key issue areas on her campaign web- site1 in which she describes herself as “working across the aisle with Republicans and

Democrats to cut wasteful spending and make our government more efficient” and highlights her role in cofounding the bipartisan United Solutions Caucus.

1kusterforcongress.com

17 Figure 2.1: Promoting Bipartisan Collaboration via Twitter

At the same time, the collaborating members must agree to both the language of the legislation and the Dear Colleague letter. Therefore, on issues that fall along the traditional left-right political spectrum, compromise will be easier if the two members are not too far apart in ideology, as measured by DW-NOMINATE scores. A strategic legislator should find a collaborator who represents a position distinct enough that the bill will garner enough support to pass, but not so different that the bill no longer accomplishes their goal. Minimizing the ideological distance between collaborators will minimize the degree to which the members must compromise their own position.

H1: Members frequently choose collaborators who can provide the broadest base of support for a bill while minimizing compromise.

However, not all bills are introduced with the expectation that they will become law. Members of Congress frequently introduce symbolic legislation to take a position on an issue or stake a claim on a policy area for future Congresses. In these cases, members are more likely to introduce the legislation without any collaborators, or work with colleagues in their own party. Here the strategy differs by party. For

18 members of the majority party there is still an incentive to collaborate within the party and demonstrate to the leadership that there is a broad base of support for the bill within the party. While the bill may not be successful in the Senate or be signed into law, the majority party can pass it in the House and point to the Senate and/or the Administration as obstructionists. For members of the minority party there is little to no hope that these messaging bills will ever be considered by the

House and so there is rarely any incentive to do more than introduce the legislation as pure position-taking.

The exception to this is the member playing the role of party soldier. For members of both parties, championing a policy priority of their party and building a broad base of support among their co-partisans is a way to earn a reputation as a team player, which is key to those with aspirations of institutional advancement. As articulated by

Friedman (1993), “Getting ahead in Congress appears to require certain qualities: a desire to follow congressional norms, a willingness to compromise and build coalitions, and a general understanding of playing the political game.” Prime committee assignments are given to those who are viewed as “responsible legislators,” described as legislators who respect the institution of Congress, are respected by their colleagues, willing to compromise, and generally seen as team players (Masters, 1961). Champi- oning party policy priorities can also be used to build support among the state and local party infrastructure back home, a subset of the voters who are less interested in bipartisan compromise and expect a measure of ideological purity from their elected representatives.

H2: Within-party collaboration is common for members of the majority party and rare for members of the minority party.

19 Once the decision is made to seek a collaborator from either the opposing or same party, the next step is for the member to find a colleague who is willing to work with them on the policy in question. Despite the high value that members place on bipartisan collaboration, interviews with congressional staff repeatedly revealed the belief that it has become more difficult to find collaborators from the other sideof the aisle. The task then is to find some common ground. One aide to a Democratic member recounted how her boss had formed a collaborative relationship with a Re- publican colleague despite “agreeing on virtually nothing.” However, a shared policy priority of their constituents led these two members from neighboring states to collaborate on a series of bills that were important to the voters in their districts. Another congressional staffer described the role that interest groups can play in developing collaborative relationships, bringing members together to work on the group’s policy priorities, as a bipartisan collaboration on the committee of jurisdiction is the best chance an interest group has to be successful in achieving their goals (Interviews,

2016).

Shared policy priorities can take several different forms, but they are a key aspect to the formation of collaborative relationships in Congress. Members frequently find collaborators who represent the same state, sit on the same committee, or co-chair a caucus together They many not have the same views, but to at least some degree they represent the same interests. Collaboration with colleagues from the same state is another way that members can appeal to voters. In this case, the collaboration is intended to deliver projects and policies that the voters care about. Teaming up to collaborate on a bill to address the needs of a prominent home-state industry or benefitting a local constituent group may increase the likelihood of passage ifthe

20 collaboration sends a signal to other members that an issue has regional support.

In some cases it may be a response to a local news story, such as H.R. 5131 in the

110th Congress, the Lance Corporal Jeremy Burris Act, which was introduced by

Representatives Ted Poe (R-TX) and Chet Edwards (D-TX) to make it a federal crime to vandalize the grave of a fallen soldier. Both members are able to promote the legislation to their constituents to demonstrate that they are responsive to local concerns, and because the bills that result from these same state collaborations often have minimal ideological content, the cost of collaboration is lower as members do not have to compromise their positions to agree on the bill language that does not fall onto the traditional left-right ideology scale. (Lee, 2009).

Collaboration with a colleague on the same committee or a caucus leader allows a member to appeal to interest groups by taking a leadership role on a group’s policy priority. Those who are members of a committee with jurisdiction over an interest group’s issue area are in the best position to champion the causes of that group.

Members who sit on the same committee not only have policy expertise in the same issue area, but are more likely to be responsive to constituencies with issues within the committee’s jurisdiction (Miler, 2007). Here we may see two members collaborating on legislation to benefit an interest group that is a substantial contributor to members of the committee, or as Miler argues, because a committee member’s familiarity with an issue makes the associated constituency more salient. In some cases, the interest group may have encouraged the collaboration, proposing the legislation and bringing two members together to work on it. An example of this sort of collaborative effort is

H.R. 6229 in the 111th Congress, a bill to allow states to award grants to local groups working to strengthen student achievement. Sponsored by Representatives Judy Chu

21 (D-CA) and David Loebsack (D-IA), who were both members of the House Education and Labor Committee at the time, the bill was also endorsed by a coalition of seven- teen education interest groups including the National Education Association and the

American Federation of Teachers. If the perception is that collaborative legislation is more successful then members who wish to demonstrate that they are actively working to advance an interest group’s policy priorities should seek out collaborators from the committee of jurisdiction and it is in the interest group’s best interest to foster these collaborative relationships as well.

For members who are not on the relevant committee but wish to build relationships with an interest group or establish themselves as a leader within the House on a policy area, they may be able to do so by taking leadership on an allied caucus. Caucuses allow members to share information and form relationships with colleagues with a common interest in a particular issue (Ringe and Victor, 2013). Members who co- chair caucuses together have chosen to establish themselves as leaders on a particular issue and are therefore more likely to collaborate.2 Whether a major national interest group with an associated PAC, or a coalition of smaller advocacy organizations, there is typically some form of organized interest that benefits from the caucus’s activities and so taking a position as a leader of that caucus and being active in caucus activities appeals to those groups.

H3: Members with shared policy goals who represent similar interests are more likely to collaborate with each other.

2While Ringe and Victor study caucus membership, I distinguish between caucus leadership and membership, focusing solely on the former. Although caucus membership facilitates information sharing, it does not necessarily indicate substantive action on an issue, while leading a caucus demonstrates a commitment to the issue.

22 Finally, personal relationships play an important role in the ability of members to find collaborative partners. Just as members cite their friendships in thecham- ber as a factor in their voting decisions (Kingdon, 1989), it is reasonable to expect that these relationships will also play a role in the drafting of legislation and other policy initiatives. In several different contexts, legislators have been shown to build reputations in their district through their collaborative relationships (Crisp, Kanthak and Leijonhufvud, 2004), their voting decisions are influenced by those with whom they spend time in close proximity (Masket, 2008; Young, 1966), and they repeatedly collaborate with the same colleagues over multiple years (Desmarais et al., 2015).

In some cases, existing relationships may overlap with shared policy goals, such as the members who became friends through their shared committee membership, but here existing relationships are intended to capture the friendships, reputations, and relationships that shape the collaboration network but cannot otherwise be explained.

Personal relationships in Congress are captured primarily through endogenous network characteristics such as triadic closure and preferential attachment. These terms represent not just the nature of the relationships between individual members, but also dyadic dependence in the policy collaboration network as a whole (Hunter et al., 2008). Triadic closure is a common network phenomenon, otherwise known as “a friend of a friend is a friend” (Goodreau, Kitts and Morris, 2009). In the context of the policy collaboration network, this may represent members who have a collaborative relationship with each other recommending additional collaborators to their colleague, or members seeking out collaborators based on their existing shared partners.3 Preferential attachment accounts for the propensity of some members to

3In a small number of cases, it also represents collaborations between more than two members on a single bill, as in the example of the Zadroga Act.

23 collaborate more than others, whether because they develop a reputation through their collaborative relationships as someone who is easy to work with and willing to collaborate, or because they are more likely to seek out collaborative partners.

More broadly, these terms capture the non-independence of members in the policy collaboration network. Although the inability to survey Congress makes it difficult to identify the existing relationships in the chamber, we can examine how the existence of a collaboration between two members affects their relationships with those around them in the network.

H4: Members are more likely to collaborate when there is an existing relationship or reputation.

Dear Colleague Letters in Congress

Obtaining a measure of collaboration in Congress that represents substantive and purposive interaction on the part of members has proved challenging. Contempo- rary members of Congress are generally unable or unwilling to participate in survey research and their staff are only somewhat more accessible. Although members acknowledge their collaborations with their colleagues through floor statements, press releases, and media appearances, these relationships are difficult to quantify ina systematic form. The cosponsorship network has been used as an approximation of collaboration, but cosponsorship does not distinguish between members who actively collaborate on legislation and those who sign on in a brief staff-level email exchange.

With a new dataset of Dear Colleague letters, I am able to directly measure collaboration by identifying the members who work together on legislation and other policy matters through their signatures to these letters. In the contemporary Congress, Dear

24 Figure 2.2: Distribution of Letters by Purpose in Sample Population

20 Percent of Total Percent

Caucus Cosign Cosponsor Floor General Invitation Letter Purpose

Colleague letters are the official correspondence among members and they are widely used to distribute information in the House. Their most common purpose is to solicit cosponsors for legislation, but they are also used to inform members about events, gather signatures on policy letters, and urge members to support or oppose floor actions. Figure 4.2 shows the distribution of letters by purpose for the 111th Congress.4

For a member of Congress seeking to quickly disseminate information in the House, the Dear Colleague letter is the most efficient tool available. Several sample letters are provided in the Appendix, showing a few of the different uses of these letters and the variation in signers.

4Based on a sample of 1600 letters.

25 Paper letters were initially distributed via the House internal mail system, but during the 105th Congress, the Dear Colleague listserv was established, allowing members and their staff to distribute Dear Colleague letters by e-mail. The popularity of the electronic distribution system has grown steadily since its introduction and in 2008, the Chief Administrative Office of the House created a web-based “e-Dear

Colleague” system to further streamline the process. The vast majority of letters are now sent electronically, with 98% of House members using the e-Dear Colleague system in the first session of the 111th Congress (Straus, 2012).

Although there has been little systematic study of Dear Colleague letters due to the difficulty of tracking letters prior to the shift to electronic Dear Colleagues, members have repeatedly cited their use and importance. Representative Daniel

Lipinski describes his use of Dear Colleague letters to gain original cosponsors, to stake a claim on an issue and to alert members to a planned amendment (Lipinski,

2009). In the literature on cosponsorship and legislative effectiveness, Dear Colleague letters are repeatedly mentioned in member and staff interviews as one of the tools by which members build support for their policy proposals (Koger, 2003; Campbell,

1982; Krutz, 2005). A study of interest group endorsements in Dear Colleague letters shows how members use these letters to send cues to their colleagues and gather cosponsors for legislation (Box-Steffensmeier, Christenson and Craig, 2013).

The volume of letters sent through the e-Dear Colleague system also demonstrates the role they play in the legislative process. During the 111th Congress, 31,994 Dear

Colleagues were sent, an average of 112 letters per legislative day.5 The letters are an essential tool for communication in the modern Congress as members and staff rely

5Letters are typically filtered by issue area to the appropriate staff member so the volume isnot unmanageable.

26 on them to know what legislation has been introduced and quickly assess the merits of a bill. As described by a former Legislative Assistant to a House member, “The

Dear Colleague system serves as a way to quickly get the best information about your legislation distributed to congressional offices. It can be a crucial factor in building support for legislation, and to get pertinent facts and figures in front of the staffers who advise the members what they should be supporting”(Interviews, 2013).

When multiple members work together on a piece of legislation or a policy issue, they typically also sign a Dear Colleague letter together based on this work. For example, in March 2009, Representative Brad Miller (D-NC) introduced H.R. 1702, the

Shelter, Land and Urban Management Assistance Act of 2009. He sent a Dear Col- league letter that he signed with Representative David Price (D-NC) asking members to sign onto what they describe as “our bill,” but in the absence of the Dear Colleague letter there would be no way of distinguishing Representative Price from the other

24 cosponsors on H.R. 1702. Here the collaboration is on an issue of importance to

North Carolinians and therefore forms between two members of the state delegation.

In another instance, Representatives Dan Burton (R-IN) and Patrick Kennedy (D-

RI) sent a Dear Colleague opposing a policy against sending condolence letters to the families of service members who commit suicide. The connection between Burton and Kennedy is less obvious, but the letter reveals that it was drafted in response to the suicide of one of Representative Burton’s constituents. As a member of the minority party, he sought a collaborator in the majority and Representative Kennedy was a natural choice as a result of his reputation on mental health issues. Regardless of how the collaboration originates, members signing a Dear Colleague letter together are sending a clear signal that the policy in question is a joint effort. The signers on

27 the letter have coordinated their efforts in a substantive manner and are now sharing credit and responsibility for the policy in question.

Research Design

My data are drawn from a comprehensive archive of the electronic Dear Colleague letters sent between members of the House of Representatives from 108th Congress through the 111th Congress.6 The dataset consists of 82,712 emails sent to the Dear

Colleague listserv over a period of eight years, which provide the complete policy collaboration network for the 108th, 109th, 110th, and 111th Congresses.7 A python script was used to extract the pertinent information from each letter, which included the sending member, the date sent, the subject of the email, any issue areas identified in the subject, the text of the letter itself, and the names of the members who signed each letter.8

Figure 2.3 shows the distribution of the number of signers per letter in the data, which ranges from one to fifteen.9 The average number of signers on a letter is 1.68.

For the 37868 letters with more than one signer, the average number of signatures is

2.49. While the majority of letters are signed by a single member, the collaboration between members is significant, particularly for pairs of legislators. Large scale collaborations of more than four members do occur, but they are far from the norm and

6Corresponding years are 2003 to 2010. 7Although some data is available for 1999-2002, I exclude the 106th and 107th Congresses from my analysis because I do not have confidence that the Dear Colleague letters sent to the electronic listserv represent the whole universe of letters as many members are not represented in the data and offices were in the process of transitioning from paper to electronic letters. 8The analysis of all 82,712 letters is consistent with the results from a sample of 3,017 letters that were hand coded by a team of research assistants. 9For visualization purposes, I combine all four Congresses into a single graph and collapse the 1828 letters with 5-15 signers into a single bin as there are fewer than 1% of all letters in each category.

28 Figure 2.3: Signers per Letter

20 Percent of Letters Percent

1 2 3 4 5+ Number of Signers per Letter

still relatively constrained in size. In comparison, for the 4867 bills introduced in the

House during the 111th Congress with at least one cosponsor, the average number of cosponsors is 22 and the range is from 1 to 425. The likelihood that a member has meaningful interactions with all 22 of the cosponsors on her bill is low in comparison to the likelihood that she has a relationship with the single colleague who cosigns a

Dear Colleague with her.

These signatures are used to construct a series of affiliation networks of members and the letters that they sign onto. In two-mode form, the events in each network are the portion of the 82,712 letters sent during that Congress and the actors are the

29 438-447 members who served in that Congress.10 In this form, the members are not

directly connected to one another, but through their common letters. However, my

primary interest is in the relationships between the members and it is therefore appro-

priate to flatten each network into a one-mode projection where the nodes represent

the members and the edges are undirected ties indicating that two members have

signed a Dear Colleague letter together in that Congress (Breiger, 1974; Wasserman

and Faust, 1994).11

Figure 2.4 shows a map of the member-level projections of the affiliation net-

works, with each network N t representing the whole collaboration network for a

given Congress.12 At t = 1, which corresponds to the 108th Congress (2003-2004),

1 I observe 439 members of Congress (Nv = 439) and 4750 collaborative relationships

1 between them (Ne = 4750). At t = 2, which corresponds to the 109th Congress

2 (2005-2006), I observe 438 members of Congress (Nv = 438) and 5725 collaborative

2 relationships between them (Ne = 5725). At t = 3, which corresponds to the 110th

3 Congress (2007-2008), I observe 447 members of Congress (Nv = 447) and 4809 col-

3 laborative relationships between them (Ne = 4809). At t = 4, which corresponds

4 to the 111th Congress (2009-2010), I observe 443 members of Congress (Nv = 443)

10The number of members in the data exceeds the number of members in the House of Representa- tives due to districts where the member was replaced mid-Congress. All members who cast a vote in each Congress are included, regardless of whether they signed a Dear Colleague letter. 11In future work, I intend to identify the “primary” signer of each letter and consider the network as directed to gain a better understanding of which members lead collaborations and which members are approached for collaboration, however it is difficult to distinguish between letters where there is a clear “first author” and those where all collaborators contributed equally so for the timebeing I treat all relationships in the network as equal. 12For visualization purposes, the isolates are excluded from the network graphs, although they are included in the analysis. There were 17 isolates in the 108th Congress, 18 isolates in the 109th Congress, 7 isolates in the 110th Congress, and 7 isolates in the 111th Congress. The majority of the isolates are members who did not serve a full term.

30 Figure 2.4: One-Mode Projections of the Member Networks

4 and 4493 collaborative relationships between them (Ne = 4493). Across all four networks we see that the structure is consistent, with every collaborating member tied to another in one large component, indicating that the policy collaboration network

31 is well-connected. We also see a clear tendency towards bipartisanship as members

of both parties are mixed together in the network, although some Congresses, such

as the 110th (t=3), are more bipartisan than others, such as the 109th (t=2).

t t The adjacency matrix for each Congress, A , is binarized so that Ai,j is equal to 1 if nodes i and j signed any Dear Colleague letter together and 0 if they do not.13

As they are projections of two-mode networks, the resulting graphs are particularly dense which limit their usefulness, but it is worth noting two interesting features of the network. First, aside from the isolates in each Congress, all of the members in each Congress are connected in one giant component, indicating that collaboration is widespread in Congress and members are well-connected to each other in the policy network. The average member collaborated with 20 of her colleagues during the 111th

Congress. Second, party attributes were added to the network so that the red nodes represent Republican members and the blue nodes represent Democratic members.

While there are obvious partisan clusters in the networks, overall they represent bipartisan networks where there are frequent collaborations between Democratic and

Republican members of Congress.

A closer look at the partisanship of the edges between members confirms that this is a network dominated by bipartisan ties. Figure 2.5 shows the breakdown of ties for each Congress by partisanship. In the 111th Congress, of the 4493 ties between members, 46% of them are between a Democrat and a Republican. 44% of the ties are between two Democrats and the remaining 10% of the ties are between two

13I do lose data by thresholding the adjacency matrix to represent the presence of any tie rather than weighting it by the number of letters signed together, however most instances in which members have multiple collaborations in a Congress represent multiple letters sent on the same bill or policy initiative rather than unique collaborations. Future work will disentangle the two and examine the strength of ties between members.

32 Republicans. As expected, bipartisan ties are a frequent occurrence across all four

Congresses, as well as collaborations between Democratic co-partisans. Somewhat surprisingly, within party collaboration does not appear to reflect whether a party is in the majority or minority as the partisanship of ties does not flip when control of the House of Representatives shifts from the Republican party to the Demo- cratic party between the 109th and 110th Congresses. Instead, the data reveal that

Democrats collaborate within their own party far more frequently than Republicans regardless of which party controls the chamber. One possible explanation for this phenomenon is the differing policy goals of the two parties, as Democrats seekto solve problems through new social policies, while Republicans maintain their ideological commitment to conservatism and limited government (Grossmann and Hopkins,

2016). The simple result is that Democrats propose more policy initiatives and therefore more frequently collaborate with each other. Alternatively, Republicans may be more strategic in their collaborative decisions, which is suggested by the marked decline in Republican-Republican collaborations from the 109th to 110th Congress and corresponding increase in bipartisan collaborations. A strategic member who suddenly finds herself in the minority party would find a greater benefit inforging new relationships with members of the majority party rather than maintaining their ties to their minority party colleagues.

My collaboration hypotheses are tested using an temporal exponential random graph model (TERGM) to identify the characteristics of the network that best ex- plain its formation. While the ERGM estimates the probability of observing a given network N, the TERGM is an extension for longitudinal networks observed in t discrete time periods (Hanneke, Fu and Xing, 2010). The advantage of the TERGM

33 Figure 2.5: Partisanship of Ties by Congress

0.5

0.4

0.3

0.2 Percent

0.1

0.0 108th 109th 110th 111th Congress

Bipartisan Democratic Republican

model is that it does not require collaborative relationships to be treated as indepen-

dent dyads when they are the result of an inherently relational process, but instead

my observations consist of four networks that reflect the whole universe of policy

collaboration in each Congress. It also allows me to incorporate both endogenous

network dependencies and exogenous covariates, which are represented in the equa-

tion as Γ(N). The probability of observing the network at some discrete period of observation (N t) is then:

exp(θ′ Γ(N t,N t−1, ..., N t−K )) P (N t|N t−K , ..., N t−1, θ) = c(θ,N t−K , ..., N t−1)

The model yields a set of coefficients that estimate the likelihood of collaboration

between members in the network based on both member attributes and the structural

34 features of the network. This makes it ideal to test my hypotheses of members

collaborating with colleagues as a result of strategic considerations, shared policy

goals, and personal relationships.14

The strategic considerations hypotheses are tested with a node-mix term that

creates separate covariates for Bipartisan, representing the likelihood of tie formation between members of opposing parties, and Majority-Majority, representing the likelihood of tie formation between two members of the majority party in a given

Congress. Collaboration between minority party members is the excluded reference category. For the bipartisan collaboration hypothesis, I expect to see a positive tendency towards tie formation between members of opposing parties, representing the tendency to use collaboration to demonstrate a broad base of support for the legislation. For the co-partisan hypothesis, I expect to see a positive tendency towards tie formation between majority party co-partisans, representing the greater incentives for within-party collaboration among majority party members when compared to minority party members.15 First and second dimension DW-NOMINATE scores are both included in the model in absolute difference terms to estimate the probability of tie formation as members move further apart on either ideology scale. I expect the effect of both to be negative, with most of the effect being on the first dimension, representing the traditional liberal to conservative scale. The second dimension has less

14Recent work by Box-Steffensmeier, Christenson and Morgan (Forthcoming) demonstrates there- sulting bias when unobserved heterogeneity is not accounted for in an exponential random graph model and the authors propose a solution in the form of incorporating a frailty term into the model. I do not do so here as the TERGM uses maximum pseudolikelihood estimation with bootstrapped confidence intervals which deals with bias in the estimates, which will be validated in futurework. 15I also estimated the model using a node-mix term for party rather than majority-minority status, the results of which are reported in the appendix.

35 explanatory power post-1980, but is included as it captures divisions within the par-

ties, such as the split between establishment and Tea Party Republicans that starts to

appear in the 111th Congress (Hare and Poole, 2016). Finally, I include a covariate for

Electoral Security, representing the percentage of the vote that each member received

in the previous election as recorded by the Federal Elections Commission.16 This is

a node-level covariate, reflecting the likelihood of forming collaborative relationships

for each individual member and as such I expect it will be negative. Members who are

vulnerable to defeat in their next election should form more collaborative relation-

ships, particularly bipartisan relationships, as it allows them to present themselves

as ideological moderates and claim credit for a greater number of policy initiatives.

Endogenous network effects are uniquely suited to capture existing relationships

in the network as they measure the probability of tie formation based on the other

relationships in the network. I include the geometrically weighted edgewise shared

partner (GWESP) statistic in the model, which captures the tendency towards tie

formation between members who have other shared partners, and the geometrically

weighted dyadwise shared partner (GWDSP) statistic, which captures the tendency

for two nodes to collaborate with the same shared partner (Hunter, 2007; Hunter

et al., 2008). When included in a model together, the GWESP statistic is a measure

of Triadic Closure, where a tie from i to j and a tie from j to k increases the likelihood of a tie from i to k, while the GWDSP is a measure of Structural Imbalance, where i and k are both tied to j, but not to each other (Wimmer and Lewis, 2010). Figure

2.6 provides a visual representation of these relationships. By including a degree weighting parameter (α), higher order triangles and k-stars have less influence on the

16For run-off elections, I recorded the vote received in the final contest.

36 Figure 2.6: GWESP and GWDSP statistics

probability of tie formation. In other words, the first shared partner between member i and member k is going to have a greater effect on the likelihood of tie formation between i and k than the fourth shared partner (Snijders et al., 2006). I expect there will be a positive tendency towards triadic closure in the network, and a negative tendency towards structural imbalance.

Preferential Attachment, or the tendency to form ties with nodes that are particularly popular or sociable in the network, is captured by the geometrically weighted degree (GWD) statistic. It estimates the likelihood of tie formation as the degree of a node increases, and similarly to the GWESP and GWDSP statistics, it is weighted so that high degree nodes do not have a disproportionate effect (Levy, 2016). More

37 simply, it will be easier for a member with thirty collaborators to form a relationship with a thirty-first member than it will be for a member with one collaborator toform a relationship with a second. I expect this will also have a positive effect, capturing the reputation effect in the network.17 Finally, I include two exogenous covariates to capture the length of time that a member has served in the House, which reflects how much opportunity they have had to form collaborative relationships. Same Class is an exogenous nodematch statistic that uses the first Congress served for each member (drawn from the Congressional Bills Project (Adler and Wilkerson, 2003-2012)) and estimates the probability of tie formation between two members who first served in the same Congress. I expect this covariate to have a positive effect as members who were elected in the same class may form relationships that persist through their tenure. At the same time, I expect that First Term, an indicator variable for whether a member is serving their first term in Congress, will have a negative effect asfirst term members have had less time to build relationships with potential collaborators.

Shared policy goals are captured by a series of exogenous covariates. Same State is a nodematch term that uses each member’s state ICPSR code (drawn from the

Congressional Bills Project (Adler and Wilkerson, 2003-2012)) and estimates the probability of tie formation between two members with matching state codes. I expect that this will have a positive effect on collaboration as members who represent the same state are more likely to find common ground on issues of local and

17In addition to using the endogenous measure for popularity, it is worth considering precisely what leads some members to be more popular in the policy collaboration network. The most obvious connection is to legislative effectiveness, as it is reasonable to expect that members who are more effective will also be more desired collaborators. However, incorporating a term for each member’s legislative effectiveness in the previous Congress creates a new issue in that it results inallfirst term members being excluded from the network, creating a selection bias issue. This will be addressed in more detail in subsequent versions of this paper.

38 regional importance. Same Committee is a set of four adjacency matrices for the

108th through 111th Congresses constructed using Stewart and Woon’s congressional

committee data (Stewart and Woon, 2011). For each of the members of the House

of Representatives in each Congress, the edge between member i and member j is coded as 1 if they served on a committee together and 0 if they did not. The matrices are then included in the model as an edge covariate which I expect will be positive, indicating that members who serve on the same committee together are more likely to collaborate with each other. Caucus Leader is a similarly constructed set of adjacency matrices using Victor’s U.S. Congress caucus leadership data (Victor, 2013). An edge is coded as 1 if members i and j co-chaired a caucus together in a given Congress and 0 if they did not. Again, I expect that the coefficient for the edge covariate will be positive, indicating that members who co-chair a caucus together are more likely to collaborate with each other as they have positioned themselves as leaders in the caucus’s area of focus.

I also expect that members of the same racial, ethnic or gender minority groups will have shared policy interests and include several terms to account for these com- monalities. Race, ethnicity, and gender covariates were coded using data from the

Congressional Research Service and the U.S. House Historian and Clerk (Office of the

Historian and Office of the Clerk, 2013; Manning and Shogan, 2012; Manning and

Brudnick, 2014; Tong, 2013). The race/ethnicity covariate is coded 1 for black members, 2 for hispanic and latino members, 3 for asian members, and 0 for all others.

An set of adjacency matrices is then created to indicate when two members are both from the same racial or ethnic minority group in each Congress. An edge between a black member i and a black member j is coded as 1, while edges between two white

39 members or a black member and a hispanic member are coded as 0. I expect that

Same Minority will have a positive effect on the likelihood of a tie between two members. Gender is a binary covariate where 1 indicates a female member and 0 indicates a male. It is included in the model using a node-mix term so that it estimates the probability of tie formation between members of the opposite gender, Opposite Gen- der, and between two female members Female-Female.18 I expect that women will be more likely to collaborate with other women and that ties will be less likely to form between members of the opposite gender.

Finally, I include a series of member-level controls to account for other influences on the decision to collaborate with a colleague. Delegation Size accounts for the size of each member’s state delegation, with the expectation that members from larger states will have a larger pool of members to work with on issues of importance to their state and therefore more opportunities for collaboration. Party Leader is an indicator variable representing whether a member is in the elected leadership of their party or the chair of a committee or subcommittee. I expect a negative effect, as members in the leadership are able to gain attention for their policy priorities on their name alone. I also control for Bills Introduced as members who introduce more legislation should have more opportunities to collaborate and therefore I expect this will have a positive effect on tie formation.

The family of exponential random graph models treats each full network as a single observation and estimates the probability of observing that network, based on the model covariates, out of all possible permutations of the network. As “all possible permutations of the network” very quickly becomes computationally impossible with

18The probability of a tie between two male members is the excluded reference category.

40 more than a handful of nodes, the model is best estimated using maximum pseu-

dolikelihood (MPLE), which converges in probability to the maximum likelihood as

the size of the network increases. However, because the MPLE underestimates the

variance of its estimates, I use bootstrapped confidence intervals to obtain a reliable

estimate of model uncertainty (Desmarais and Cranmer, 2012).

Analysis

Model results are shown in Table 2.1. The model was estimated using the xergm

package (Leifeld and Cranmer, 2015) with coefficients representing the conditional

log-odds of a tie between two nodes as a result of the covariate and 95% confidence

intervals based on 1,000 bootstrap iterations. Three of my four hypotheses are sup-

ported by the model results. Looking first at the strategic considerations hypotheses,

I find that three of the five coefficients yield significant results in the expecteddirec-

tion, providing strong support for my first hypothesis, but no support for my second.

Members of the opposite party are significantly more likely to collaborate than co-

partisans, indicating that members of both parties find an advantage in reaching

across the aisle to find a collaborator for a bill or policy initiative so that theycan

claim it is a bipartisan effort. The ideology coefficients are also significant andin

the expected directly, indicating that even as members seek out bipartisan collab-

orators, they also try to minimize the ideological distance between themselves and

their collaborators so that they also minimize the amount of compromise that must

be done.

Contrary to my expectations, neither Majority-Majority or Electoral Security are significant in the model. Although I expect that members of the majority party

41 would find particular benefit in collaborating with their co-partisans to buildtheir relationships within their own party, as shown in figure 2.5, within-party collaboration follows a partisan trend rather than a majority-minority one. Democrats collaborate with co-partisans more frequently than Republicans regardless of which party is in control of Congress.19 I also find that a member’s electoral security has no bearing on their likelihood to form collaborative relationships with their colleagues. Members from both safe and vulnerable districts appear to find benefit in collaborating with their colleagues.

Four of the five coefficients used to test the personal relationships hypothesis yield the expected results. There is a definite tendency towards triadic closure in the policy collaboration network, indicating that when controlling for the exogenous covariates included as well as the other structural terms, members of Congress are more likely to form collaborations with colleagues when they share collaborative partners. If members i and k both collaborate with member j, then a collaboration is significantly more likely to form between them, with the probability increasing the more shared partners they have. This likely accounts for collaborations that form as a result of both existing collaborative relationships and unobserved personal relationships. Members i and k may collaborate because their previous work with j made them realize they would be well-suited to a collaboration, or because i, j, and k all have a personal relationship not otherwise captured in the model. Structural Imbal- ance is both significant and positive, indicating that there are more instances where members i and k are connected to member j but not to each other than expected

19This is supported by estimating the model with a partisan mixing term rather than a majority- minority mixing term. The results, reported in the appendix, show a positive and significant result for the likelihood of tie formation between two Democratic members, although the result is not as strong as what is reported for bipartisan collaboration.

42 Table 2.1: Probability of Policy Collaboration

Estimate 2.5% 97.5% Strategic Considerations Bipartisan 0.571∗ 0.337 1.032 Majority-Majority -0.046 -0.096 0.283 1st DW-NOMINATE -0.484∗ -0.602 -0.365 2nd DW-NOMINATE -0.187∗ -0.232 -0.167 Electoral Security -0.001 -0.003 0.001

Personal Relationships Triadic Closure 0.445∗ 0.377 0.529 Structural Imbalance 0.030∗ 0.025 0.042 Preferential Attachment 2.648∗ 2.132 3.554 Same Class 0.155∗ 0.138 0.192 First Term -0.080∗ -0.210 -0.002

Shared Policy Goals Same State 1.119∗ 1.033 1.201 Same Committee 0.566∗ 0.546 0.587 Caucus Leader 2.508∗ 2.165 2.884 Same Minority 0.866∗ 0.698 0.949 Female-Female 0.322∗ 0.136 0.588

Controls Delegation Size -0.004∗ -0.005 -0.004 Party Leader -0.002 -0.049 0.049 Bills Introduced -0.002∗ -0.002 -0.001 Opposite Gender -0.142∗ -0.208 -0.083 Edges -8.945∗ -10.716 -8.068 ∗ 0 outside the confidence interval

43 (Leifeld and Schneider, 2012). My expectation was that it would have a negative

effect as I expect members with shared partners to generally collaborate witheach

other, however the effect is small in magnitude. Preferential Attachment has a positive and significant effect on tie formation indicating that members who collaborate more frequently with colleagues, either because they are more prone to reach out to other members (sociability) or because they develop a reputation as an appealing collaborator (popularity) are more likely to attract additional collaborators as a result of their reputation. Same Class and First Term both yield small but significant effects

in the expected directions, providing more support for my hypothesis that personal

relationships influence the decision to collaborate with a colleague. Members who

were elected in the same class are more likely to collaborate with one another, sug-

gesting the relationships formed in the beginning of their tenure persist as they stay

in Congress. At the same time, members in their first term are slightly less likely to

form collaborative relationships as they do not have the robust personal network that

more senior colleagues have cultivated.

The shared policy goals hypothesis is strongly supported by all five coefficients

included in the model. As expected, members from the same state, who sit on the

same committee, and who co-chair caucuses together are significantly more likely

to collaborate, indicating that these relationships are key to members seeking to find

common ground with their colleagues in Congress. Members from the same state work

together on issues of local and regional importance, while members who sit on the

same committee and co-chair caucuses together work on issues within the jurisdiction

of their committee or the policy priorities of the caucus.20 Members of the same racial,

20The size of the caucus leader effect is partly attributable to the Dear Colleague letters sentouton behalf of the caucus to disseminate information or recruit additional caucus members. However,

44 Figure 2.7: Probability of Tie Formation Coefficients

Bipartisan ● Majority−Majority ● 1st DW−NOMINATE ● 2nd DW−NOMINATE ● Electoral Security ● Triadic Closure ● Structural Imbalance ● Preferential Attachment ● Same Class ● First Term ● Same State ● Same Committee ● Caucus Leader ● Same Minority ● Female−Female ● Delegation Size ● Party Leader ● Bills Introduced ● Opposite Gender ●

−1 0 1 2 3 4

Bars denote CIs.

ethnic, and gender minority groups are also more likely to collaborate with each other,

as evidenced by the positive effects of Same Minority and Same Gender:F.

To better illustrate the degree to which covariates explains tie formation in the policy collaboration network, Figure 2.7 plots the model coefficients using the texreg

caucus letters represent only 2% of the Dear Colleague letters sent, so at least some of these collaborations between caucus leaders are on other issues.

45 package in R (Leifeld, 2013). Significant coefficients are printed in black, while the

three insignificant coefficients are in gray. Here we can more directly see themag-

nitude of the coefficients as they effect the probability of tie formation in thepolicy

collaboration network. Preferential Attachment and Caucus Leader have the largest

effect on the structure of the network, highlighting the importance of endogenous

network relations in explaining policy collaboration. There are also clear positive

effects for Bipartisan, Triadic Closure, Same State, Same Committee, and Same Mi-

nority, indicating that these terms have a significant influence on the overall network structure. 1st DW-NOMINATE and 2nd DW-NOMINATE have significant negative

effects, indicating that these covariates discourage tie formation in the network.

As hypothesized, strategic considerations, shared policy goals and personal rela-

tionships all play a meaningful role in the decision to collaborate with a colleague.

Members seek collaborators from the opposing party when they wish to demonstrate

a broad base of support for a bill and from their own party when they are champi-

oning a party policy priority. Once that decision is made, members turn to colleagues

with whom they have shared policy priorities and personal relationships to identify

individual collaborative partners who are willing to work with them.

These collaborative relationships are important not only for the signals that they

send about the underlying legislation, but also because members can and do use them

to send signals to different audiences within and outside of the House. Bipartisan col-

laboration can be used to signal to voters that a member is a moderate, to colleagues

in the chamber that a member is willing to compromise, and to interest groups that

a member is prioritizing their policy priorities. Within party collaboration can signal

to colleagues and the party leadership that a member is a team player and willing

46 Figure 2.8: Goodness-of-fit Diagnostics

to push the policy priorities of the party. Collaborating with colleagues from the same state can demonstrate that the member is working on the policy priorities of his constituents, while collaborating with colleagues on the same committee can be used to show support to interest groups with business before the committee.

As previously discussed, the temporal exponential random graph model uses maximum pseudolikelihood (MPLE) to derive consistent coefficient estimates and a bootstrap sample of MPLEs to estimate 95% confidence intervals. The best assessment

47 of the model’s fit is to compare summary statistics for 100 simulated networks from each Congress against those of the actual networks. Figure 2.8 shows the goodness of fit plots for the dyad-wise shared partners, edge-wise shared partners, degree, in- degree, geodesic distance, and modularity.21 The solid line represents the statistics for the observed networks, while the grey line is a series of boxplots reflecting the range of values from the sampled networks. An ideal fit is one in which the statistics for the observed networks are statistically indistinguishable from the simulated networks. The simulated networks closely approximate the observed networks for dyad-wise shared partners, edge-wise shared partners, and geodesic distances. While the simulated networks are not a perfect fit on the degree statistic, there are only six points where there is a statistically significant difference between the simulations and the observed networks. As a result, I am confident that my model is a good fit for the observed networks.

Discussion

Members of Congress frequently collaborate with colleagues in the legislature, but until now, we have had very little understanding of what these relationships look like and how they form. I present a theory of policy collaboration in which the decision to collaborate with a colleague is the result of three factors: strategic considerations, shared policy goals, and existing relationships. These collaborative relationships are important because they allow members of Congress to forge partnerships within the legislature that help them advance their agenda in a polarized and gridlocked legislature. Using a unique dataset of congressional Dear Colleague letters that provide

21As these are undirected networks, the degree and in-degree statistics are the same for both the actual networks and the simulated versions.

48 a measure of purposive collaboration and an exponential random graph model that explains network structure using both endogenous and exogenous covariates, I find strong support for three of my four hypotheses.

Members are undeniably strategic when deciding who to collaborate with on policy. Bipartisan collaboration is seen as the “gold standard” that members strive towards when seeking collaborative partners. Of course it is not always possible to find a collaborator from the other party, particularly on social policy, but members are conscious of the fact that legislation with bipartisan support is more likely to be signed into law and when a bill or policy is bipartisan from its initial development, it will be easier to obtain support from both parties (Interviews, 2016). At the same time, collaboration necessarily requires some degree of compromise as the two (or more) members who are collaborating on a bill or policy must agree on the underlying language as well as the promotional materials. These agreements are likely easier to reach when the members are closer on the ideological spectrum and therefore we see a decreased likelihood of collaboration the further apart two members are ideologically.

However even bills with bipartisan support can languish in committee. As it becomes increasingly difficult for members to pass legislation, they have to find other ways to demonstrate that they are working on behalf of their constituents and other interested parties and collaboration has become a tool by which they can accomplish this. As explained by one congressional staff member, public opinion of Congress is so poor that even if a member cannot get a single bill passed under their own name, they can claim credit and build their reputation just by working towards an audience’s priorities (Interviews, 2016). Seeking collaborators of any kind sends a

49 clear signal to the interested audience that the member has prioritized their desired policy, whether it is a within-state collaboration on behalf of local constituents, a bipartisan committee effort on behalf of an interest group, or a collaboration with co-partisans on a party platform bill.

The prevalence of collaboration in Congress and the nuances of who collaborates with whom have important implications for how we study both congressional behavior and the legislative process. As demonstrated here, personal relationships and reputations play a significant role in the formation of collaborative relationships between members. In an ideal world we would be able to survey members of Congress and obtain a complete picture of who members consider to be their friends and allies in the House. In the absence of such data, treating Congress as a network rather than independent members or dyads allows us to account for these unobserved relationships by modeling the endogenous network structure. Furthermore, estimating the likelihood of collaboration in the network without accounting for these endogenous network features yields a model that is a poor fit for our observed data and as a result significantly overstates the effect size for the partisan and ideological covariates.

Members are generally assumed to introduce legislation that reflects their own preferences, but when collaboration between multiple sponsors is obscured by the

House rules allowing for only a single sponsor per bill, we miss the compromise that occurs before the bill is even introduced. Many bills represent not just the priorities and goals of a single sponsor, but of a small group of legislators who spent time working out the details of the legislation and finding common ground while drafting the bill. When Representative Maloney brought Representatives Nadler and King on board the James Zadroga 9/11 Health and Compensation Act of 2010, the bill

50 had to be written in a way that all three members could support. For Representative

Maloney, whatever compromises had to be made to appease Representative King were worth the benefit of being able to claim that the bill was a bipartisan effort. Andin this case, their effort paid off as the bill finally passed the House and was signed into law after several years of work.

51 Chapter 3: Lone Wolves and Team Players: Policy Collaboration Networks and Legislative Effectiveness in the House of Representatives

Congresswoman Ileana Ros-Lehtinen (R-FL) was one of the most effective Repub- lican members of the 110th Congress (Volden and Wiseman, 2014). Where many of her Republican colleagues failed to maintain their effectiveness when the Republicans lost control of the House of Representatives, Congresswoman Ros-Lehtinen was the second most effective Republican, significantly outperforming the expectations fora member with her experience, position, and minority party status. She introduced twenty-five bills, six of which passed the House of Representatives, and four ofwhich were signed into law. Compared to the 2.8% success rate for all bills introduced in the House during the 110th Congress, Congresswoman Ros-Lehtinen’s record is particularly impressive (Office of the Clerk, 2014). She was also one of themost collaborative members of the 110th Congress, working with 74 of her colleagues on legislation, policy letters, and events, 68% of which were with Democratic members of the House.

Collaborative relationships such as the ones formed by Congresswoman Ros-Lehtinen and her colleagues are a common occurrence in the House of Representatives, yet they have received limited attention from political scientists. Much of the literature

52 on Congress treats members as independent actors and while recent scholarship has paid more attention to the connections between members, attempts to describe and estimate the impact of these relationships have been largely focused on cosponsorship and co-voting (e.g. Cranmer and Desmarais, 2011; Kirkland, 2011; Rogowski and Sinclair, 2012; Craig et al., 2015). Matthews and Stimson (1975) and Kingdon

(1989) were two early scholars who recognized the relationships between members in the House, highlighting the value that members place on cues from their colleagues when making voting decisions. Fowler (2006a) sparked a series of network studies of

Congress with his research that uses shared cosponsorships to create a measure of

“connectedness” for members. But where previous studies have used measures like cosponsorship as proxies for policy collaboration, I leverage a new dataset of Dear

Colleague letters to create the first measure of substantive collaboration between members of the U.S. House of Representatives based on the decision to coauthor and promote policy initiatives together in the formative stages of the policy process.

With this measure, I examine how members of Congress benefit from collaborating with their colleagues. Between 2003 and 2010, 46% of the 82,712 Dear Colleague letters sent in the U.S. House of Representatives were signed by more than one member, indicating that policy collaboration is a widespread practice. Yet it is not immediately apparent what the benefit is to the individual member. Collaboration takes time and effort as members must first agree to work with each other and thenagree on the substance of the policy and the language of the Dear Colleague letter. What is the pay off in return for this effort? Is collaboration rewarded in Congress? Are members who form more relationships with their colleagues more effective legislators? Does it matter who a member collaborates with? By studying the relationship

53 between policy collaboration and legislative effectiveness, I provide a greater under-

standing of how collaboration between members fits into one of their three primary

goals: creating good public policy (Fenno, 1973a).22 I argue that different aspects of

collaboration matter at different stages of the legislative process, from members who

are able to increase their overall legislative effectiveness by collaborating with other

highly effective members, to the value of bipartisanship in building early support for

policy initiatives. The result is a greater understanding of why some members are

more effective than others and new insights into the collaborative nature of Congress

itself.

Collaboration in Congress

The cosponsorship network is the most commonly used measure of collaboration

in Congress. In this network members are connected when member i cosponsors a bill

introduced by member j. Because cosponsorship is a widespread practice, this yields a large, dense network in which a third of the relationships in the House are direct ties from one member to another and no legislators are more than four connections away from any other member (Fowler, 2006b). The cosponsorship studies find that

members who are more connected to their colleagues pass more amendments and

garner more votes on their legislation (Fowler, 2006a), and that Congress represents

a “small world” network that passes more important legislation when the members

are more closely tied to each other (Cho and Fowler, 2010).

22As Fenno (1973a) articulated, members of Congress are largely driven by three key goals: reelection, institutional advancement, and good public policy.

54 Fowler (2006a) also distinguishes between what he calls “active” and “passive” cosponsors. He defines active cosponsors as those who help write or promote legislation in some way, while passive cosponsors do nothing more than add their names to the bill. Hundreds of members can sign onto the same bill with no interaction or discussion between the cosponsor and bill sponsor beyond a one sentence email from one office to another. Yet without a means to distinguish between active and passive cosponsorship, these are treated as equivalent relationships in the network.

If a member who is more connected within the cosponsorship network is a more successful legislator, it is reasonable to expect that this effect will be stronger once those members who actively collaborate on legislation are distinguished from those who simply agree on policy.

Kirkland and Gross (2012) argue that a single cosponsorship between two members in a given Congress is not necessarily indicative of a collaborative relationship and instead disaggregate the cosponsorship network into weeks of activity. They claim that not only is the relationship between bill sponsor and cosponsor a collaborative one, but also that members who cosponsor the same piece of legislation within a five day window are collaborating with each other. Again, I argue this is indicative of policy agreement rather than policy collaboration, as two members who cosponsor the same bill are unlikely to have any interaction regarding that legislation regardless of whether they cosponsored a bill in the same week. Other scholars have opera- tionalized collaboration as having cosponsored multiple bills together (Oleszek, 2010;

Rippere, 2016), which may indicate a relationship, but more likely represents two members with similar policy agendas. Desmarais et al. (2015) argue that “the major

55 shortcoming of extant measures of legislative networks is that overlap between legislators may be driven by correlated preferences (e.g., for legislation or policy areas) and/or institutional forces (e.g., the committee assignment process) and may not be indicative of active collaboration between or among legislators” [p.44]. Instead they use a novel dataset of Senate press events to measure collaboration based on which members promote legislation together and find that collaborating on a press event is a significant predictor of co-voting in the Senate.

Regardless of how ties in the cosponsorship network are interpreted, it is clear from the existing literature that relationships in legislatures matter. In addition to cosponsorship and participation in Senate press events, social relationships, connections between staff, and physical proximity are all associated with co-voting (Arnold,

Deen and Patterson, 2000; Peoples, 2008; Ringe, Victor and Gross, 2009; Young,

1966; Masket, 2008). Legislators who interact with each other also influence each other, and members of Congress are strategic in their decision of whom to collaborate with, seeking out collaborators they believe will increase the likelihood of legislative success, but no work to date has examined the utility of collaboration to the individual member (Craig, 2016).23 I argue that members of Congress who put in the effort to form collaborative relationships with their colleagues are more successful legislators because they will have stronger and more strategic relationships to draw on when trying to advance legislation. As described by one member of the U.S. Senate,

“...if the other senators know and like you, it increases your effectiveness” (Matthews,

1959, p.1066).

23Future work will examine the benefits of collaboration at the bill level rather than themember level.

56 Legislative Effectiveness

The vast majority of bills introduced in Congress are never signed into law. Of the 13,683 bills and resolutions introduced in the 111th Congress, only 1892 (13.8%) passed a vote on the House floor and 383 (2.8%) were signed into law. The content of the legislation matters, but there is also evidence that some members are simply more effective at advancing their legislative agenda (Adler and Wilkerson, 2005). Whydoes

Congresswoman Maxine Waters (D-CA) pass seven of the 26 bills she introduced in the 111th Congress, while Congresswoman Lucille Roybal-Allard passes none? Both women are California Democrats with similar experience in the House, serving on influential committees, yet Congresswoman Waters is far more successful by standard metrics of legislative effectiveness.24

Much of the legislative effectiveness research has focused on factors largely out of the member’s direct control such as seniority, party, and committee leadership positions. Majority party members, committee chairs, and more senior members all have more success in advancing their legislative agenda (Anderson, Box-Steffensmeier and Sinclair-Chapman, 2003; Krutz, 2005; Volden and Wiseman, 2014). But there is also evidence that the actions a member takes can affect their own legislative success.

Matthews (1959) provides several quotes from senators emphasizing that members should be careful not to speak too often or on issues they are not as knowledgeable on. This is reinforced by Anderson, Box-Steffensmeier and Sinclair-Chapman (2003) who find that members who take a balanced approach to legislative activities suchas bill cosponsorship and floor speeches are more likely to be successful. Members who

24Congresswoman Waters was first elected in 1990 and sits on the Financial Services Committee. Congresswoman Roybal-Allard was first elected in 1992 and sits on the Appropriations Committee.

57 gather more cosponsors and speak on the floor are more successful (Krutz, 2005) as are those who are more loyal to their party (Hasecke and Mycoff, 2007).

Members who work hard to advance their legislation get results. I expect that in addition to the previously measured forms of effort such as cosponsorship, floor speeches, and campaign contributions, collaboration is a tool that members can use to increase their effectiveness. Members who are well-connected within the House should have more success in advancing their legislative agenda because they are more likely to have a relationship with someone who can help, such as a member who sits on the appropriate committee, or who has the ear of the party leadership. Yet there is a balance to be found between collaborating enough that a member is able to tap into their personal network to advance their agenda, and not collaborating so frequently that a member is not taken seriously. Just as members who moderate their floor speeches and cosponsorship activities are more effective, so too should we expect to observe greater success from members who are balanced in their collaborative relationships. As described in Matthews (1959), “The really effective senators are those who speak only on the subjects they have been dealing with at close quarters, not those who are on their feet on almost every subject all the time” [p.1068].

Effort Hypothesis: Members who collaborate with their colleagues are more effective legislators so long as they do not overextend themselves in the network.

In addition to the overall amount of collaboration a member engages in, there is reason to expect that whom a member collaborates with matters. Members of

Congress seek out collaborators who are well-positioned to help advance their policy agenda (Craig, 2016). Members of the minority party should be more successful when they collaborate with majority party colleagues. As described by one congressional

58 staff member, “Being in the minority [...] you have a much better chance of actually passing something or getting traction if you have someone from the majority on a bill.” (Interviews, 2016). In some cases, members of the minority party go so far as to do the majority of the work preparing a bill before passing it off to a majority party colleague for introduction as they expect it will be more likely to pass.25 Similarly, more junior members of Congress should find benefit from collaborating with senior colleagues, who have a greater understanding of the rules and strategies of the House

(Fenno, 1973b; Frantzich, 1979).

Members also seek out collaborators who are institutionally positioned to achieve success, whether that entails collaborating with someone who sits on the committee of jurisdiction or someone who has a good relationship with the House leadership.

Hasecke and Mycoff (2007) find that majority party members who are more loyalto the party in their voting records and who give more money to other party members’ campaign committees are more successful. A member who is not in the good graces of their party may find greater success by collaborating with colleagues who can appeal directly to the leadership to get their bill scheduled for a vote, which is a common occurrence for those in a position to do so, like one Republican member whose staff reported, “He’s been around a while and has a good relationship with the Speaker so if it’s an issue he cares about he speaks directly to him” (Interviews, 2016). One common theme in over twenty interviews with congressional staff was that members would seek out collaborators who were on the committee of jurisdiction, particularly if they were working on an issue outside their own committee’s jurisdiction.

25Similar anecdotes were reported by staff to minority party members in three different interviews. However, our traditional metrics of legislative success do not give any credit to the minority party member in instances like these, an issue I address in greater detail later in this article.

59 There is also a reputational aspect to the decision of whom to collaborate with.

When members of Congress send out Dear Colleague letters together, they are associated with each other in the eyes of their colleagues. When asked how their boss decided whom to collaborate with, several staff members were direct about the role a member’s reputation plays. “If it’s a colleague he doesn’t think is a very serious member, he may not want to be associated with that member.” Or, as another staff member put it, “We’re looking for a member with credibility” (Interviews, 2016).

Majority party members, those with more seniority, and members who are well positioned within the House, whether it is due to their institutional position or their reputation with the leadership and their colleagues, have all been shown to be more effective members (Anderson, Box-Steffensmeier and Sinclair-Chapman, 2003; Cox and Terry, 2008; Frantzich, 1979; Hasecke and Mycoff, 2007; Volden and Wiseman,

2014). I expect that members who collaborate with highly effective colleagues are going to reap the benefits of that effectiveness as well, whether it is because effective members seek out other effective colleagues to work with, or because the less effective colleague is learning how to be more effective from their collaborative partner.26 These sort of relational influences are well-documented in social networks, from academic departments benefitting from the prestige of the departments they both hirefrom and place their own PhDs in (Burris, 2004) to actors who collaborate with elite peers being more likely to receive award nominations (Rossman, Esparza and Bonacich,

2010).

26Distinguishing between influence (more effective members causing the effectiveness of theircol- laborators to increase) and homophily (more effective members choosing to work together) isa common issue in social networks as detailed in Aral, Muchnik and Sundarajan (2009) but beyond the scope of this article.

60 Learning/Homophily Hypothesis: Members of Congress who collaborate with more

effective colleagues are more effective themselves.

It is not only minority party members who benefit from bipartisan collaboration.

A rational member of the majority party would not collaborate with a minority party

member unless there was some benefit for her. And indeed there is. Bipartisan

legislation is more likely to move through the legislative process smoothly be signed

into law (Binder and Lee, 2015; Fowler and Marshall, 2017; Harbridge, 2015). Not

only will it be easier to pass the 60 vote hurdle in the Senate, it is likely to move

more quickly through the House as well. “You’ll face less opposition on the floor. If

it’s something more minor, you’re more likely to get it on the suspension calendar,

which requires a bipartisan vote” (Interviews, 2016). Of the 385 bills signed into

law in the 111th Congress (2009-2010), 75% passed the House either by voice vote

or unanimous roll call vote.27 Despite the Democrats holding control of the House,

Senate and presidency in the 111th Congress, only 8.1% of the bills that became law were significantly partisan, which I define as bills that passed the House withfewer than 50 votes from the minority party. Only five bills became law without any support from Republicans in the House, indicating that even under unified party government, bipartisanship is advantageous for members who want to see their bills enacted.28

27Data were collected from the govtrack.us website and coded by the author. For bills that received two votes in the House (pre and post Senate consideration) I recorded the result for the initial vote to reflect the partisanship of the House bill before it was modified to pass the 60 votethreshold in the Senate. The one exception to this coding rule was H.R. 3590, the Patient Protection and Affordable Care Act which originated as the Service Members Home Ownership Tax Actof2009 and was changed in whole by the Senate. In this case I recorded the results of the final House vote because the final version of the bill was unrelated to the bill at introduction. 28The five bills were H.R. 1, the American Recovery and Reinvestment Act of 2009, H.R.4314, To permit continued financing of government operations, H.R. 3590, the Patient Protection and Affordable Care Act, H.R. 4872, the Health Care and Education Reconciliation Act of2010,and H.R. 4173, the Dodd-Frank Wall Street Reform and Consumer Protection Act.

61 When two members from opposing parties send out a Dear Colleague letter it sends a clear signal to the other members of the House that an issue is bipartisan.

“It makes it easier to be noticed and lets people know there isn’t any sort of agenda or traps or anything like that. It makes a bill more inviting to the other side of the aisle. If you get an endorsement from someone in the other party, people will be more open to it.” Dear Colleague letters are frequently used to send out this sort of signal, whether by advertising the other members who have signed onto a bill or policy level, or by listing the interest groups that have endorsed legislation (Box-Steffensmeier,

Christenson and Craig, 2013). Members are conscious of the signals they are sending with their choice of collaborative partner(s). “If we are trying to make something bipartisan, we don’t go to the Freedom Caucus members. We consider how willing people are to work with others” (Interviews, 2016). The effect of Dear Colleagues as a signaling mechanism is likely to be greatest early in the legislative process when members and their staff have little information about a bill and use the names ofthe collaborating members as a quick cue to decide whether a policy is worth supporting.

Signaling Hypothesis A: Members who collaborate on a bipartisan basis are more effective legislators.

Signaling Hypothesis B: Bipartisan collaboration will have the greatest effect in the earliest stages of the legislative process.

The legislative politics and social networks literatures give us four mechanisms by which we can expect collaboration in Congress to be associated with increased legislative effectiveness: 1) effort, 2) learning, 3) homophily, and 4) signaling. In order to test the resulting hypotheses we need a measure of meaningful collaboration in Congress that goes beyond cosponsorship to reflect effort, influence, and signaling

62 on the part of members. Dear Colleague letters are an ideal source to identify these relationships.

Dear Colleagues in Congress

To capture substantive and purposive collaboration between members of Congress, we need to distinguish between the members of Congress who sign onto a policy pro- posal after it has been drafted and those who were involved in its creation. However, the legislative record only lists a single bill sponsor regardless of how many members were involved in its drafting, and it is limited to the bills introduced. To identify the members of Congress who collaborate on policy across the spectrum of congressional activities, I turn to Dear Colleague letters.

Members of Congress have a long history of sending letters to their colleagues to build support for legislation. Early House rules did not give individual members the power to introduce legislation and instead required “explicit approval of the full chamber.” As a result, letters between members were a commonly used tool to build this support. The term “Dear Colleague,” which comes from the traditional salu- tation on these letters, came into usage in the early 20th century (Peterson, 2005).

Letters were distributed to all 435 member offices and the House committees solely by internal mail until 1998 when the Dear Colleague listserv was established. The listserv provided members with the ability to send Dear Colleague letters to every

House office via email, resulting in a more efficient distribution system.29

29It should be noted that the transition from paper letters to email distribution was not immediate and the usage of the Dear Colleague listserv in its early years was limited to the “early adopters” while many members continued to send paper letters. By 2003 there was widespread usage of the Dear Colleague listserv.

63 Figure 3.1: Distribution of Dear Colleague Letters by Purpose

20 Percent of Total Percent

Caucus Cosign Cosponsor Floor General Invitation Letter Purpose

Members typically send out a Dear Colleague letter either before or shortly after they introduce a bill to provide a brief summary of the legislation and solicit cosponsors. They have been described in the legislative politics literature as one of the primary ways that members build support for their bills, particularly in the earliest stage of the legislative process (Lipinski, 2009; Koger, 2003; Campbell, 1982; Krutz,

2005). However, they are also used for other purposes in the policy process, such as soliciting cosigners for letters to administrative agencies and committees within the House, inviting members to participate in briefings and other legislative events, and urging support for legislation on the floor. Figure 4.2 provides the distribution of letters by purpose for the 111th Congress and demonstrates that while soliciting cosponsors is the primary usage of Dear Colleague letters, they serve several purposes in the policy process: recruiting members for congressional caucuses, soliciting

64 cosigners to policy letters, soliciting cosponsors for legislation, urging floor support

or opposition for bills and amendments, providing general information on an issue,

and inviting members and staff to events.30 As a result, they are a valuable tool for measuring policy collaboration that goes beyond bills and their cosponsors.

When multiple members sign a Dear Colleague letter together it is a clear indi- cation of policy collaboration. In July 2009, Congressman Brad Sherman (D-CA) introduced H.R. 3516, the “Enable Divestment from Sudan and Iran Act of 2009.”

According to the legislative record, the bill was introduced by Congressman Sherman and had twenty cosponsors, eighteen of whom were so-called “original cosponsors” meaning they added their name to the bill before it was introduced. The natural assumption is that Congressman Sherman authored this legislation alone, but the

Dear Colleagues tell a different story. In May 2009, a Dear Colleague letter was sent out to solicit cosponsors for this legislation and it was signed by Congressman Sher- man and Congresswoman Ileana Ros-Lehtinen (R-FL), sending a clear signal to their colleagues that the bill was a joint effort.

Figure A shows a letter sent by Congressman Peter Visclosky (D-IN) and Con- gressman Frank LoBiondo (R-NJ) in support of H.R. 6045, the Bulletproof Vest Part- nership Grant Act of 2008, which would end up with 171 cosponsors and was signed into law on October 15, 2008. Again, if you look at the legislative record, there is no way to distinguish Congressman LoBiondo’s involvement with the bill from the other copsonsors, while the letter draws a clear distinction between Congressman Lo-

Biondo, who signs the letter as a co-author, and the other 108 cosponsors who had signed on at the time the letter was sent. From the letter, it is clear that Congressmen

30Based on a sample of 1600 letters.

65 Figure 3.2: Sample Dear Colleague Letter from 110th Congress

“Body Armor is one of the most important pieces of equipment an officer can have and often means the difference between life and death.” Chuck Canterbury, National President, Fraternal Order of Police

Protect our law enforcement officers! Join the 109 Bipartisan Cosponsors of H.R. 6045, the Bulletproof Vest Partnership Grant Act of 2008

Supported by the National Association of Police Officers, Southern States Police Benevolent Association, the Fraternal Order of Police, the National Narcotics Officers’ Associations’ Coalition, and the International Association of Chiefs of Police

Current Cosponsors (109): LoBiondo, Stupak, Ramstad, Crenshaw, Baca, Bishop (GA), Udall (CO), Neal, Brady (PA), Jackson Lee, Gonzalez, Sutton, Moran (VA), Brown (FL), Holden, Holt, Kildee, Ackerman, Gerlach, King (NY), McHugh, Schwartz, Rothman, Michaud, Grijalva, Costello, Napolitano, Etheridge, Young (AK), Shays, Hinojosa, Farr, Engel, Hall (TX), Nadler, Pascrell, Abercrombie, Ellsworth, Hill, Payne, Moore (KS), McCollum, Pallone, Ryan (OH), Scott (VA), Peterson (MN), Allen, Price (NC), McIntyre, Donnelly, McNulty, Gene Green, Udall (NM), Carney, Lowey, Lipinski, Saxton, Carson, Murphy (PA), Eddie Bernice Johnson, Oberstar, Christenson, McGovern, Larsen, Turner, Hayes, DeLauro, Filner, Dicks, Langevin, Kuhl, Mica, Edwards, Cuellar, Kucinich, Rahall, Jackson, Butterfield, Bilirakis, Lewis (GA), Sires, Boyda, Boswell, Giffords, Ortiz, Emanuel, Altmire, Snyder, Shea-Porter, Schakowsky, Higgins, Sestak, Capps, Olver, Hinchey, Tiahrt, Coble, Dingell, Matsui, Walz, Courtney, Lofgren, Berry, Davis (TN), Fattah, Mitchell, Waters, Mack, Lee

Dear Colleague:

In conjunction with National Police Week, May 11-17, 2008, we have introduced H.R. 6045, the Bulletproof Vest Partnership Grant Act of 2008, to reauthorize the highly successful Bulletproof Vest Partnership Grant Program for another three years.

Bulletproof vests and body armor have saved thousands of law enforcement officers since the introduction of the modern bulletproof material. However, they cannot protect the lives of those who do not have access to them. The Bulletproof Vest Partnership Grant Program was created in 1999 to provide state, local, and tribal law enforcement officers with needed protection by aiding the purchase of protective equipment.

In Fiscal Year 2007 alone, the Bulletproof Vest Partnership Grant Program provided $28.6 million to state and local law enforcement agencies across America. This funding provided up to 50 percent matching grants for the purchase of approximately 180,173 new bulletproof vests. Below is a state-by-state breakdown of the FY 2007 BVP grants. For a jurisdictional level list of the FY 2007 awards, please visit: http://www.ojp.usdoj.gov/bvpbasi/docs/FY07_Awards_Reports.pdf

Our law enforcement officers are depending on us to guarantee that this program continues. We must ensure that we protect those who risk their lives every day protecting our communities. Help save lives by becoming a cosponsor of the Bulletproof Vest Partnership Grant Act of 2008. To become a cosponsor or request additional information, please contact Seren Orgel at 5-2461 (Visclosky) or Lance Seibenhener at 5-6572 (LoBiondo.)

Sincerely, Peter J. Visclosky Frank A. LoBiondo Member of Congress Member of Congress

66 Visclosky and LoBiondo intend to convey that the bill is the work of both members

as beyond signing the letter together, the language they use is collaborative. “In

conjuction with National Police Week, May 11-17, 2008, we have introduced H.R.

6045, the Bulletproof Vest Partnership Grant Act of 2008, to reauthorize the highly successful Bulletproof Vest Partnership Grant Program for another three years.” I argue that members who sign Dear Colleague letters together have true collaborative relationships and these ties are what form the policy collaboration network.

One of the chief issues with using cosponsorship as a measure of collaboration is that it is a noisy measure (Fowler, 2006a; Kirkland and Gross, 2012). The ties between members can represent either policy collaboration or policy agreement, with no clear way to distinguish the two. The average bill in the 110th Congress had

18 cosponsors while the average Dear Colleague letter in the same Congress had 1.7 signers.31 Looking solely at the bills and letters where there is a relationship between members, the average bill had 30 cosponsors while the average Dear Colleague had

2.4 signers. The range of cosponsors on a bill in the 110th Congress was one to 406, while Dear Colleague letters had between one and fifteen signers, with 99% signed by no more than five members. These are small scale collaborations that allow usto distinguish substantive collaboration from policy agreement. Figure 3.3 compares the distribution of signers on Dear Colleague letters to the distribution of bill cosponsors for the 108th-111th Congresses (2003-2011). It is reasonable to assume that the average 2-3 signers on a Dear Colleague letter are more likely to have meaningful interactions than the average 30 cosponsors on a bill and therefore it is a stronger

31Cosponsorship data are from the Congressional Bills Project (CBP). The CBP codes a bill’s primary sponsor as the first cosponsor which I maintain to ensure a fair comparison between cosponsorship and Dear Colleague signatures.

67 Figure 3.3: Distribution of Signers by Source

Percent 20

0 01 02−05 06−10 11−15 16−20 21−25 26+ Signers

Cosponsorship Dear Colleague

measure of interaction than cosponsorship (Borgatti and Halgin, 2011). Although

Dear Colleague letters have received little attention in the field as a comprehensive archive was previously unavailable, they are an important part of the policy process that provide key insights into the formative stage of the policy process.

Data and Methods

The data used for this paper consist of 82,712 letters sent to the Dear Colleague listserv between 2003 and 2011. Each session of Congress represents a complete population as the composition of the House network changes with each election and there are Congress-level factors that affect the structure of the network such as ideological

68 polarization, divided government, and recent turnover. Furthermore, all the bills that failed to pass in the previous session have to be reintroduced in the new Congress and members once again go through the process of finding collaborative partners and sending Dear Colleague letters to promote their policies. I use the 108th through 111th

Congresses for this analysis because I am confident I have the complete networks which is important to ensure accuracy of the network measures (Borgatti, Carley and

Krackhardt, 2006).32

Each office writes and sends their own letters which results in substantial variation in their content and formatting. The key piece of information for this study is the names of the members who signed each letter so I wrote a scraper that retrieved the content following “Sincerely,” in each letter and flagged the letters that used an alternate valediction for hand coding.33 In addition to the members who signed each letter, I obtained the date the letter was sent, the title of the Dear Colleague, the issue areas of the letter as designated by congressional staff, the body of the letter, the bill number when one was specifically referenced, and the names of any interest groups that endorsed the bill.

With a complete list of all of the members who signed each letter, I create the policy collaboration network for each Congress. Each network begins as a two-mode affiliation matrix in which members are not directly connected to each otherbut instead are connected to “events” which in this case are the Dear Colleague letters they sign in a given Congress. However, because I am interested in the relationships

32As Borgatti, Carley and Krackhardt (2006) point out, network centrality measures are robust under small amounts of error, however as members were still transitioning from paper letters to the electronic listserv in the 107th Congress, I am not confident that the resulting measurement error would be under the acceptable threshold. 33Approximately ten percent of letters were hand coded as a result.

69 Figure 3.4: Policy Collaboration Network for the 110th Congress

between members, I transform the two-mode matrices into one mode projections which is accomplished by multiplying the affiliation matrix for each Congress by its transpose. The resulting adjacency matrices become four networks in which the nodes represent the members who served in a given Congress, and the edges are undirected ties indicating that two members signed a Dear Colleague letter together during that Congress (Breiger, 1974; Wasserman and Faust, 1994).34 Figure 3.4 provides a

34In future work I intend to explore the policy collaboration network with directed ties and weighted edges, however there are substantial measurement issues that need to be addressed to ensure I have an accurate count of unique collaborations between members and am able to properly identify the primary signer of each letter.

70 Table 3.1: Policy Collaboration Network Summary Statistics

108th 109th 110th 111th 2003-2004 2005-2006 2007-2008 2009-2010 Nodes 439 438 447 443 Edges 4750 5725 4809 4493 Bipartisan Edges 1809 2234 2360 2059 Isolates 17 18 7 7 Average Degree 21.64 26.14 21.52 20.28 Maximum Degree 97 97 93 90 Density 0.049 0.060 0.048 0.046 Transitivity 0.293 0.290 0.190 0.201 Letters 10,278 14,909 25,857 31,668

visualization of the policy collaboration network for the 110th Congress. Nearly all of

the 447 members in this network are connected to each other in one large component,

with seven isolates who either did not send a Dear Colleague letter during the 110th

Congress or did not sign a letter with a colleague.35

Table 3.1 provides summary statistics for each of the four networks. As described

above, Nodes represents the number of members in each network and Edges represents the number of collaborative relationships between them. Bipartisan Edges reflects the number of collaborative relationships in the network that are between Democratic and Republican members. Isolates indicates the number of members who are not connected to anyone else in the network. Density is a measure of how well-connected

the network as a whole is and is calculated by dividing the actual number of observed

35The network for the 110th Congress includes twenty seven members who did not serve a full term. Of the seven isolates, four are members who served partial terms (Rep. Don Cazayoux (D-LA), Rep. Jo Ann Davis (R-VA), Rep. Robert Latta (R-OH), and Rep. Steve Scalise (R-LA)). The other three are Rep. Gresham Barrett (R-SC), Rep. David Davis (R-TN), and Rep. Kenny Marchant (R-TX).

71 ties in each network by the number of potential ties. So for the 110th Congress,

4.8% of all possible ties in the network are realized. Transitivity captures the number of closed triangles in the network in relation to all possible closed triangles, or the

“a friend of a friend is my friend” phenomenon. If member i and memberj both collaborate with member k, the transitivity measure reflects how often members i and j will collaborate with each other. In the 110th Congress, 19% of all possible triangles were closed. The degree measures will be discussed in further detail below.

My interest is in the relationship between how collaborative a member of Congress is and their legislative effectiveness, so I create a series of measures to capture different aspects of policy collaboration. First, I calculate the degree centrality of every member, which is the total number of connections a node has to other nodes in the network. In this case Degree Centrality represents the number of other members that each member collaborates with in a given Congress (Freeman, 1979). For example, in the 110th Congress, Representative Dan Boren (D-OK) had a degree centrality of

21, indicating that over the course of the Congress he signed Dear Colleague letters with twenty-one of his colleagues. Degree centrality is used as a measure of immediate influence, or the extent to which members’ legislative effectiveness is influenced by their collaborators (Borgatti, 2005).36 As seen in table 3.1, the average member collaborated with about 22 of her colleagues in the 110th Congress and the most collaborative member, Congressman Christopher Shays (R-CT), was tied to 93 of his colleagues.

36Another common measure of centrality is eigenvector centrality, which reflects not only the number of ties a node has, but also the number of ties the connected nodes have. It is appropriate for measuring similar parallel flow processes, but I use degree centrality here because I am interested in the effect of the size of a member’s personal collaboration network.

72 The effort hypothesis does not suggest that the benefits of collaboration continu- ally increase the more collaborators a member has. If that were the case, the average member of Congress should collaborate with far more than 22 of her colleagues. In- stead, I expect that the benefits of collaboration come from striking the right balance.

A member like Congressman Wayne Gilchrest (R-MD) who collaborated with 53 colleagues in the 110th Congress and only introduced four bills of his own may not be seen as a member who is deliberate in his choice of collaborative partners. At the same time, a member like Congressman Howard Coble (R-NC) who introduced 49 bills in the 110th Congress and only collaborated four times may not be seen as a team player. To examine the relationship between total collaboration and legislative effectiveness, I create a measure I call Tactical Collaboration which reflects the overall amount of collaboration a member engages in relative to the number of bills they introduce and the average collaboration in the chamber. For member i in a network with n nodes at time t, their tactical collaboration score is calculated as the absolute value of the ratio of collaborative relationships to bills introduced for member i subtracted from the average ratio of collaborative relationships to bills introduced for the whole chamber. ⏐ n ( ) ⏐ ⏐ 1 ∑ degreeit degreeit ⏐ TC = ⏐ − ⏐ it ⏐n billsintro billsintro ⏐ ⏐ i=1 it it ⏐ I incorporate the number of bills a member introduces into my measure of tactical collaboration for two reasons. First, because I expect the “ideal” amount of collaboration for any member should be proportional to the number of bills they introduce.

Both members who engage in a lot of collaboration without introducing many bills of their own, and members who introduce a lot of legislation but rarely collaborate are penalized. Second, our existing measures of legislative effectiveness are based on

73 the “one sponsor per bill” model and so collaborators get no credit for the successes of legislation they worked on if they did not introduce it themselves.37 I relate each member’s activities to the chamber mean, which is 1.8 collaborative relationships for every bill introduced, because my expectation is behavior that adheres to the chamber norm will be most rewarded.

To test the learning/homophily hypothesis, I create a spatial lag term by multiplying the NxN adjacency matrix (A) for each Congress by the Nx1 outcome vector

(Y ) for each dependent variable. So, to model bill success for a single Congress, the predictor variable (AY ) is created by taking the adjacency matrix of collaborative relationships for that Congress and multiplying it by the vector of bill passage rates for each member. In this case, because the adjacency matrix is binary and undirected, the result is a covariate that reflects the sum of all the bills passed by a member’s collaborators, or the total effectiveness of her colleagues. For member i who collaborates with n members j at time t, her network collaboration is calculated as the sum of the outcome variable yj for all of i’s collaborative partners.

n ∑ NCit = yjt j

Finally, there is the signaling hypothesis, which I test by creating a measure that captures a member’s propensity for bipartisanship. I measure Bipartisan Collabora- tion as simply the number of ties a member has to colleagues in the other party.38 I use the total number of collaborations with the opposing party rather than the percentage of collaborations that are bipartisan because the latter rewards low-degree

37This is an issue I will address in more depth in future work by establishing a measure of legislative effectiveness that rewards collaboration. 38Where present in the data, Congressman Bernie Sanders (I-VT) is coded as a Democrat.

74 members and as a signaling mechanism, I expect that a member with twenty bipartisan collaborations out of forty Dear Colleagues signed is going to be viewed by colleagues as more bipartisan than a member with five bipartisan collaborations out of five Dear Colleagues sent.

Table 3.2 lists the highest and lowest ranking members of the 110th Congress by degree centrality, tactical collaboration score, and bipartisan collaboration, along with their legislative effectiveness scores as calculated by Volden and Wiseman (2014). As it has become more difficult for a member of Congress to see their initiatives signed into law, Congress scholars have explored several methods to measure legislative effectiveness that are not based solely on the number of bills they introduce that are signed into law. Anderson, Box-Steffensmeier and Sinclair-Chapman (2003) examine legislative success in stages: the number of bills a member introduces that are reported out of committee, the number of bills that pass a vote on the floor, and the number of bills that are signed into law. Cox and Terry (2008) disregard lawmaking entirely and consider only the number of a member’s bills that are reported from committee and passed by the House. The Volden and Wiseman (2014) composite score is based on a member’s success at five different stages of the legislative process:

1) the number of bills they introduced, 2) the number of their bills that receive action in committee, 3) the number of their bills reported out of committee, 4) the number of their bills that pass the house, and 5) the number of their bills that are signed into law. They then construct a legislative effectiveness score for three different classes of legislation: 1) ceremonial bills, 2) substantive legislation, and 3) substantive and significant legislation. The member’s activities are weighted by importance and calculated as a proportion of all legislative activities in a Congress. The result is a score

75 Table 3.2: Most and Least Collaborative Members of the 110th Congress

Member DG LES Member DG LES 1) Christopher Shays (R-CT) 93 0.34 440) Randy Neugebauer (R-TX) 1 0.29 2) Ileana Ros-Lehtinen (R-FL) 74 2.36 439) John Murtha (D-PA) 1 1.27 3) Lois Capps (D-CA) 72 2.13 438) Michael McNulty (D-NY) 1 0.23 4) Carolyn Maloney (D-NY) 70 4.71 437) Virgil Goode (R-VA) 1 0.26 5) Michael Castle (R-DE) 64 1.29 436) Terry Everett (R-AL) 1 0.04 6) Ron Kind (D-WI) 63 1.34 435) Vern Buchanan (R-FL) 1 0.40 7) Henry Waxman (D-CA) 63 4.71 434) Niki Tsongas (D-MA) 2 0.33 8) Barbara Lee (D-CA) 62 1.22 433) Adrian Smith (R-NE) 2 0.06 9) James McGovern (D-MA) 61 0.87 432) William Jefferson (D-LA) 2 0.83 10) Mark Kirk (R-IL) 60 0.70 431) Ander Crenshaw (R-FL) 2 0.01

Member TC∗ LES Member TC LES 1) Timothy Bishop (D-NY) 0.00 0.84 440) John Shimkus (R-IL) 40.18 0.01 2) Charles Dent (R-PA) 0.00 0.40 439) John Linder (R-GA) 13.68 0.19 3) Joe Knollenberg (R-MI) 0.00 0.14 438) Wayne Gilchrest (R-MD) 11.43 0.14 4) David Reichert (R-WA) 0.00 0.90 437) Roger Wicker (R-MS) 8.18 0.22 5) Barbara Lee (D-CA) 0.01 1.22 436) Spencer Bachus (R-AL) 8.18 0.22 6) Joe Barton (R-TX) 0.02 0.19 435) Mark Souder (R-IN) 7.18 0.16 7) Howard McKeon (R-CA) 0.02 0.45 434) Donald Manzullo (R-IL) 5.18 0.06 8) Christopher Smith (R-NJ) 0.02 1.44 433) Carolyn Kilpatrick (D-MI) 5.02 0.08 9) Timothy Walz (D-MN) 0.02 0.37 432) Mike Pence (R-IN) 4.93 0.05 10) Carol Shea-Porter (D-NH) 0.03 0.33 431) Lynn Westmoreland (R-GA) 4.85 0.68

Member BC LES Member BC∗∗ LES 1) Christopher Shays (R-CT) 70 0.34 440) GK Butterfield (D-NC) 0 1.07 2) Ileana Ros-Lehtinen (R-FL) 50 2.36 439) Ciro Rodriguez (D-TX) 0 1.36 3) Mark Kirk (R-IL) 46 0.70 438) Jeb Hensarling (R-TX) 0 0.16 4) Michael Castle (R-DE) 46 1.29 437) Jerry Lewis (R-CA) 0 0.12 5) James Ramstad (R-MN) 42 0.70 436) Joe Sestak (D-PA) 0 1.40 6) Phil English (R-PA) 37 0.85 435) David Obey (D-WI) 0 3.40 7) Wayne Gilchrest (R-MD) 36 0.14 434) Jose Serrano (D-NY) 0 1.30 8) Earl Pomeroy (D-ND) 35 0.63 433) Virgil Goode (R-VA) 0 0.26 9) Steven LaTourette (R-OH) 34 1.01 432) Joe Baca (D-CA) 1 0.53 10) Thomas Davis (R-VA) 33 1.40 431) Diane Watson (D-CA) 1 1.17 Isolates and members who served less than a year are excluded from the rankings. ∗For tactical collaboration (TC), lower scores indicate the member is closer to the chamber mean. ∗∗For bipartisan collaboration (BC), degree centrality (DG) was used as a tie-breaker.

that “captures the relative share of all legislative activities in any two-year Congress that can be attributed to each lawmaker” [p.25]

76 I am interested in the impact of collaboration both across the legislative process

as a whole and at individual stages and therefore in this paper I consider four dif-

ferent measures of legislative effectiveness: 1) a composite score based on the Volden

and Wiseman Legislative Effectiveness Score, 2) the average number of cosponsors a

member attracts to the bills she introduces, 3) the number of bills a member intro-

duces that pass the House, and 4) the number of bills a member introduces that are

signed into law.39 Each measure is discussed in more detail below.

The Volden and Wiseman Legislative Effectiveness Score (henceforth referred to

as LES) is valuable as a single composite score that reflects a member’s success ina

given Congress relative to the activities of Congress as a whole. Volden and Wiseman

show, as with most models of legislative effectiveness, that seniority, majority party

status, and whether a member is the chair of a committee or subcommittee are the

primary determinants of a member’s LES. However, I am interested in collaboration

as a tool by which members can take their legislative fate into their own hands and

so I use an alternate measure created by Volden and Wiseman that captures the

degree to which members of Congress over- or underperform expectations which I

call Legislative Performance or LPS. I create this score in three steps. First, I regress a member’s LES on their seniority, majority party status, and whether a member is the chair of a committee or subcommittee as shown in equation (1). I take the resulting coefficients and use them to estimate the predicted value of a member’s

LES based on their seniority, majority party status, and position within the chamber

39Unfortunately all of these measures are still based on the one sponsor per bill model and do not reward collaborators for their efforts. Nor do they account for bills that do not pass asstand alone legislation and are instead incorporated into another bill, which is an increasingly common occurrence in the contemporary House. The creation of such a measure will be addressed in future work.

77 as in equation (2). This gives me each member’s benchmark score (LESˆ ) which reflects the expected LES for a member with their experience and position. Finally, as shown in equation (3), the LPS is simply the ratio of a member’s LES to their benchmark score in a given Congress.40

LESit = β0 + β1Seniorityit + β2Majorityit + β3Chairit + β4SCChairit + ϵ (3.1)

ˆ ˆ ˆ ˆ ˆ ˆ LESit = β0 + β1Seniorityit + β2Majorityit + β3Chairit + β4SCChairit (3.2) LES LP S = it (3.3) it ˆ LESit

The LPS reflects a member’s overall effectiveness compared to expectations, butIam

also interested in examining the relationship between collaboration and legislative ef-

fectiveness at specific stages of the legislative process. The second dependent variable

in my analysis is Average Cosponsors which is the average number of cosponsors a

member has across all of the legislation she introduced in each Congress. The aver-

age number of cosponsors is used because it is a measure of the level of support a

member has across all of her legislative initiatives, whereas total number of cospon-

sors is somewhat skewed by the few members who introduce high profile legislation

that garners over half the chamber as cosponsors. In addition to the cosponsorship

model, I also estimate models using Bills Passed and Laws Enacted as the dependent

variables. These are the raw count of the number of bills a member introduced in

each Congress that passed a vote on the House floor and were signed into law re-

spectively. Following Anderson, Box-Steffensmeier and Sinclair-Chapman (2003), the

count model is preferred to the proportion of bills introduced that passed because it

represents the information most easily conveyed to constituents and other audiences.

40Benchmark scores are discussed in more detail on Volden and Wiseman’s website, www.thelawmakers.org.

78 A member who introduces twenty bills and passes three of them will likely be seen by others as more successful than a member who introduces and passes one bill.

With these data I use a temporal network autocorrelation model (TNAM) to estimate the relationship between collaboration and legislative effectiveness. The

TNAM builds on the spatial autocorrelation model, in which the outcome Y is a function of exogenous covariates X and a spatial weight matrix W multiplied by Y so that it represents the outcome values of an observation’s neighbors. While originally intended to capture physical proximity, the spatial autocorrelation model is easily transformed to capture connections in a network by using the adjacency matrix A for each network as the weight matrix W (White, Burton and Dow, 1981)

Y = α + ρAY + Xβ + ϵ

In this way my estimations of a member’s legislative effectiveness do not require the assumption that members of Congress are independent from one another and instead accounts for the likelihood that a member’s legislative effectiveness, whether measured as performance compared to expectations or a count of successes, is influenced by the legislative effectiveness of her colleagues. In addition to removing bias that results from interdependent observations, the network autocorrelation model can also be used to test social influence hypotheses, which I do here as the AY term is also the predictor variable used to test my learning/homophily hypothesis. Because I am interested in the cumulative influence of direct connections, I maintain the weight matrix in its simplest form, A (Leenders, 2002). Leifeld and Cranmer provide an extension of the spatial autocorrelation model through the xergm package in the form of the TNAM (Leifeld, Cranmer and Desmarais, 2014). The TNAM provides a flexible structure that can be used to account for a wide array of network and

79 temporal dependencies from the behavior of a node’s direct or indirect connections at any observed time to the structure of the network itself. I use the TNAM to model the relationship between collaboration and legislative effectiveness across four different

Congresses while incorporating fixed effects that account for the expectation that there are Congress-level factors that influence both the collaborative and legislative dynamics of the chamber.

Of course I expect that a member’s legislative effectiveness is influenced by more than just the collaboration network and the effectiveness of her collaborators. SoI include several other traditional determinants of legislative success as controls in my models. Whether a member is in the Majority party in a given Congress is consistently one of the strongest predictors of legislative effectiveness in the literature as discussed above. I expect a strong, positive effect for the dichotomous variable in the cosponsorship, bill passage, and lawmaking models. I expect a negative effect in the legislative performance model because majority party status is incorporated into the benchmark score for each member and I expect that the average majority party member will be more likely to underperform expectations than a minority party member. Seniority is another predictor frequently associated with legislative effectiveness, measured here as the number of terms that a member of Congress has served, with members in their first term coded as one. As with Majority, I expect a strong, positive effect for seniority in the cosponsorship, bill passage, and lawmaking models and a negative effect in the performance model because it will be harder for senior members to outperform expectations. Two other covariates frequently associated with legislative effectiveness are whether a member isa Committee Chair or Subcommittee Chair. These are metrics of a member’s power in the House and

80 I expect that both committee chairs and subcommittee chairs will be successful at advancing their agenda across all four measures of effectiveness.

Beyond the four covariates included in a member’s benchmark score, I expect several other covariates will be associated with both collaboration and legislative effectiveness. I include a variable to represent a member’s ideological Chamber Extrem- ity which is the absolute value of the difference between a member’s first dimension

DW-NOMINATE score and the median for all House members in a given Congress.

Similarly, Party Extremity represents the absolute value of the difference between a member’s DW-NOMINATE score and the median for their party in that Congress.

The extremity scores test whether a member who is far from the ideological center of either the chamber or their party is punished for that distance. I expect that the legislative process rewards both ideological moderates and those who toe the party line and therefore Chamber Extremity and Party Extremity will be negative across all four dependent variables. I also account for whether a member is a part of the majority party leadership with Majority Leader, which I expect will be positive in the legislative performance, bill passage and law models as members of the leadership will be in the best position to put their bills on the House calendar. I do not expect an effect in the cosponsorship model, as members of the leadership will be lessreliant on cosponsors to gain attention for their bills.

Bills Introduced is a count of the total number of bill introduced by the member in that Congress. I expect that increasing the number of bills a member introduces will decrease the average number of cosponsors, as the more legislation they introduce, the more likely it is that one or more of those bills will have zero or one cosponsor. I expect it will increase the degree to which a member over performs their expected legislative

81 effectiveness as well as the number of bills passed and signed into law as introducing

more legislation provides more opportunities for a member to pass something. Finally,

Electoral Margin represents the percentage of the vote that each member received in

the previous election as recorded by the Federal Elections Commission.41 I expect more electorally vulnerable members will be more effective across all measures of effectiveness as the leadership will look for opportunities to help those members be successful. The data for all controls were drawn from the Lawmakers Project (Volden and Wiseman, 2014), the Congressional Bills Project (Adler and Wilkerson, 2003-

2012) and VoteView (Poole and Rosenthal, 2011).

Analysis

The results of four models testing the effort and learning/homophily hypotheses are reported in table 4.2. The LPS model shows support for both the effort and learning/homophily hypotheses. Members who are further away from the chamber average in their collaborative activities are less likely to outperform their expected legislative effectiveness, while members whose co-collaborators have a higher LPSare more likely to outperform their own expectations. Substantively, a one standard deviation increase in a member’s Tactical Collaboration measure in the 110th Congress

is associated with a -0.06 point decrease in Legislative Performance. Members who

deviate from the chamber norms in their collaborative behavior are more likely to

underperform expectations in their legislative effectiveness. Members are also influ-

enced by the legislative performance of their collaborators. A one point increase in

41For run-off elections, I recorded the vote received in the final contest.

82 Table 3.3: Relationship Between Collaboration and Legislative Effectiveness

LPS Model Cosp Model Pass Model Law Model Tactical Collab −0.025∗ 0.299∗ −0.057∗∗ −0.028 (0.012) (0.128) (0.020) (0.024) Network Collab 0.004∗∗ 0.008∗∗∗ 0.001∗ 0.000 (0.001) (0.001) (0.001) (0.002) Majority Party −0.776∗∗∗ 6.333∗∗∗ 0.521∗∗∗ 0.297 (0.135) (1.468) (0.116) (0.167) Seniority −0.086∗∗∗ 0.086 0.015∗∗ 0.037∗∗∗ (0.008) (0.084) (0.006) (0.008) Comm Chair 0.456∗∗ −4.207∗∗ 1.183∗∗∗ 1.198∗∗∗ (0.145) (1.572) (0.074) (0.105) Subcomm Chair 0.307∗∗∗ −0.081 0.278∗∗∗ 0.255∗∗ (0.083) (0.896) (0.056) (0.083) Chamber Extremity −0.185 6.064∗∗ −0.324 −0.536∗ (0.201) (2.183) (0.169) (0.249) Party Extremity −0.717∗ 1.305 −0.135 −0.593 (0.302) (3.272) (0.243) (0.361) Majority Leader 0.635∗∗∗ 3.728 0.389∗∗∗ 0.442∗∗ (0.176) (1.917) (0.108) (0.159) Bills Introduced 0.040∗∗∗ −0.228∗∗∗ 0.025∗∗∗ 0.023∗∗∗ (0.003) (0.031) (0.002) (0.002) Electoral Margin −0.010∗∗∗ 0.079∗∗ −0.007∗∗∗ −0.003 (0.002) (0.024) (0.002) (0.003) 109th Congress −0.144 −1.322 −0.110 −0.203∗ (0.079) (0.859) (0.067) (0.088) 110th Congress −0.142 1.004 0.103 −0.366∗∗∗ (0.080) (0.868) (0.067) (0.091) 111th Congress −0.131 0.228 −0.088 −0.536∗∗∗ (0.080) (0.870) (0.067) (0.096) Constant 2.294∗∗∗ 6.016∗∗ 0.027 −0.598∗ (0.209) (2.263) (0.170) (0.247) AIC 5504.812 13880.940 BIC 5592.299 13968.454 Log Likelihood -2736.406 -6924.470 Deviance 2334.610 275794.080 2427.865 1982.510 Num. obs. 1751 1754 1754 1754 ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05

83 the total legislative performance score for all of a member’s collaborators is associated with a 0.004 increase in that member’s own legislative performance score. It is a substantively small effect, but when taken in consideration with the fact thatthe average legislative performance score across all four Congresses is 1.153, it is easy to see how a member’s legislative performance score could quickly increase depending on her collaborators.42

The second model is the cosponsorship model, in which the dependent variable is the average number of cosponsors a member has across all of the bills she introduces, which serves as a measure of a member’s ability to draw support from her colleagues.

Here, Tactical Collaboration is surprisingly positive, suggesting that adhering to the chamber norm is not necessary when it comes to attracting cosponsors and may even be detrimental. A one standard deviation in a member’s Tactical Collaboration measure in the 110th Congress is associated with a 0.670 increase in the average number of cosponsors a member has across her legislation. Although this is perhaps not as surprising as it seems considering that members who collaborator more are likely to send out more Dear Colleague letters, which are the primary means by which members solicit cosponsors. The Network Collaboration covariate yields the expected results, with a one member increase in the sum of the average cospsonsors for all of a member’s collaborators associated with a 0.008 increase in that member’s average cosponsors for their own legislation.

Next I look at one of the most traditional measures of legislative effectiveness: the number of a member’s bills that pass the House. Across all four Congresses in my

42In a separate analysis I looked at the relationship between total collaboration and legislative effectiveness and found no evidence that the number of collaborative partners a member hasis associated with either an increase or decrease in legislative effectiveness, which provides further support for my assertion that the “right” amount of collaboration is the middle ground.

84 data, the median number of bills passed by any member is one, which highlights the difficulty members of Congress have finding success on the House floor. But theresults suggest collaboration is associated with an increase in the number of bills a member passes, although the effect is predominantly in the Tactical Collaboration covariate.

Unlike the first two models, here the dependent variable is a count of the number of bills passed and so the linear regression is no longer appropriate. Fortunately, the

TNAM allows for a variety of link functions to accommodate different distributions of Y . Here, my dependent variable is overdispersed, with a mean ranging from 1.378 to 2.051 and a variance ranging from 4.208 to 8.574 so I use a quasi-poisson model to account for the overdispersion (Ver Hoef and Boveng, 2007). Again, the model provides support for both the effort and learning/homophily hypotheses. For the average majority party member in the 110th Congress, a one standard deviation increase in their tactical collaboration score from 0 to 2.239 is associated with a decrease in the expected number of bills passed from 1.560 to 1.372.43 Regarding the learning/homophily hypothesis, for the average majority party member in the 110th

Congress, if they have no successful collaborators, adding a single bill to the count of bills passed by their collaborator is associated with an increase in the predicted number of bills passed from 1.429 to 1.431. Although substantively small, when taken in consideration with the difficulty of passing legislation in the first place, this becomes more meaningful. For a member who is already collaborating with successful colleagues, the effect is smaller in magnitude. A member whose collaborators have passed 15 bills between them is predicted to pass 1.455 bills of their own while a

43Predicted probabilities were calculated holding continuous variables at their mean and discrete variables at their median in the 110th Congress. For the spatial lag term, I set the value to 18 to reflect that the median member of the 110th Congress had 18 collaborators and themedian number of bills passed for all members was one.

85 member whose collaborators have passed 20 bills between them is predicted to pass

1.464 bills of their own

Finally, I look at the effect of collaboration on the number of bills a member has signed into law. Again, I use the quasi-poisson link function in the TNAM and find no discernible effect of either the number of collaborators or the success ofthose collaborators influencing a member’s own success. However, considering the median member had zero bills signed into law, it would be hard for a member to find successful collaborators to learn from.

Across all four models, the control variables yield results largely as expected, with two notable exceptions. First, contrary to the majority of the existing findings in the literature, I find no relationship between being a member of the majority party and the number of bills a member sees signed into law. This appears to be a peculiarity of the time period I study as the result is robust across several different model specifications. While this makes sense for the 110th Congress, as the House was controlled by the Democratic party and the presidency by the Republican party, I would expect any effect unique to the divided government of 2007-2008 would be outweighed by the unified party government in the other Congresses here. It may be a reflection of the increasing difficulty all members are finding seeing their bills translated intolaw and a new trend in lawmaking, which merits further study. Considering the majority of studies on legislative success rely on pre-2005 data and the increasingly negative coefficients for the fixed effects in the lawmaking model, this seems likely. Theother unexpected result is the relationship between ideological extremity and the average number of cosponsors on a bill. While ideological moderates do appear to be rewarded when it comes to the number of bills that are signed into law, those who are

86 further from the chamber median have a much higher average number of cosponsors

on their legislation, which I expect is driven by the need for members who are further

from the ideological median to rely on cosponsors to gain attention for their legis-

lation. Extremity within a member’s own party is associated with performing lower

than expected in legislative effectiveness, but has no relationship with the number of

cosponsors a member attracts or the bills passed.

Next, I turn my attention to the signaling hypotheses and the results in table 4.3.

My expectation is that members will benefit from bipartisan collaboration, particu-

larly in the earliest stages of the legislative process, and yet that does not appear to

be the case. In fact, the legislative performance model suggests that every additional

collaborator a member has in the opposing party is associated with a -0.013 decrease

in the ratio of their LES to their benchmark score, suggesting that members are pun-

ished for bipartisanship by the gatekeepers of their party, likely because committee

chairs and party leaders are more likely to be party loyalists than the average rank

and file member. Neither the bill passage or public law models show any relationship

between legislative success and bipartisan collaboration, however there is some evi-

dence that bipartisan collaboration helps members be more successful in attracting

cosponsors.

The initial cosponsorship model did not show any effect for the bipartisan col-

laboration score. To further examine the unexpected results in the bipartisan col-

laboration models, I created an interaction term between a member’s bipartisan col-

laboration score and whether they are in the majority party, Bipartisan*Majority.

The results, reported as Cosp 2 in table 4.3, show that for members of the majority party, there is a clear benefit to bipartisan collaboration in the form of increased

87 Table 3.4: Relationship Between Bipartisanship and Legislative Effectiveness

LPS Model Cosp 1 Cosp 2 Pass Model Law Model Bipartisan Collab −0.013∗∗ −0.019 −0.158∗ −0.001 −0.002 (0.005) (0.056) (0.065) (0.003) (0.005) Bipartisan*Maj 0.298∗∗∗ (0.073) Tactical Collab −0.025∗ 0.298∗ 0.351∗∗ −0.057∗∗ −0.028 (0.012) (0.128) (0.128) (0.020) (0.024) Network Collab 0.007∗∗∗ 0.008∗∗∗ 0.008∗∗∗ 0.001 0.001 (0.002) (0.001) (0.001) (0.001) (0.003) Majority Party −0.838∗∗∗ 6.238∗∗∗ 3.627∗ 0.515∗∗∗ 0.275 (0.137) (1.493) (1.618) (0.122) (0.174) Seniority −0.083∗∗∗ 0.089 0.094 0.015∗∗ 0.037∗∗∗ (0.008) (0.084) (0.084) (0.006) (0.008) Comm Chair 0.440∗∗ −4.230∗∗ −4.420∗∗ 1.181∗∗∗ 1.193∗∗∗ (0.145) (1.574) (1.568) (0.074) (0.106) Subcomm Chair 0.314∗∗∗ −0.068 −0.402 0.279∗∗∗ 0.257∗∗ (0.082) (0.897) (0.897) (0.056) (0.084) Chamber Extremity −0.285 5.904∗∗ 6.069∗∗ −0.332 −0.570∗ (0.204) (2.231) (2.221) (0.177) (0.261) Party Extremity −0.637∗ 1.416 0.582 −0.128 −0.560 (0.303) (3.288) (3.280) (0.246) (0.368) Majority Leader 0.626∗∗∗ 3.706 3.752∗ 0.389∗∗∗ 0.441∗∗ (0.176) (1.918) (1.910) (0.108) (0.159) Bills Introduced 0.040∗∗∗ −0.227∗∗∗ −0.228∗∗∗ 0.025∗∗∗ 0.023∗∗∗ (0.003) (0.031) (0.031) (0.002) (0.002) Electoral Margin −0.011∗∗∗ 0.077∗∗ 0.072∗∗ −0.007∗∗∗ −0.003 (0.002) (0.025) (0.025) (0.002) (0.003) 109th Congress −0.136 −1.301 −1.291 −0.109 −0.201∗ (0.079) (0.861) (0.857) (0.067) (0.088) 110th Congress −0.111 1.051 1.101 0.101 −0.365∗∗∗ (0.081) (0.879) (0.875) (0.068) (0.091) 111th Congress −0.106 0.271 0.468 −0.090 −0.535∗∗∗ (0.081) (0.879) (0.876) (0.068) (0.096) Control 2.411∗∗∗ 6.207∗∗ 7.879∗∗∗ 0.039 −0.560∗ (0.213) (2.329) (2.354) (0.183) (0.261) AIC 5499.427 13882.818 13867.987 BIC 5592.382 13975.802 13966.440 Log Likelihood -2732.713 -6924.409 -6915.993 Deviance 2324.784 275774.890 273141.270 2427.826 1982.265 Num. obs. 1751 1754 1754 1754 1754 ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05

88 cosponsorship for their legislation. For members of the minority party, adding an

additional bipartisan collaborator is associated with a 0.157 decrease in the average

number of cosponsors on that member’s bills. However, for majority party members,

an additional bipartisan collaborator is associated with a 0.140 increase in the average

number of cosponsors. While this may seem surprising at first glance, it fits with the

reports from congressional staff provided at the beginning of this article. Formem-

bers of the minority party, the benefit of bipartisan collaboration comes from being

the collaborator rather than the lead sponsor on legislation, further demonstrating

the need to account for these collaborators in our measures of legislative effectiveness.

For members of the majority party, bipartisan collaboration signals that a bill “is not

a trap” and is therefore likely to garner more support from both sides of the aisle.

The remainder of the results in the bipartisan collaboration models are largely

consistent with the findings in table 4.2. Tactical Collaboration continues to be beneficial for members seeking to be more effective in navigating the legislative process, but somewhat detrimental to gathering cosponsors and the network collaboration term remains positive and significant except in the bill passage model. Here thead- dition of the Bipartisan Collaboraiton term, while not significant itself, results ina loss of significance for the Network Collaboration term.

Discussion and Conclusion

The findings presented here demonstrate that there are clear legislative benefits to collaboration in Congress. However, members who wish to be more successful in advancing their legislative agenda through the traditional process are best served not by collaborating with as many members as possible, but by being careful in their

89 collaborative choices and finding the right balance between collaborating on policy and introducing their own legislation. The closer a member is to the chamber average of 1.8 collaborative relationships per bill introduced, the more likely they are to pass legislation and outperform their expected legislative effectiveness. Beyond the factors that are largely out of a member’s control such as majority party status, seniority, and position within the House, members of Congress who put in the effort to advance their legislative agenda by forming smart relationships with their colleagues in the

House are more successful.

There is also a clear benefit to collaborating with other successful members. While it is difficult to determine whether the effect is one of successful members choosing to work together, or members learning from their more successful colleagues, it is clear that members who collaborate with other effective members are more likely to be effective themselves. This finding also highlights the inherent interdependence of the legislature. Regardless of whether legislative effectiveness is measured using the legislative performance score or as the number of bills that passed the House, the success of one member is clearly influenced by the effectiveness of the colleagues she works with. Members of Congress are not independent from one another and should not be treated as such.

Of my four hypotheses, the signaling hypotheses have the least support. Members who collaborate with more colleagues in the opposing party are not more successful in the legislative process than those who collaborate largely within their own party and in fact appear to be more likely to underperform expectations for legislative effectiveness.

The one exception to this provides partial support for my second signaling hypothesis in that increased bipartisan collaboration is associated with an increase in the support

90 a member can gain through cosponsorships, but only for members of the majority party. This does not necessarily mean there is no benefit to bipartisanship, but rather that there is no apparent benefit to bipartisanship in the aggregate. Members who build a reputation for bipartisanship within the legislature are not rewarded in the legislative process, however bipartisan bills are still more likely to become law and congressional staff still claim there is benefit to working with colleagues intheother party. It is likely that the legislative benefit of bipartisanship is at the bill level, rather than the member level. This will be explored in future work, along with the possibility that bipartisanship is a signal not for members within the chamber, but for constituents at home.

In addition to highlighting the benefit of collaboration in the House of Repre- sentatives, this paper draws attention to the shortcomings of our existing measures of legislative effectiveness as they do not give credit to collaborative partners who may not have their name on the bill. Volden and Wiseman (2014) acknowledge that their measure does not account for people who work behind the scenes but then they dismiss the idea that it occurs frequently enough to merit consideration. I provide a measure that captures this sort of collaboration for the first time and demonstrate that collaboration is widespread in Congress and valued by members as a tool for success. A true measure of legislative effectiveness should move beyond the question of how well members succeed at advancing the legislation in their name through the process and account for the realities of the contemporary Congress. Members of

Congress can play a key role in crafting policy without necessarily being the one to sponsor the bill and this sort of effort should be accounted for when we consider what it means to be a successful legislator in the modern Congress.

91 Chapter 4: Running from Washington: Policy Collaboration as Symbolic Representation

Members of Congress frequently seek to distance themselves from Washington,

DC when interacting with their constituents, a phenomenon that has been widely documented in the legislative politics literature, most notably by Fenno (1978). As he writes in Home Style, “If a member discusses the House as an institution in order to point out its institutional strengths, he or she runs the risk of being associated with an unpopular institution. So members tend their own constituency relations and even attack Congress from time to time to reenforce their customized political support at home” [p.246].

For many members of Congress, this is a rational decision. The U.S. Congress is consistently one of the least popular institutions in the country, with an approval rating under 30% since the fall of 2009, while a majority of voters generally approve of their own representative (Gallup, 2017). Although there is some evidence that despite the historical disparity between the approval ratings for individual members of Congress and the institution as a whole, members’ reelection prospects are affected by the popularity of Congress (Born, 1990; McDermott and Jones, 2003; Jones, 2010), research has shown that individual members fare poorly in their reelection campaigns

92 when they are seen as too in step with their party over their constituents (e.g. Canes-

Wrone, Brady and Cogan, 2002; Carson et al., 2010). As a result, we see members of

Congress frequently “running away from Washington” in their campaigns, attempting to portray themselves as different and therefore better than their colleagues inan unpopular legislature.

How do members of Congress establish this distance between themselves and their counterparts in Washington, DC? Much has been written about the relationship between voting records and electoral performance, while less attention has been paid to the other ways members present themselves as different to their constituents. In this paper, I argue that one of the ways in which members of Congress attempt to differentiate themselves from their colleagues is by collaborating with those samecol- leagues. Policy collaboration in Congress serves as a form of symbolic representation that members use to promote an image of themselves as collaborative and bipartisan when they are in their districts.

Using a measure of policy collaboration constructed from a novel dataset of Dear

Colleague letters sent by members of Congress between 2003 and 2011, I examine the relationship between policy collaboration and electoral margins. I find that members of Congress whose seats are considered safe are less likely to collaborate with their colleagues. However, for the most electorally vulnerable members, collaborating with more colleagues in the other party is associated with a significant increase in vote share. For members who have to fight for their seat every two years, engaging inmore bipartisan collaboration in Washington, DC allows them to cultivate a reputation among their constituency for being willing to “work across the aisle” with members of the other party and improves their electoral performance.

93 Collaboration in Washington

Collaboration is a widespread yet understudied practice in Congress. Much of the research on policy collaboration in Congress has used the cosponsorship network as the measure of collaboration, which its proponents concede is a noisy measure that cannot distinguish between policy agreement and policy collaboration (Fowler, 2006a;

Kirkland and Gross, 2012). Cosponsorship is a valuable tool for members of Congress who wish to show their support for a particular policy. Members cosponsor legislation for a number of reasons, including position taking, signaling support to constituents and other outside interests who can help a member in their reelection efforts, and intra-legislative signaling to advance favored policy initiatives (Kessler and Krehbiel,

1996; Wawro, 2000; Koger, 2003; Highton and Rocca, 2005).

However, cosponsorship is not necessarily indicative of collaboration. While some cosponsors may be involved in the drafting and promotion of legislation, the majority have little engagement on the bill beyond reviewing the contents and agreeing to add their name. This is evidenced by both the number of cosponsors who sign onto legislation and the timing of their support. The average bill in the 111th Congress had 18 cosponsors, with the total number of cosponsors on any bill ranging from one to 425.44 And most of those cosponsors are added to the bill after it has been introduced. If they sit on the committee of jurisdiction they may have the opportunity to sign onto the bill during a committee mark up, but for the most part, cosponsors are accepting the underlying bill as-is.

I argue that true policy collaboration involves members of Congress working together on legislation or other policy issues in a substantive and purposive manner.

44Data are from the Congressional Bills Project (Adler and Wilkerson, 2003-2012).

94 Collaborators work together to shape the substance of legislation and agree on the language they use to promote their policy initiatives. However, there is a cost to collaboration. Members may have to compromise on the content of legislation in order to reach agreement with their collaborator and legislation takes longer to draft when more people are involved. Yet, they still choose to do so. In a series of interviews with over twenty congressional staff, I asked how often their boss collaborated with colleagues on policy initiatives and all but one staff member reported that it wasa common occurrence. Responses to the open-ended question “How often does your boss coauthor legislation and policy letters with other members?” included “All the time,” “A lot of time,” “Frequently,” “Very often,” and “He doesn’t always work well with others.” When asked why their boss chose to collaborate, staff reported the perceived legislative benefits, particularly of bipartisan collaboration, the messaging benefits, and a belief that collaboration is a normative good. “He believes inacol- laborative approach and the best way to enact public policy is to write policy that has broad support” (Interviews, 2016).

While it may not be apparent in the legislative record as it only allows for a single member to be listed as the sponsor of each bill, these sort of coauthoring relationships are a frequent occurrence in the modern Congress. In the 111th Congress,

98% of members collaborated with at least one colleague and the average member collaborated with 20 of her colleagues. Members of Congress choose their collaborative partners on the basis of strategic considerations, personal relationships, and shared policy goals and a plurality of these relationships are bipartisan (Craig, 2016).

Members who adhere to the chamber norm in their collaborative patterns and those

95 who collaborate with other effective members of Congress are more likely to beef- fective themselves (Craig, 2015). But collaboration also allows members of Congress to return to their districts and declare that they are working with their colleagues, particularly those on the other side of the aisle to distinguish themselves from a legislature bound by gridlock and acrimony.45

Collaboration at Home

Much of the research on Congress considers the legislative and electoral work that members of Congress engage in to exist in two separate arenas, or is focused on policy congruence between a legislator and her constituents. In one of the seminal works linking the policy preferences of a member’s constituency to their voting behavior,

Miller and Stokes (1963) find that a representative’s voting behavior is influenced by both their own preferences and their perception of their constituents’ policy preferences, but neither the member or constituents have a clear understanding of the other’s policy preferences. Other works connecting voting behavior to constituency preferences have argued that members’ behavior in either the legislative or electoral arena is influenced by their expected outcomes in the other (Denzau, Riker andShep- sle, 1985), members are more likely to represent the position of the district’s median voter in ideologically homogenous districts (Gerber and Lewis, 2004), and members compensate for poor information on constituency opinion by using previous vote margins to guide their voting (Henderson and Brooks, 2016).

45In theory, a member of Congress could return to their district and claim that they are collaborating with colleagues without actually doing so. However, if they are making the claim frequently enough for it to be a prominent messaging point, there is a significant risk of being exposed through opposition research if they do not have evidence to support their claim.

96 However, Eulau and Karps (1977) argue that policy congruence is merely one form of responsiveness and we should consider representation as consisting of four different mechanisms by which a member represents their district: policy responsiveness, service responsiveness, allocation responsiveness, and symbolic responsiveness.46 In the

Eulau and Karps framework, symbolic responsiveness is distinct from the other three forms in that it does not necessarily require a particular behavior from members of

Congress, but rather “involves public gestures of a sort that create a sense of trust and support in the relationship between representative and represented” [p.241]. Studies of the relationship between symbolic representation and electoral performance have found that members with a larger legislative portfolio are viewed more favorably by their constituents, as are senior members who have had time to build a relationship with their voters. Electoral support increased for members who introduced more bills as well as more local interest bills, while members who gave more floor speeches were punished for their showmanship (Box-Steffensmeier et al., 2003). Senators who are electorally vulnerable are more likely to give floor speeches expressing identification with or empathy for their constituents, particularly if they are up for reelection (Hill and Hurley, 2002) and there is also evidence that members of Congress take positions in their direct communications to voters that are more in line with constituency preferences while voting behavior is more likely to reflect the position of the party (Grose and Middlemass, 2010).

I argue that while collaboration in Washington, DC has clear policy implications, it is also a form of symbolic representation that members use to differentiate themselves from their colleagues in Congress. Members, particularly those from electorally

46Subsequent scholars have expanded the concept of representation to include descriptive representation and collective national representation (e.g. Mansbridge, 1999; Weissberg, 1978).

97 vulnerable districts, like to present themselves to their constituents as being “above the fray” when it comes to congressional in-fighting. One example is the 2013 creation of the “Problem Solvers Caucus,” a bipartisan legislative member organization that meets regularly “to build a sense of trust in the spirit of bipartisanship” (Tam, 2013).

The caucus has legislative goals and are working to establish their influence within the House, but there is a clear electoral interest as well. The six members who lead the caucus in the 115th Congress all represent competitive districts and frequently tout their involvement in press releases and social media.47

Similarly, members of Congress promote their collaborative relationships to their constituents. In an editorial printed in The Record and promoted on Facebook and

Twitter, Congressman Josh Gottheimer (D-NJ) shared a report on his first hundred days in Congress with his voters, repeatedly emphasizing the collaborative nature of the legislation he introduced. “I [...] introduced the Regulatory Improvement Act with a group of colleagues on both sides of the aisle to cut out-of-date red tape and help our businesses grow.” He also described how he was “working to pass bipartisan legislation with Republican Brian Fitzpatrick to keep lead out of the drinking water in our children’s schools” and “helped introduce bipartisan legislation to create a comprehensive plan to destroy ISIS” (Gottheimer, 2017). Members of the majority party engage in this sort of promotion as well, as seen in figure 4.1 which provides two examples of how Congressman Carlos Curbelo (R-FL) used Twitter to promote his collaborations with Congressman Marc Veasey (D-TX) and Congressman Scott Peters

(D-CA). Notably these are congressional communications sent a year and a half prior

47The six caucus leaders are Rep. Josh Gottheimer (D-NJ), Rep. Tom Reed (R-NY), Rep. Carlos Curbelo (R-FL), Rep. Kurt Schrader (D-OR), Rep. Mike Gallagher (R-WI), and Rep. Tom Suozzi (D-NY).

98 Figure 4.1: Promotion of Policy Collaboration via Twitter

to the next election, suggesting that rather than using collaboration as a campaign messaging strategy, Congressman Curbelo is trying to build a reputation for himself as someone who works with other members of Congress on both sides of the aisle. Of course, these collaborative relationships can play into campaign messaging as well, as evidenced by Congressman Tim Walberg’s (R-MI) 2016 campaign ad in which a community leader said, “He’s out there fighting for us. The legislation that’s out there you’ll see it’s all bipartisan. You’ll see that Tim’s leading that charge. Tim is able to work with the other side. He’s out there doing this because it’s the right thing to do. And we need more of that in this country.”48 Members of Congress are engaging in collaboration, frequently with members of the other party, and then highlighting those collaborative relationships in their communications with their constituents.

For many members of Congress, using collaboration as a form of symbolic representation makes sense. We know that there is electoral value in position-taking, even if members are not successful in advancing their legislation (Mayhew, 1974). We

48Advertisement script retrieved from www.walbergforcongress.com.

99 also know that while members generally do not position themselves at the ideological median of their district, they are punished at the ballot box if they are seen as being too out of step with their district or too partisan (Clinton, 2006; Canes-Wrone,

Brady and Cogan, 2002; Carson et al., 2010). And the media tend to reward these collaborative relationships with positive coverage. Congressman Scott Rigell (R-VA) received positive coverage in Virginia’s largest daily newspaper when he collaborated with three other members to form the “Fix Congress Now” caucus, a bipartisan pre- decessor to the “Problem Solvers Caucus” (Bartel, 2012). In June 2017, at least three different local newspapers ran articles touting the bipartisanship of their local members of Congress. The Milwaukee Journal Sentinel praised Representatives Ron Kind

(D-WI) and James Sensenbrenner (R-WI) for earning positive scores on the Lugar

Center’s “Bipartisan Index,” Louisiana’s largest daily newspaper printed a feature on the friendship and collaborations between Representatives Steve Scalise (R-LA) and

Cedric Richmond (D-LA), and the newspaper of State College, PA highlighted the multiple collaborations between Representatives Glenn Thompson (R-PA) and Jim

Langevin (D-RI) (Gilbert, 2017; Grace, 2017; Falce, 2017).

At the same time, collaboration, particularly with members of the other party, may not always be in the member’s best interest. While voters want Congress as an institution to be bipartisan, partisan voters want partisan representatives and people prefer partisan legislation over the ideal of bipartisan compromise, which is seen as “losing” (Harbridge and Malhotra, 2011; Harbridge, Malhotra and Harrison,

2014). Furthermore, members of the majority party benefit electorally from a strong, homogenous party and a positive approval of both their party and Congress as a whole

(Hall and Shepsle, 2014; Jones, 2010, 2015). For members representing safe districts

100 and those in the majority party, this suggests that they would be best served by representing the partisan interests of their constituents and promoting their party’s leadership rather than running away from Congress. Electorally vulnerable members, however, would be better off distancing themselves from their party and an unpopular legislature, instead using symbolic representation to promote an image of themselves as a bipartisan, independent representative. This theory yields four hypotheses:

H1: Electorally safe members of Congress will engage in less collaboration with their colleagues.

H2: For electorally vulnerable members, engaging in more collaboration will be associated with an increase in electoral performance.

H3: Electorally vulnerable members who engage in more bipartisan collaboration will see a greater increase in their electoral performance.

H4: For members of the majority party, engaging in more bipartisan collaboration will be associated with a decrease in electoral performance.

Dear Colleague Letters

To test these hypotheses and better understand the relationship between policy collaboration and electoral performance, we first must find a meaningful measure of collaboration between members of Congress. As detailed above, the dominant measure of collaboration in the literature is cosponsorship, which cannot distinguish between members who coauthor legislation together and those who cosponsor legislation to demonstrate their support for a policy. In some instances, such as Congressman

Gottheimer’s legislation on lead in drinking water with Congressman Fitzpatrick, the role that Congressman Fitzpatrick played can be inferred from the legislative record

101 Figure 4.2: Distribution of Dear Colleague Letters by Purpose

20 Percent of Total Percent

Caucus Cosign Cosponsor Floor General Invitation Letter Purpose

because he is the sole original cosponsor of the bill.49 In other instances, such as

the SUPER Act introduced by Congressman Peters and referenced by Congressman

Curbelo in figure 4.1, the legislative record does not distinguish between Congress-

man Curbelo, who is clearly claiming ownership of the legislation, and the other six

members who were original cosponsors.50 However, with Dear Colleague letters, we can identify the members who collaborate on legislation and other policy initiatives in the formative stage of the legislative process.

49H.R. 2094 in the 115th Congress. 50H.R. 2858 in the 115th Congress.

102 Dear Colleague letters are authored by members of Congress and distributed via electronic listserv to all House member and committee offices (Peterson, 2005).51 The letters have received little systematic study in the legislative politics literature due to previous unavailability of the data, however they are frequently mentioned as a tool that members use to build support for their policy proposals (Campbell, 1982;

Koger, 2003; Krutz, 2005). The letters are most commonly used to build support for legislation in the form of cosponsorship before or shortly after it is introduced, but they are also used to solicit cosigners for policy letters to administrative agencies or committees within Congress, as well as urge members to support or oppose a bill on the floor, recruit members for congressional caucuses, disseminate news articles and reports a member finds interesting, and invite members and staff to events. Figure

4.2 provides the distribution of letters by purpose for the 111th Congress.52 Dear

Colleague letters are a valuable tool for members of Congress as they provide an efficient means for members to quickly disseminate information about their policy initiatives and preferences across Capitol Hill. As described by Congressman David

Price (D-NC) in his account of the process involved in passing the Home Equity Loan

Consumer Protection Act, “Having discovered a promising policy gap and feeling anxious lest other members might be getting similar ideas, I hurried to draft a bill and to circulate a ‘Dear Colleague’ letter inviting other members to join me as cosponsors”

(Price, 2004, p.97-98).53

51The U.S. Senate has their own Dear Colleague system, which is beyond the scope of this study. Bicameral Dear Colleague letters constitute less than one percent of letters sent in the House during the 111th Congress. 52Based on a sample of 1600 letters. 53A more detailed discussion of the history and usage of Dear Colleague letters in Congress can be found in Craig (2015).

103 Figure 4.3: Sample Dear Colleague Letter from 111th Congress

Cosponsor H.R. 2743, The Automobile Dealer Economic Restoration Act

Deadline to become an original cosponsor: 6pm today

Current Cosponsors: Maffei, Kratovil, Van Hollen, Hoyer, McMahon, Bartlett, Sutton, Hall, Posey

Dear Colleague:

Please join us in protecting our nation’s autodealers by cosponsoring H.R. 2743, the Automobile Dealer Economic Restoration Act, which would require Chrysler and General Motors to continue to honor their commitments to auto dealers.

Automobile dealers are one of the largest private sector employers in the United States, providing tens of thousands of local jobs and contributing millions of dollars in tax revenues to states. Forcing the closure of automobile dealers would have an especially devastating economic impact in rural communities, where dealers play an integral role in the community, provide essential services and serve as a critical economic engine. This legislation requires that auto manufacturers in which the Federal Government has an ownership interest, or to which the federal government is a lender, continue to honor this commitment and not deprive economic rights to the dealers.

Previously GM and Chrysler seemingly arbitrarily notified dealers that their relationship was ending, essentially immediately, leaving dealers with millions of dollars invested in car stock, no options for consolidation and little leverage for liquidation. There was no transparency to the system that would shutdown many profitable dealerships that have been local institutions for decades. Furthermore, the decisions lacked any proof from the automakers that these closing would actually benefit them financially. This legislation builds on the efforts of Congress in a letter sent to the Treasury Department Auto Task Force on May 19, and a letter sent to President Obama today.

Specifically, H.R. 2743 accomplishes these goals: • Restores the economic rights of General Motors and Chrysler car dealers as they existed prior to each company’s bankruptcies. • Preserves General Motors and Chrysler car dealers’ rights to recourse under state law • At the request of an automobile dealer, requires General Motors and Chrysler to reinstate franchise agreements in effect prior to each company’s bankruptcies. • Makes clear that the legislation is not intended to make null and void the court-ordered transfer of assets from Chrysler LLC to New CarCo Acquisition LLC or the transfer of General Motors assets that could be approved by a court after the introduction of the Act.

If you have any questions or would like to become a cosponsor please contact Hasan Sarsour of Rep. Maffei’s staff at [email protected] or Ben Abrams of Rep. Kratovil's staff at [email protected].

Daniel B. Maffei Frank M. Kratovil Member of Congress Member of Congress

104 While a slight majority of Dear Colleague letters are signed by a single member of

Congress, 46% of the letters sent between 2003 and 2011 were signed by two or more members. When members sign a Dear Colleague letter together, I argue that it is the internal messaging equivalent of Congressman Curbelo’s tweets about the legislation he introduced with Representatives Veasey and Peters. Members are jointly claiming ownership over the legislation or policy initiative in question and signaling to their colleagues that it is a collaborative effort. A sample letter is presented in figure A.

In this letter, Congressman Daniel Maffei (D-NY) and Congressman Frank Kratovil

(D-MD) are promoting the legislation they authored to protect automobile dealers who were harmed by the bankruptcies of General Motors and Chrysler. The bill was introduced by Congressman Maffei in June 2009 and by the end of the111th

Congress was supported by 286 cosponsors, of which Congressman Kratovil was one.

However, even in the earliest stage of the legislative process, on the day that the bill was introduced, the letter distinguishes Congressman Kratovil from the other eight members who had signed on as cosponsors and are listed at the top. The

Automobile Dealer Economic Restoration Act was coauthored by Representatives

Maffei and Kratovil and the associated Dear Colleague letter sends a clear signal to other members that the bill is a collaborative effort.

Dear Colleagues are internal communications within the House so an argument could be made that they are not an appropriate measure of symbolic representation as they are intended for other members of Congress rather than constituents. It is certainly unlikely that the average voter has ever seen a Dear Colleague letter.

However, what I am interested in is the relationships that can be identified within each letter rather than the letters themselves. While members of Congress may talk

105 about their collaborative work in press releases, newsletters, and social media, Dear

Colleague letters allow for the measurement of each member’s overall propensity for collaboration and bipartisanship. Furthermore, while members of Congress may not write Dear Colleague letters for their constituents, they are a part of the public record and are often provided to outside interests so members are conscious that the contents of any letter may reach their constituents (Straus, 2012). Even if some of the relationships in Dear Colleague letters were not entered into with the intent that they be broadcast to the public, it is unlikely that any member would sign a letter with a collaborator they wanted to keep secret from their constituents.

Data and Methods

The policy collaboration network is drawn from the 82,712 Dear Colleague letters sent via electronic listserv in the House of Representatives between 2003 and 2011.54

In order to capture the collaborative relationships between members of Congress, I first must identify the signatures on each letter, which I accomplish with apython script. Each email is read by the script and the content that follows the valediction

“Sincerely,” is identified.55 In addition to the members who signed each letter, I identify the date and Congress that the letter was sent, the subject line of the email, the relevant issue areas as designated by congressional staff, and the body of the letter itself. The result is an affiliation matrix for each Congress consisting of all the letters sent and the names of the members who signed each one.

54Corresponding Congresses are the 108th, 109th, 110th, and 111th. 55About ten percent of letters use alternate valedictions such as “Yours Truly” or lack formal signa- ture blocks. These letters were flagged by the script and hand-coded.

106 Each Congress is its own network as the membership of the House changes with each election as do the relationships within it. Some collaborative relationships persist from one Congress to the next as members reintroduce legislation and policy initiatives that were unsuccessful in the previous Congress and members in certain institutional positions will continue to collaborate on different issues, such as the chairman and ranking member of a committee. However, many of the relationships change from one Congress to the next as issues rise and fall in prominence and political dynamics both inside and outside of the House shift. I use the 108th, 109th,

110th, and 111th Congresses in this study for two reasons. First, three different political dynamics are represented: unified Republican control (108th, 109th), divided government with a Democratic House and Senate and Republican President (110th), and unified Democratic control (111th). Second, I am confident I have the complete policy collaboration network for each of these four Congresses which allows me to ensure the robustness of my network measures (Borgatti, Carley and Krackhardt,

2006).

The affiliation matrix for each Congress yields a two-mode network in which members are connected to the letters they signed. However, for the purposes of this study, what I am interested in is not which members signed each letter but which members signed letters with each other at any point during a given Congress which reflects the collaborative tendencies of members rather than how collaborative each letter is. I therefore create one-mode projections of each two-mode network of letters and members so that each member of Congress is represented in the network by a node and the ties between them are a binary indicator of whether two members signed a Dear Col- league letter together at any point during that Congress (Breiger, 1974; Wasserman

107 and Faust, 1994).56 Ties in the network are undirected as it is unclear in many of the letters which member initiated the collaborative relationship. As members claim credit for the bills they coauthor regardless if they were the one who introduced it in the House this should not affect my measure of propensity for collaboration. Again, see figure 4.1 for an illustration of how members message their collaborations. Con- gressman Curbelo is not the lead sponsor of either of the bills he is promoting on

Twitter, but a voter reading those tweets would not know that unless they looked up the legislation.

Figure 4.4 shows the ego networks for Congressman Henry Cuellar (D-TX) and

Congressman Glenn Thompson (R-PA) in the 111th Congress to illustrate what the collaborative relationships for each member look like. The ego networks show all of the colleagues that each member collaborated with during that Congress with the nodes sized to represent the number of collaborations for each member in the larger network. Congressman Cuellar, denoted as “HC” in the first network, signed Dear

Colleague letters with fourteen of his colleagues, only two of whom were Republicans.

He represents a relatively safe district and was elected in 2008 with 69% of the vote and again in 2010 with 56% of the vote. Congressman Thompson is shown in the second graph as collaborating with thirteen of his colleagues in the 111th Congress, eleven of whom were Democrats. He was elected in 2008 with 57% of the vote and again in

2010 with 69% of the vote. Both members were close to the median number of total collaborations for the 111th Congress, which was sixteen. Regarding the partisanship

56In some instances members, particularly those who co-chair caucuses together or are the chair and ranking member of a committee or subcommittee, have multiple collaborations in a single Congress. This is something I intend to explore in future work examining the strength of ties in the policy collaboration network. However, first I must distinguish unique collaborations in the data from repeated messages on the same bill to properly weight the edges.

108 Figure 4.4: Ego Networks for Congressman Henry Cuellar (D-TX) and Congressman Glenn Thompson (R-PA) in the 111th Congress (2009-2010)

109 of their ties, Congressman Cuellar was less bipartisan than the median Democrat

who collaborated with seven Republicans, while Congressman Thompson was more

bipartisan than the median Republican who collaborated with nine Democrats.

From the policy collaboration networks I create two measures of propensity for

collaboration for each member in a given Congress. The first, Total Collaboration

is the degree centrality score for each member in the network, which is simply a

count of the other members that each member collaborates with (Freeman, 1979).

So for Congressman Cuellar in the 111th Congress, his degree centrality, or Total

Collaboration is 14, while Congressman Thompson’s is 13. The expectation given by H1 is that an increase in a member’s electoral margin will be associated with a decrease in Total Collaboration as electorally secure members collaborate with fewer of their colleagues. The second collaboration measure is Bipartisan Collaboration, which is a count of the number of collaborations each member has with colleagues in the other party. For Congressman Cuellar in the 111th Congress, his Bipartisan

Collaboration score is two, while Congressman Thompson’s is 11. The expectation for this measure is that its effects will be conditioned on whether a member is electorally vulnerable or a member of the majority party, as discussed in further detail below.

For both of these measures I use the count of total collaborations rather than a normalized or proportional measure because even a highly informed voter following a member on social media is likely to notice the number of times a member mentions their collaborations with colleagues rather than the proportion of all messages that feature collaboration.57

57In a separate analysis I examine the relationship between bipartisanship and electoral margin using the percentage of a member’s collaborations that are bipartisan and find no result.

110 My expectation, based on the varying voter perceptions of compromise and bipartisanship depending on partisan strength and alignment, is that the electoral benefits of collaboration and bipartisanship will be concentrated among more vulnerable incumbents. To test hypotheses two and three, I first create a dichotomous measure of whether a member is Vulnerable based on their performance in the previous election.

All members who were elected with less than 60% of the vote at t − 1 are coded as one, while those who received 60% more of the vote in their previous election are coded as zero. I use 60% as the cut off rather than the more traditional 55% because my data cover a somewhat volatile period in congressional elections in which three of the four elections saw one party gain twenty or more seats in the House. Of the 31

Democratic members in the 111th Congress who were elected with between 55 and

59% of the vote in 2008, only fifteen of them were reelected in 2010. With the electoral vulnerability measure, I create two interaction terms: Vulnerable*Total, which is the dichotomous vulnerability variable interacted with each member’s Total Col- laboration, and Vulnerable*Bipartisan, which is Vulnerable interacted with Bipartisan

Collaboration for each member.

The congressional elections literature also suggests that members of the majority party benefit from a strong party and therefore should be less inclined to collaborate with minority party members, instead collaborating more with members of their own party. If majority party members working with colleagues in the other party hurts majority party approval or improves minority party approval, we may see a negative electoral effect for bipartisanship conditional on whether a member is in the majority party. To test this hypothesis, I create an indicator variable, Majority which is one if

111 a member is in the majority party in a given Congress and zero otherwise and interact

it with Bipartisan Collaboration to create the interaction term Majority*Bipartisan.

The dependent variables in the first two models testing the relationship between past electoral performance and propensity for collaboration are Total Collaboration and Bipartisan Collaboration. Electoral Margin is the dependent variable in the models testing hypotheses two through four. Electoral Margin is drawn from Federal

Election Commission records and represents the percentage of the vote each member

received in the election following a given Congress. So for the 111th Congress which

ran from 2009-2010, the electoral margin recorded for each member is the percentage

of the vote they received in 2010. Members who did not run for reelection or ran

unopposed are excluded from the analysis.58 As an independent variable in the first

model, I use Electoral Margint−1, or the percentage of the vote each member received

in the election prior to a given Congress. Table 4.1 provides the summary statistics

for the dependent and key independent variables for each of the four Congresses.

Because I am working with network data and my key independent variables all

reflect the degree to which members of Congress work with their colleagues invarious

forms, I use a temporal network autocorrelation model (TNAM) to account for the

relational dynamics in each model. The TNAM provides a way around the assumption

in standard regression models that the observations are independent from one another

and allows me to account for the possibility that the electoral margin of member i

58Members were defined as unopposed if they had no challenger from the other major party and any third party candidates in the race did not receive more than 2% of the vote. Members who lost in the primary election were excluded from the analysis, which removed a total of twelve members across all four Congresses. Within the time period covered here, it was very rare for a member to lose in a primary outside of redistricting scenarios where two incumbents were pitted against one another. An analysis of current data would have to more carefully consider the role of primary challenges as they became a greater threat after the 111th Congress.

112 Table 4.1: Summary Statistics for Dependent and Key Independent Variables

Congress Variable Min Median Mean Max N 108 Total Collaboration 0.00 18.00 21.64 97.00 439 Bipartisan Collaboration 0.00 7.00 8.24 44.00 439 Vulnerable 0.00 0.00 0.20 1.00 439 Vulnerable*Total 0.00 0.00 3.39 61.00 439 Vulnerable*Bipartisan 0.00 0.00 1.30 35.00 439 Majority 0.00 1.00 0.53 1.00 439 Majority*Bipartisan 0.00 0.00 4.12 44.00 439 Electoral Margin 38.00 67.00 70.02 100.00 402 Congress Variable Min Median Mean Max N 109 Total Collaboration 0.00 22.00 26.14 97.00 438 Bipartisan Collaboration 0.00 8.00 10.20 66.00 438 Vulnerable 0.00 0.00 0.20 1.00 438 Vulnerable*Total 0.00 0.00 4.30 97.00 438 Vulnerable*Bipartisan 0.00 0.00 1.93 66.00 438 Majority 0.00 1.00 0.54 1.00 438 Majority*Bipartisan 0.00 0.00 5.10 66.00 438 Electoral Margin 39.00 65.00 67.51 100.00 404 Congress Variable Min Median Mean Max N 110 Total Collaboration 0.00 18.00 21.52 93.00 447 Bipartisan Collaboration 0.00 9.00 10.56 70.00 447 Vulnerable 0.00 0.00 0.30 1.00 447 Vulnerable*Total 0.00 0.00 5.48 93.00 447 Vulnerable*Bipartisan 0.00 0.00 3.16 70.00 447 Majority 0.00 1.00 0.54 1.00 447 Majority*Bipartisan 0.00 1.00 5.28 35.00 447 Electoral Margin 40.00 66.00 68.55 100.00 397 Congress Variable Min Median Mean Max N 111 Total Collaboration 0.00 16.00 20.28 90.00 443 Bipartisan Collaboration 0.00 7.00 9.30 55.00 443 Vulnerable 0.00 0.00 0.30 1.00 443 Vulnerable*Total 0.00 0.00 4.52 66.00 443 Vulnerable*Bipartisan 0.00 0.00 2.45 48.00 443 Majority 0.00 1.00 0.59 1.00 443 Majority*Bipartisan 0.00 2.00 4.65 31.00 443 Electoral Margin 33.44 63.09 63.31 100.00 395

113 at time t is influenced in some way by the electoral margin of member j at time t − 1. The model builds on the spatial autocorrelation model, which accounts for non-independence of observations at a single time point by incorporating a spatial weight matrix W multiplied by the dependent variable Y so it captures the outcome values of an observation’s neighbors. In the network context, we use the adjacency matrix A as the spatial weight matrix (White, Burton and Dow, 1981). The adjacency matrix is the NxN representation of the connections in each network. In the case of the policy collaboration network, each Congress is its own adjacency matrix in which a one indicates that two members signed a Dear Colleague letter together in that congress and zero indicates that they did not. The result is that AY reflects the sum of the electoral margins for all of member i’s collaborators.

Y = α + ρAY + Xβ + ϵ

However, my interest is not limited to a single Congress and so I use the temporal network autocorrelation model developed by Leifeld, Cranmer and Desmarais (2014) and implemented in the xergm package. The TNAM is a flexible structure that can account for various network and temporal dependencies. In this case, I make use of two key features. First, the temporal component allows me to include all four

Congresses in the model rather than estimating a separate spatial autocorrelation model for each one. Second, it allows for various operationalizations of the adjacency matrix. In the models testing hypotheses two through four, my expectation is that it is not the sum of the electoral margins for all of member i′s collaborators that will be correlated with their own performance, but rather the average electoral return of their collaborators. Therefore, I normalize the adjacency matrix by row so that every outgoing tie for member i is weighted equally. The weight for the degree to which

114 member i is influenced by member j is then:

aij wij = ai

in which wij represents the observation in the weight matrix corresponding to member j’s influence on member i. Using the adjacency matrix, aij = 1 if member i and

member j collaborate and ai is the sum of the ith row of the adjacency matrix (Leen-

ders, 2002). The resulting weight matrix W is multiplied by the outcome variable, in this case Electoral Margin, and WY then captures the average electoral margin for all of member i’s collaborators in the election, Collaborator Margin. I expect that members of Congress will collaborate with colleagues who see similar electoral returns. A homophily effect will be present as safe members will collaborate with other safe members, while vulnerable members will work with others whose seats are less secure.59

I also incorporate eight control variables that are commonly used in the congressional elections literature, following Canes-Wrone, Brady and Cogan (2002) and

Carson et al. (2010). Ideological Extremism is calculated using each member’s first- dimension DW-NOMINATE score provided by Poole and Rosenthal (2011). The extremity score tests whether members who are far from the ideological center of

Congress are punished at the ballot box and is calculated by taking the absolute value of the difference between each member’s DW-NOMINATE score and the chamber median for that Congress. I expect that there will be a negative relationship between ideological extremity and electoral margin, consistent with Canes-Wrone,

59The model does not distinguish between homophily and influence, however as I lack a theoretical explanation for why the electoral margin of member i would be influenced by the average electoral margin of her collaborators in the same election, I assume if an effect is present, it is the result of homophily.

115 Brady and Cogan (2002).60 I also control for District Partisanship as the share of the two-party vote that the presidential candidate from a member’s party received in the district in the most recent election. This is a commonly used measure of district preferences and I expect that members representing districts where the presidential candidate from their party performed well will see a corresponding increase in their own electoral returns (Jacobson, 2009).

Spending Gap is the difference between the campaign spending of the challenger and the incumbent as recorded by the Federal Elections Commission, calculated so that higher values of Spending Gap reflect races where the challenger matched or outspent the incumbent. This also serves as an indicator of the competitiveness of the race and I expect higher challenger spending relative to incumbent spending will be associated with a decrease in electoral margin. Freshman is incorporated as a dummy variable indicating whether a member is serving their first term in Congress with the expectation that first term members will have less of an incumbency advantage than more senior members (Canes-Wrone, Brady and Cogan, 2002).

I also include four controls that reflect the political conditions likely to affecta member’s electoral performance following Carson et al. (2010). In Party is a dichotomous variable reflecting whether a member of Congress is in the same party asthe

President. Midterm is a categorical variable that captures whether the election following a given Congress is a midterm election, and if so, whether the member is in the president’s party to account for the widely documented phenomenon that members of the president’s party are more likely to perform poorly in midterm elections.

60Carson et al. (2010) make a compelling case that this is not the result of ideologically extreme members being punished, but rather members who are seen as being too loyal to their party particularly as the parties move further apart in ideology. This may be the case, but teasing out the different effects of ideological extremity and party unity are beyond the scope of thisstudy.

116 This variable is coded as -1 for members of the president’s party when the election following a given Congress is a midterm election, 1 for members of the out party in midterm election years, and 0 for presidential election years. Presidential Approval is measured using the national level responses to the Gallup poll question, “Do you approve or disapprove of the way [president] is handling his job as president?” taken from the most recent poll completed before the election. I subtract 50% from the approval rating and multiply it by -1 for members who are not in the president’s party so that popular presidents benefit members of their party while unpopular presidents help members of the other party. Finally, Change in Personal Income is the national level percent change in real personal income measured by the Bureau of Economic

Analysis from the beginning of the election year to its end. Again, this is multiplied by -1 for members of Congress who are not in the president’s party to account for the assumption that a growing economy will benefit in-party members, while a sluggish economy will benefit out-party members.

Analysis

The results for the models testing the relationships between Electoral Margint−1 and Total Collaboration and Bipartisan Collaboration are provided in table 4.2.61 As predicted in my first hypothesis, electorally secure members are less likely to engage in collaboration in general and bipartisan collaboration. A one percentage point increase in a member’s vote share in the previous election is associated with a -0.036 decrease in the total number of collaborators a member has in the next Congress and a

-0.086 decrease in the number of collaborators in the other party. This is particularly

61A more complete analysis of the determinants of collaboration can be found in Craig (2016). This is intended merely as a test of the relationship between electoral performance and collaboration.

117 Table 4.2: Relationship Between Past Electoral Performance and Collaboration

Total Bipartisan ∗∗∗ ∗∗∗ Electoral Margint−1 −0.036 −0.086 (0.008) (0.009) Ideological Extremism −1.971∗∗ −6.156∗∗∗ (0.742) (0.819) Freshman −1.178∗∗∗ −2.796∗∗∗ (0.336) (0.374) Majority 0.723 −3.368∗∗∗ (0.491) (0.543) Midterm Election 0.194 −0.466∗ (0.211) (0.234) Collaborator Total 0.023∗∗∗ (0.000) Collaborator Bipartisan 0.024∗∗∗ (0.000) Constant 6.704∗∗∗ 12.804∗∗∗ (0.769) (0.854) AIC 10238.733 10589.909 BIC 10282.550 10633.726 Log Likelihood -5111.367 -5286.955 Deviance 33675.391 41079.481 Num. obs. 1767 1767 ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05

noteworthy as Straus (2012) finds that electorally secure members are more likely to send Dear Colleague letters, suggesting that when members who were elected with close margins do send Dear Colleagues, they are much more likely to be collaborative letters.

The remainder of the findings are largely as expected. Members of Congress who are further from the ideological median in a given Congress are less likely to collaborate with colleagues, as are first term members. As collaboration necessitates a

118 certain amount of compromise as members work together to craft a bill or policy ini-

tiative, it is understandable that more ideologically extreme members are less likely

to find collaborative partners, particularly collaborative partners in the other party.

Members in their first term are not as established in Congress and have fewer personal

connections in the chamber which may make it difficult for them to find collaborative

partners. Members of the majority party are less likely to collaborate with the other

party than minority party members, but there is no apparent relationship between

being a majority party member and total collaboration. This fits with the findings in

Craig (2016) that there is little benefit to minority party members collaborating with

each other. I include Midterm Election in this model to examine whether members of the president’s party are more likely to collaborate when they are going into a midterm election and the results suggest they may, at least when it comes to bipartisan collaboration. Members of the president’s party going into a midterm election have, on average 0.466 more bipartisan collaborations than all members going into a presidential election.62 Collaborator Total and Collaborator Bipartisan represent the

AY term in each model, in this case neither weighted nor lagged. The results show

that both are positive and significant, suggesting a homophily effect: high collabora-

tors work with other high collaborators.

Next, I look at the relationship between collaboration and vote share, conditioned

on electoral vulnerability. Models 1 and 2 in table 4.3 show the results. From model

1 we see that there is no apparent relationship between Total Collaboration and

62The remainder of the election-specific variables that are included in the next set of models are excluded in table 4.2 because while members of Congress know when they are going into a midterm election, they do not know what the president’s approval rating or economic conditions will be going into the election so there is no reason to expect those terms to influence collaboration. District Partisanship is excluded due to high collinearity with Electoral Margint−1

119 Table 4.3: Relationship Between Collaboration and Electoral Performance

Model 1 Model 2 Model 3 Total Collaboration −0.023 (0.012) Bipartisan Collaboration −0.044 −0.013 (0.027) (0.034) Vulnerable −7.585∗∗∗ −7.691∗∗∗ (0.710) (0.666) Vulnerable*Total 0.050 (0.027) Vulnerable*Bipartisan 0.113∗ (0.046) Majority −4.144∗∗∗ (1.135) Majority*Bipartisan 0.018 (0.046) Collaborator Margin 0.031∗ 0.030∗ 0.032∗ (0.015) (0.015) (0.016) Ideological Extremism −3.204∗ −3.531∗ −3.484∗ (1.476) (1.471) (1.560) District Partisanship 0.557∗∗∗ 0.556∗∗∗ 0.640∗∗∗ (0.021) (0.021) (0.021) Spending Gap −0.000 −0.000 −0.000 (0.000) (0.000) (0.000) Freshman −0.539 −0.362 −3.860∗∗∗ (0.601) (0.612) (0.588) In-Party −5.110∗∗∗ −5.098∗∗∗ (1.155) (1.153) Midterm Election −4.333∗∗∗ −4.506∗∗∗ −3.216∗∗∗ (0.539) (0.530) (0.593) Presidential Approval 0.021 0.014 0.167∗∗∗ (0.031) (0.031) (0.019) ∆ Personal Income 0.261 0.276∗ 0.775∗∗∗ (0.136) (0.135) (0.120) Constant 36.904∗∗∗ 37.114∗∗∗ 29.729∗∗∗ (1.592) (1.631) (1.619) AIC 10067.105 10066.028 10240.202 BIC 10141.453 10140.376 10309.239 Log Likelihood -5019.552 -5019.014 -5107.101 Deviance 71919.645 71867.905 80849.770 Num. obs. 1496 1496 1496 ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05

120 Electoral Margin for vulnerable members of Congress. Despite the findings in table

4.2 that members of Congress who were elected with a smaller percentage of the vote engage in more collaborative relationships, there is no evidence of a benefit to that collaboration when considered over all relationships. However, as model 2 shows, there is a distinct benefit to bipartisan collaboration for vulnerable members. As

Vulnerable is a dichotomous variable, the interpretation of the interaction is relatively straightforward. For members who received less than 60% of the vote in the previous election, each additional bipartisan collaborator that they work with is associated with a 0.069 percentage point increase in their electoral margin in the subsequent election. The effect is small in magnitude when considered in terms of addinga single collaborative relationship, but consider the effect in terms of the difference between a member with a high number of bipartisan collaborations who has built a reputation among their constituents as someone who is willing to work across the aisle versus a member who works predominately within their own party. In the

111th Congress, vulnerable members collaborated with anywhere between zero and

48 members of the other party. A one standard deviation increase in the number of bipartisan collaborators a member works with is associated with a 0.57 percentage point increase in vote share. When dealing with members who may win reelection by margins of one percent or less, this is a significant benefit. The results of model 3in table 4.3 show no evidence in support of my fourth hypothesis. An increase in the number of bipartisan collaborators a member works with has no effect for members of either party.

The model results also show evidence of a homophily effect as collaborating with other members who receive a higher percentage of the vote is associated with an

121 increase in a member’s electoral margin.63 More important than the size of any

homophily effect, this demonstrates the value in accounting for the non-independence

of members in electoral models as the electoral margin for a member of Congress is

clearly related to the average electoral margin of her collaborators. Most of the other

model controls yield results as expected. Ideological Extremism has a strong negative

effect on electoral margin, while District Partisanship has a clear positive effect. The

stronger the district-level support is for the presidential candidate from the member’s

party in the most recent presidential election, the better off a member will be in their

reelection campaign, while members whose voting records place them far from the

chamber median ideologically are penalized. Members of the president’s party do

poorly, particularly in midterm elections.64

There are a few surprising findings among the controls which warrant further

study. First, Spending Gap has no apparent effect on electoral margin in any model

where it typically has a negative effect. It is unclear if this is a peculiarity unique to

the 2004-2010 congressional elections possibly due to the wave elections in 2006 and

2010 or the result of accounting for the interdependence of observations.65 Models

1 and 2 also yield insignificant coefficients on Freshman and Presidential Approval

and model 1 shows no effect for Change in Personal Income. This is somewhat surprising, as these are generally reliable predictors of electoral margin, however as they yield the expected results in model 3, I assume the null results are the byproduct

63In an alternate specification, I use Electoral Margin at t-1 to test the possibility that collaborating with members who did well in the previous election has some benefit, however I found no support for that theory. 64It should be noted that the midterm election years represented in the data are 2006 and 2010, both of which resulted in the president’s party losing control of the House. 65Excluding the collaboration measures does not alter this result. It may also be the result of measurement error, which is another reason further investigation is needed.

122 of including the Vulnerable variable which includes the observations most likely to be impacted by their lack of incumbency advantage or the larger political and economic circumstances.

Discussion and Conclusion

Bipartisanship is frequently seen as a normative good. Political dysfunction of all stripes including a gridlocked Congress is attributed to excessive partisanship by the public and in the media. And as Harbridge and Malhotra (2011) show, the public generally wants Congress as an institution to be bipartisan. However, recent scholarship has increasingly suggested that voters favor partisanship in their representation and compromise of any sort is viewed as losing (Harbridge and Malhotra, 2011; Har- bridge, Malhotra and Harrison, 2014). One might infer from recent scholarship that reelection minded members of Congress would be best served by becoming partisan bomb-throwers with little interest in compromise or collaboration. However, as I demonstrate here, the world may not be quite so bleak.

By providing the first measure of substantive collaboration among members of

Congress, I demonstrate that there is value in collaboration as a form of symbolic representation. When presented with specific issues, voters may prefer their position over any sort of compromise but they still reward members who engage in more bipartisan collaboration and are able to build a history of being someone willing to work across the aisle. The results presented here show that electorally safe members engage in less collaboration, which is a rational decision on their part as they receive no apparent benefit for collaboration with members of either party. However, for electorally vulnerable members, collaborating with members from the other party and

123 creating a reputation for bipartisanship can have significant electoral returns. Voters in marginal districts reward their representatives for distinguishing themselves from their more partisan colleagues through the symbolic act of cross-partisan collaboration.

This paper represents a first attempt at examining the relationship between collaboration and electoral returns, accounting for the symbolic representation and position taking that members engage in in addition to their voting behavior. Uncovering the relationship between bipartisanship and electoral performance provides important insights for our understanding of symbolic representation and congressional elections.

Future work will examine the impact of collaboration across different types of policy efforts to determine whether members are benefitting from collaboration on substantive policy issues or merely symbolic initiatives. One of the key assumptions made in this paper is that members of Congress explicitly promote their collaborative relationships to their constituencies in the district and thereby develop a reputation for bipartisanship. While that assumption is strongly supported by anecdotal evidence in the form of op-eds, social media postings, and press releases issued by members, work to more explicitly connect collaboration to these communications will also be explored now that this study has shown it to be a fruitful area of research. The result will be a greater understanding of how members of Congress use their activities in

Washington to promote their desired image of themselves at home and what that means for the functioning of the U.S. House of Representatives.

If there is a normative good to bipartisanship, it is all the more reason for con- cern regarding the increasing efforts to create electorally safe congressional districts.

Although there is evidence that the incumbency advantage may be on the decline

124 (Jacobson, 2015), members representing districts crafted to ensure representation by a particular party are more likely to win by significant margins and therefore have no electoral incentive to work with members in the other party. This is detrimental to Congress as an institution. As described by one congressional staff member, “The benefit of continuing to work with members on different issues is you just builda better relationship with those people. I have a better relationship with the staff on

[the member’s committee] and the [member’s state] delegation and they’ll think of

[the member] when they’re introducing a bill that they’re trying to get support for”

(Interviews, 2016). A Congress where fewer members collaborate with each other or only collaborate within their own party is one where there are fewer relationships between members and greater discord. Uncovering this relationship between electoral safety and partisanship should provide new ammunition to those arguing for a less partisan redistricting process as more members elected in close races will result in increased bipartisanship within Congress.

125 Chapter 5: Conclusion

In the preceding three essays on policy collaboration in the U.S. House of Represen- tatives, I demonstrate that collaboration between members of Congress is widespread, that it is driven by strategic considerations, personal relationships, and shared policy goals, and that the greatest legislative benefits are for members who are moderate in their collaborative decisions, avoiding collaborating either too much or too little, while electoral benefits are strongest for vulnerable members who work across the aisle and are able to project an image of bipartisanship to their constituents.

With the introduction of new data that allow us to measure substantive policy collaboration between members of Congress, this represents only the beginning of what will be a thorough examination of the role of collaboration in Congress and its effects for individual members of Congress, for specific policy proposals, andfor the legislature as a whole. These new data also allow for the reexamining of several of our longstanding theories and assumptions about congressional politics with the understanding that legislation is not always single-authored, and members are not independent from one another.

The next phase of my research agenda will revolve around two tracks. The first will continue the study of collaboration in Congress, examining some of the unanswered questions raised in the previous chapters such as the benefits of collaboration for

126 individual policies rather than members, and reconsidering how we measure legislative effectiveness. The second will tap into other insights that Dear Colleague letters can provide regarding the formative stage in the policy process, such as the issues that members choose to prioritize in their portfolios, and the language they use when communicating with each other to promote their agenda.

The next step in the policy collaboration research is to shift to a bill-level analysis that will build on my legislative effectiveness findings to further examine one of the repeated assertions in my interviews with congressional staff: that bipartisan legislation is generally more successful. This paper will examine which sorts of bills and policy initiatives are more likely to be collaborative and then compare outcomes for legislation that is single authored to those of bills that are the result of both co-partisan and cross-partisan collaboration. Recent advances in data collection will allow me to consider not only whether collaborative bills are more likely to pass, but also whether they are more likely to be incorporated into other legislation which then goes on to pass as I expect they will be. I expect that bills that are moderate in scope and represent second tier policy issues will see greater success in the legislative process when they are collaborative and when the measure of success accounts for the bundling of legislation.

Another issue raised in the preceding studies that can now be addressed with Dear

Colleague data is the failure to account for collaboration when considering what constitutes legislative success. Once all of a bill’s coauthors are associated with its success or failure and the fate of bills that are incorporated into other pieces of legislation is accounted for, this will allow for the creation of a much more comprehensive measure of legislative success that gives credit to members who coauthor successful bills and

127 considers incorporation into a successful bill just as valid a measure of success as passing your own legislation. With such a measure I will be able to reexamine longstanding questions on the determinants of legislative effectiveness with a particular emphasis on the value of strategies available to members of the minority party.

Other possible studies involving the policy collaboration network may include examining the evolution of issue networks over multiple Congresses and the associated methodological issues with network comparisons, the ego networks of individual members and how they relate to legislative style and effectiveness, studying the policy collaboration network as a multiplex network in conjunction with committee and caucus networks, and leveraging an unexpected shock to the network such as the 2009 party switch of Representative Parker Griffith from Democrat to Republican to study the effects of changes in policy collaboration relationships.

Finally, I intend to present a more nuanced view of the policy collaboration network in a study that accounts for the strength of the collaborative relationships between members of Congress and, when possible, the direction of those relationships.

Which members are more likely to initiate collaboration, which members are more desired collaborators, and why? What is the relationship between strong cross-partisan ties and vote agreement? How does the strength of ties in the network affect the legislative process and the functioning of Congress itself? Although there are a few measurement hurdles to overcome first, the Dear Colleague data in conjunction with qualitative evidence will allow me to identify the members who initiated collaborative relationships, and the number of times that members collaborated on unique policy initiatives.

128 The second, and broader track of my research agenda uses Dear Colleague letters to further our understanding of congressional decision-making and the policy process by examining the strategies that members of Congress use to appeal to six key audiences: voters, colleagues, the party, interest groups, contributors, and the media.

Policy collaboration will play a part in these studies, but the broader interest is in how members choose to prioritize and promote different aspects of their policy agenda.

Understanding congressional behavior through the framework of members and their audiences represents an evolution of our traditional goal-oriented theories of Congress to reflect the constraints and complexities of the 21st century legislature.

Members of Congress are undeniably motivated by reelection, good public policy, and institutional advancement (Fenno, 1973a; Mayhew, 1974). However, Congress has changed since the writings of Fenno and Mayhew in several important ways. Power has shifted from congressional committees to party leadership, floor procedures are more restrictive, and the combination of the 24-hour news networks, online distribution of printed news, and rise of political blogs and social media have diminished candidate control over the news and information disseminated about them to their constituents. As a result, while members may be motivated by reelection, good public policy, and institutional advancement, an individual member of Congress today has far less control over whether she achieves those goals than in the 1970s.

Dear Colleague letters provide a window into the earliest stages of the legislative process. They will allow us to distinguish between the bills that members are serious about advancing and those that are introduced for the purposes of position-taking either to stake a claim on an issue among members of Congress, or show support for a local constituency. As multiple members of Congress have recounted in interviews

129 and memoirs, when they want to build support for a bill, they do so by sending out a

Dear Colleague letter. Yet only a third of the bills introduced in the House have Dear

Colleague letters associated with them. Examining which bills members of Congress choose to devote their limited resources towards advancing will provide rich insights into questions of reputation and policymaking.

Understanding the true policy priorities of individual members of Congress will also lead to a greater understanding of the institutional constraints within the House.

One of the more striking observations of the policy collaboration network in Congress is how bipartisan it is compared to the cosponsorship network and the co-voting network, suggesting that the leadership’s agenda control in the modern Congress may have an even greater role in fostering partisan divisions in Congress than previously suspected. One of the longstanding questions of congressional politics is the degree to which political parties influence the legislative process. I intend to use Dear Colleague letters to examine issue attention within the House, comparing the issues that rank and file members of Congress prioritize and send Dear Colleagues on to the priorities of the leadership as determined by the floor agenda. This will allow for a greater understanding of the leadership’s role and provide important insights into the power

(or lack thereof) of individual members of Congress when it comes to policy change.

Dear Colleague letters along with recent advances in text analysis also allow for the study of the language that members use when promoting their policy agenda, and in particular, whether the language used in internal communications is different from that used in external communications. Available datasets of press releases and newsletters sent by members of Congress make this a potentially rich and fruitful avenue of study, although I expect that contrary to public perception, the language

130 members of Congress use in their internal communications will be strikingly similar to their public messages. If this is the case, it will only make any marked differences in communication style that much more noticeable.

What policy collaboration, agenda prioritization and the language that members use all have in common is that they are actions that individual members take that are almost entirely under their own control. Whereas votes are limited to the bills the leadership is willing to consider, committee assignments are dictated by senior party members, and even media coverage is shaped by the issues the media finds interesting, Dear Colleague letters allow for the examination of the issues that members of Congress choose to prioritize, the colleagues they choose to work with, and the language they choose to promote their agenda. Which brings us back to the idea of Congress and its audiences. Across all three actions, I expect members of Congress to engage in different strategies to appease each of the six distinctive, yet overlapping audiences. Some of these distinctions are already apparent from this study of policy collaboration as I revealed that for vulnerable members of Congress, bipartisan collaboration was a powerful signal for the audience of voters, while policy-minded legislators can foster beneficial relationships with their colleagues in the House by striking the right balance between collaborating enough to build relationships and be a team player, while not collaborating so much that they appear as a dilettante.

Members may prioritize one audience over another, and indeed in many cases, such as with the vulnerable members of Congress, this can be crucial. Prioritization can also evolve over time, depending on where the member is in their career. Some favor a more district-focused approach to legislating, and others prefer the national spotlight,

131 but all six audiences are critical to a member who wishes to be successful in achieving their goals and cannot be neglected entirely.

132 Bibliography

Adler, Scott E. and John Wilkerson. 2003-2012. “Congressional Bills Project: (2003-

2012), NSF 00880066 and 00880061.”.

Adler, Scott E. and John Wilkerson. 2005. “The Scope and Urgency of Legislation:

Reconsidering Bill Success in the House of Representatives.” Presented at the 2005

American Political Science Association Annual Meeting, Washington, DC.

Anderson, William D, Janet M Box-Steffensmeier and Valeria Sinclair-Chapman.

2003. “The Keys to Legislative Success in the US House of Representatives.” Leg-

islative Studies Quarterly 28(3):357–386.

Aral, Sinan, Lev Muchnik and Arun Sundarajan. 2009. “Distinguishing Influence-

Based Contagion from Homophily-Driven Diffusion in Dynamic Networks.” Pro-

ceedings of the National Academy of Sciences of the United States of America

106(51):21544–21549.

Arnold, Laura W, Rebecca E Deen and Samuel C Patterson. 2000. “Friendship and

Votes: The Impact of Interpersonal Ties on Legislative Decision Making.” State &

Local Government Review pp. 142–147.

Bartel, Bill. 2012. “Rigell, Other Lawmakers Start Fix Congress Group.” The

Virginian-Pilot May 17, 2012.

133 Binder, Sarah A. and Frances E. Lee. 2015. Making Deals in Congress. In Solutions

to Political Polarization in America, ed. Nathaniel Persily. Cambridge University

Press.

Black, Duncan. 1948. “On the Rationale of Group Decision Making.” Journal of

Political Economy 56(1):23–34.

Borgatti, Stephen P. 2005. “Centrality and Network Flow.” Social Networks 27:55–71.

Borgatti, Stephen P. and Daniel S. Halgin. 2011. “Analyzing Affiliation Networks.”

The Sage Handbook of Social Network Analysis pp. 417–433.

Borgatti, Stephen P., Kathleen M. Carley and David Krackhardt. 2006. “On the

Robustness of Centrality Measures Under Conditions of Imperfect Data.” Social

Networks 28:124–136.

Born, Richard. 1990. “The Shared Fortunes of Congress and Congressmen: Members

May Run from Congress, but They Can’t Hide.” Journal of Politics 52(4):1223–

1241.

Box-Steffensmeier, Janet M., David C. Kimball, Scott R. Meinke and Katherine Tate.

2003. “The Effects of Political Representation on the Electoral Advantages ofHouse

Incumbents.” Political Research Quarterly 56(3):259–270.

Box-Steffensmeier, Janet M., Dino Christenson and Jason W. Morgan. Forthcoming.

“Modeling Unobserved Heterogeneity in Social Networks with the Frailty Exponen-

tial Random Graph Model.” Political Analysis .

134 Box-Steffensmeier, Janet M., Dino P. Christenson and Alison W. Craig. 2013. “Cue-

Taking in Congress: Interest Group Signals from Dear Colleague Letters.” Pre-

sented at the 2013 American Political Science Association annual meeting, Wash-

ington, DC.

Breiger, Ronald L. 1974. “The Duality of Persons and Groups.” Social forces

53(2):181–190.

Burris, Val. 2004. “The Academic Caste System: Prestige Hierarchies in PhD Ex-

change Networks.” American Sociological Review 69:239–264.

Campbell, James E. 1982. “Cosponsoring Legislation in the US Congress.” Legislative

Studies Quarterly pp. 415–422.

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(01):127–140.

Carson, Jamie L., Gregory Koger, Matthew J. Lebo and Everett Young. 2010. “The

Electoral Costs of Party Loyalty in Congress.” American Journal of Political Sci-

ence 54(3):598–616.

Cho, Wendy K Tam and James H Fowler. 2010. “Legislative Success in a Small

World: Social Network Analysis and the Dynamics of Congressional Legislation.”

The Journal of Politics 72(1):124–135.

Clinton, Joshua. 2006. “Representation in Congress: Constituents and Roll Calls in

the 106th House.” Journal of Politics 68(2):397–409.

135 Cox, Gary W. and William C. Terry. 2008. “Legislative Productivity in the 93rd-105th

Congresses.” Legislative Studies Quarterly 33(4):603–618.

Craig, Alison W. 2015. “Lone Wolves and Team Players: Policy Collaboration Net-

works and Legislative Effectiveness in the House of Representatives.” Presented as

a part of the 2015 Legislative Studies Virtual Workshop Series.

Craig, Alison W. 2016. “The Room Where it Happens: Collaborative Strategies in the

U.S. House of Representatives.” Presented at the 2016 American Political Science

Association Meeting, Philadelphia, PA.

Craig, Alison W., Skyler J. Cranmer, Bruce A. Desmarais, Christopher J. Clark and

Vincent G. Moscardelli. 2015. “The Role of Race, Ethnicity, and Gender in the

Congressional Cosponsorship Network.” Presented at the 2015 Political Networks

Conference, Portland, OR.

Cranmer, Skyler J. and Bruce A. Desmarais. 2011. “Inferential Network Analysis with

Exponential Random Graph Models.” Political Analysis 19:66–86.

Crisp, Brian F., Kristin Kanthak and Jenny Leijonhufvud. 2004. “The Reputations

Legislators Build: With Whom Should Representatives Collaborate?” American

Political Science Review 98(4):703–716.

Denzau, Arthur, William Riker and Kenneth A. Shepsle. 1985. “Farquharson and

Fenno: Sophisticated Voting and Home Style.” American Political Science Review

79(4):1117–1134.

136 Desmarais, Bruce A., Vincent G. Moscardelli, Brian F. Schaffner and Michael S.

Kowal. 2015. “Measuring Legislative Collaboration: The Senate Press Events Net-

work.” Social Networks 40:43–54.

Desmarais, Bruce and Skyler J. Cranmer. 2012. “Statistical Mechanics of Networks:

Estimation and Uncertainty.” Physica A 391:1865–1876.

Downs, Anthony. 1957. “An Economic Theory of Political Action in a Democracy.”

Journal of Political Economy 65(2):135–150.

Eulau, Heinz and Paul D. Karps. 1977. “The Puzzle of Representation: Specifying

Components of Responsiveness.” Legislative Studies Quarterly 2(3):233–254.

Falce, Lori. 2017. “Rep. Gleen Thompson Talks Importance of Bipartisanship in

Congress.” Centre Daily Times June 19, 2017.

Fenno, Richard F. 1973a. Congressmen in Committees. Little, Brown Boston.

Fenno, Richard F. 1973b. The Internal Distribution of Influence: The House. In

Congress and America’s Future, ed. David Truman. Englewood Cliffs, NJ: Prentice

Hall pp. 52–76.

Fenno, Richard F. 1978. Home Style: House Members in Their Districts. Longman

Publishing Group.

Fowler, James H. 2006a. “Connecting the Congress: A Study of Cosponsorship Net-

works.” Political Analysis 14(4):456–487.

Fowler, James H. 2006b. “Legislative Cosponsorship Networks in the US House and

Senate.” Social Networks 28(4):454–465.

137 Fowler, Linda L. and Bryan W. Marshall. 2017. “Veto-Proof Majorities, Legislative

Procedures, and Presidential Decisions, 1981-2008.” Political Research Quarterly

70(2):348–362.

Frantzich, Stephen. 1979. “Who Makes Our Laws? The Legislative Effectiveness of

Members of the U.S. Congress.” Legislative Studies Quarterly 4(3):409–428.

Freeman, Linton C. 1979. “Centrality in Social Networks Conceptual Clarification.”

Social Networks 1:215–239.

Friedman, Sally. 1993. “Committee Advancement of Women and Blacks in Congress.”

Women and Politics 13(2):27–52.

Gallup. 2017. “Congress and the Public.” http://www.gallup.com/poll/1600/congress-

public.aspx.

Gerber, Elizabeth R. and Jeffrey B. Lewis. 2004. “Beyond the Median: Voter Pref-

erences, District Heterogeneity, and Political Representation.” Journal of Political

Economy 112(6):1364–1383.

Gilbert, Craig. 2017. “Ranking Wisconsin’s Federal Lawmakers for ‘Bipartisanship’.”

Milwaukee Journal Sentinel June 5, 2017.

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.

Gottheimer, Josh. 2017. “A Hundred-Plus Days of Working for You.” The Record

April 28, 2017.

138 Grace, Stephanie. 2017. “A Bipartisan Friendship Plays out in Congress, and on the

Baseball Diamond.” The Advocate June 14, 2017.

Grose, Christian R. and Keesha M. Middlemass. 2010. “Listen to What I Say, Not

How I Vote: Congressional Support for the President in Washington and at Home.”

Social Science Quarterly 91(1):143–167.

Grossmann, Matt and David Hopkins. 2016. Asymmetric Politics: Ideological Repub-

licans and Group Interest Democrats. Oxford University Press.

Hall, Andrew B. and Kenneth A. Shepsle. 2014. “The Changing Value of Seniority

in the US House: Conditional Party Government Revised.” Journal of Politics

76(1):98–113.

Hanneke, Steve, Wenjie Fu and Eric P. Xing. 2010. “Discrete Temporal Models of

Social Networks.” Electronic Journal of Statistics 4:585–605.

Harbridge, Laurel. 2015. Is Bipartisanship Dead? Policy Agreement and Agenda-

Setting in the House of Representatives. Cambridge University Press.

Harbridge, Laurel and Neil Malhotra. 2011. “Electoral Incentives and Partisan Con-

flict in Congress: Evidence from Survey Experiments.” American Journal of Polit-

ical Science 55(3):494–510.

Harbridge, Laurel, Neil Malhotra and Brian F. Harrison. 2014. “Public Prefer-

ences for Bipartisanship in the Policymaking Process.” Legislative Studies Quarterly

39(3):327–355.

139 Hare, Christopher and Keith T. Poole. 2016. Measuring Ideology in Congress. In

Handbook of Social Choice and Voting, ed. Jac C. Heckelman and Nicholas R.

Miller. Northampton, ma ed. Edward Elgar Publishing.

Hasecke, Edward B. and Jason D. Mycoff. 2007. “Party Loyalty and Legislative

Success.” Political Research Quarterly 60(4):607–617.

Henderson, John and John Brooks. 2016. “Mediating the Electoral Connection: The

Information Effects of Voter Signals on Legislative Behavior.” Journal of Politics

78(3):653–669.

Highton, Benjamin and Michael S. Rocca. 2005. “Beyond the Roll-Call Arena:

The Determinants of Position Taking in Congres.” Political Research Quarterly

58(2):303–316.

Hill, Kim Quaile and Patricia A. Hurley. 2002. “Symbolic Speeches in the U.S. Senate

and Their Representational Implications.” Journal of Politics 64(1):219–231.

Hunter, David R. 2007. “Curved Exponential Family Models for Social Networks.”

Social Networks 29(2):216–230.

Hunter, David R, Mark S Handcock, Carter T Butts, Steven M Goodreau and Martina

Morris. 2008. “ergm: A Package to Fit, Simulate and Diagnose Exponential-Family

Models for Networks.” Journal of Statistical Software 24(3):nihpa54860.

Interviews, Staff. 2013. “Personal interviews conducted by the author with member

and CAO staff in the House.”.

Interviews, Staff. 2016. “Personal interviews conducted by the author with member

staff in the House.”.

140 Jacobson, Gary C. 2009. The Politics of Congressional Elections. 7th edition ed. New

York: Pearson.

Jacobson, Gary C. 2015. “It’s Nothing Personal: The Decline of the Incumbency

Advantage in US House Elections.” Journal of Politics 77(3):861–873.

Jones, David R. 2010. “Partisan Polarization and Congressional Accountability in

House Elections.” American Journal of Political Science 54(2):323–337.

Jones, David R. 2015. “Partisan Polarization and the Effect of Congressional Perfor-

mance Evaluations on Party Brands and American Elections.” Political Research

Quarterly 68(4):785–801.

Kessler, Daniel and Keith Krehbiel. 1996. “Dynamics of Cosponsorship.” American

Political Science Review 90(3):555–566.

King, Peter. 2010. “The James Zadroga 9/11 Health and Compensation Act of 2010.”

Congressional Record 156(133):H7245.

Kingdon, John W. 1989. Congressmen’s Voting Decisions. University of Michigan

Press.

Kirkland, Justin H. 2011. “The Relational Determinants of Legislative Outcomes:

Strong and Weak Ties Between Legislators.” The Journal of Politics 73(03):887–

898.

Kirkland, Justin H and Justin H Gross. 2012. “Measurement and Theory in Leg-

islative Networks: The Evolving Topology of Congressional Collaboration.” Social

Networks .

141 Koger, Gregory. 2003. “Position-Taking and Cosponsorship in the US House.” Leg-

islative Studies Quarterly 28:225–246.

Krutz, Glen S. 2005. “Issues and Institutions: “Winnowing” in the US Congress.”

American Journal of Political Science 49(2):313–326.

Lee, Frances E. 2009. Beyond Ideology: Politics, Principles, and Partisanship in the

U.S. Senate. Chicago, IL: University of Chicago Press.

Leenders, Roger Th.A.J. 2002. “Modeling Social Influence Through Network Auto-

correlation: Constructing the Weight Matrix.” Social Networks 24:21–47.

Leifeld, Philip. 2013. “texreg: Conversion of Statistical Model Output in R to LATEX and HTML Tables.” Journal of Statistical Software 55(8):1–24.

Leifeld, Philip and Skyler J. Cranmer. 2015. “The Temporal Network Autocorrelation

Model (TNAM).” Presented at the 2015 Political Networks Conference, Portland,

OR.

Leifeld, Philip, Skyler J. Cranmer and Bruce A. Desmarais. 2014. “xergm: Extentions

for Exponential Random Graph Models.”.

Leifeld, Philip and Volker Schneider. 2012. “Information Exchange in Policy Net-

works.” American Journal of Political Science 53(3):731–744.

Levy, Michael. 2016. “gwdegree: Improving Interpretation of Geometrically-Weighted

Degree Estimates in Exponential Random Graph Models.” Journal of Open Source

Software 1(3).

142 Lipinski, Daniel. 2009. Navigating Congressional Policy Processes: The Inside Per-

spective on How Laws Are Made. In Congress Reconsidered, ed. Lawrence C. Dodd

and Bruce I. Oppenheimer. 9 ed. Congressional Quarterly.

Maloney, Carolyn. 2010. “The James Zadroga 9/11 Health and Compensation Act of

2010.” Congressional Record 156(173):H8961.

Manning, Jennifer E. and Collen J. Shogan. 2012. African American Members of

the United States Congress:1870-2012. Technical Report RL30378 Congressional

Research Service.

Manning, Jennifer E. and Ida A. Brudnick. 2014. Women in the United States

Congress, 1917-2014: Biographical and Committee Assignment Information, and

Listings by State and Congress. Technical Report RL30261 Congressional Research

Service.

Mansbridge, Jane. 1999. “Should Blacks Represent Blacks and Women Represent

Women? A Contingent Yes.” Journal of Politics 61(3):628–657.

Masket, Seth. 2008. “Where You Sit is Where You Stand: The Impact of Seating

Proximity on Legislative Cue-Taking.” Quarterly Journal of Political Science 3:301–

311.

Masters, Nicholas A. 1961. “Committee Assignments in the House of Representatives.”

American Political Science Review 55(2):345–357.

Matthews, Donald R. 1959. “The Folkways of the United States Senate: Conformity

to Group Norms and Legislative Effectiveness.” American Political Science Review

53(4):1064–1089.

143 Matthews, Donald R and James A Stimson. 1975. Yeas and Nays: Normal Decision-

Making in the US House of Representatives. Wiley New York.

Mayhew, David R. 1974. Congress: The Electoral Connection. Yale University Press.

McCollum, Betty. 2014. “McCollum Introduces Bipartisan Legislation to Improve

Access to Health Care for Native Americans.” Congresswoman Betty McCollum

Press Releases.

McDermott, Monika L. and David R. Jones. 2003. “Do Public Evaluations of Congress

Matter? Retrospective Voting in Congressional Elections.” American Politics Re-

search 31(2):155–177.

Merica, Dan. 2013. “Washington Gridlock Linked to Social Funk.” CNN Politics

January 25, 2013.

Miler, Kristina C. 2007. “The View from the Hill: Legislative Perceptions of the

District.” Legislative Studies Quarterly 32(4):597–628.

Miller, Warren E. and Donald E. Stokes. 1963. “Constituency Influence in Congress.”

American Political Science Review 57(1):45–56.

Office of the Clerk, US House of Representatives. 2014. “Resume of Congressional

Activity.” Congressional Record, 80th Congress - 113th Congress.

Office of the Historian and Office of the Clerk. 2013. Hispanic Americans inCongress

1822-2012. Technical Report 108-225 U.S. House of Representatives.

Oleszek, Mark J. 2010. Collaboration and Lawmaking in the Contemporary United

States Senate PhD thesis University of California, Berkeley.

144 Peoples, Clayton D. 2008. “Interlegislator Relations and Policy Making: A Sociological

Study of Roll-Call Voting in a State Legislature1.” Sociological Forum 23(3):455–

480.

Peterson, R. Eric. 2005. “Dear Colleague Letters: A Brief Overview.” Congressional

Research Service Congressional Report No. RS21667.

Poole, Keith T and Howard L Rosenthal. 2011. Ideology and congress. Vol. 1 Trans-

action Books.

Price, David E. 2004. The Congressional Experience. 3rd edition ed. Cambridge, MA:

West View Press.

Ringe, Nils and Jennifer Nicoll Victor. 2013. Bridging the Information Gap: Leg-

islative Member Organizations in the United States and the European Union. Ann

Arbor: University of Michigan Press.

Ringe, Nils, Jennifer Nicoll Victor and Justin H Gross. 2009. “Keeping Your Friends

Close and Your Enemies Closer? Information Networks in Legislative Politics.”

British Journal of Political Science 1(1):1–28.

Rippere, Paulina. 2016. “Polarization Reconsidered: Bipartisan Cooperation Through

Bill Cosponsorship.” Polity 48(2):243–278.

Rogowski, Jon C and Betsy Sinclair. 2012. “Estimating the Causal Effects of Social

Interaction with Endogenous Networks.” Political Analysis 20(3):316–328.

Rossman, Gabriel, Nicole Esparza and Phillip Bonacich. 2010. “I’d Like to Thank the

Academy, Team Spillovers, and Network Centrality.” American Sociological Review

75(1):31–51.

145 Snijders, Tom A. B., Philippa E. Pattison, Garry L. Robins and Mark S. Handcock.

2006. “New Specifications for Exponential Random Graph Models.” Sociological

Methodology 36:99–153.

Stewart, Charles and Jonathan Woon. 2011. “Congressional Committee Assignments,

103rd to 112th Congresses, 1993-2011.”.

Straus, Jacob R. 2012. “Use of ’Dear Colleague’ Letters in the US House of Repre-

sentatives: A Study of Internal Communications.” Journal of Legislative Studies

19:60–75.

Tam, Ruth. 2013. “Members of Congress Introduce Bipartisan Legislation as the

‘Problem Solvers’.” Washington Post July 18, 2013.

Thomas, Cal. 2017. “When Members of Congress Knew Each Other.” The Day April

3, 2017.

Tong, Lorraine H. 2013. Asian Pacific Americans in the United States Congress.

Technical Report 97-398 Congressional Research Service.

Ver Hoef, Jay M. and Peter L. Boveng. 2007. “Quasi-Poisson vs. Negative Bino-

mial Regression: How Should We Model Overdispersed Count Data?” Ecology

88(11):2766–2772.

Victor, Jennifer Nicoll. 2013. “U.S. Congress Caucus Data.” Data available for schol-

arly use from http://mason.gmu.edu/ jvictor3/Data/.

Volden, Craig and Alan E. Wiseman. 2014. Legislative Effectiveness in the United

States Congress: The Lawmakers. New York: Cambridge University Press.

146 Warren, Mark. 2014. “Help, We’re in a Living Hell and Don’t Know How to Get

Out.” Esquire .

Wasserman, Stanley and Katherine Faust. 1994. Social Network Analysis: Methods

And Applications. Cambridge University Press.

Wawro, Gregory. 2000. Legislative Entrepreneurship in the U.S. House of Represen-

tatives. Ann Arbor: University of Michigan Press.

Weissberg, Robert. 1978. “Collective vs. Dyadic Representation in Congress.” Amer-

ican Political Science Review 72(2):535–547.

White, Douglas R., Michael L. Burton and Malcolm M. Dow. 1981. “Sexual Division

of Labor in African Agriculture: A Network Autocorrelation Analysis.” American

Anthropologist 83(4):824–849.

Wimmer, Andreas and Kevin Lewis. 2010. “Beyond and Below Racial Homophily:

ERG Models of a Friendship Network Documented on Facebook.” American Jour-

nal of Sociology 116(2):583–642.

Young, James Sterling. 1966. The Washington Community, 1800-1828. Columbia

University Press.

147 Appendices

148 Appendix A: Sample Dear Colleague Letters

For illustration, I present three sample Dear Colleague letters in Figures A.1, A.2 and A.3. The first letter shows a bipartisan network of signers: Representatives Zach

Space (D-TN), Lee Terry (R-NE), Marion Berry (D-AR), and Jerry Moran (R-KS).

This letter was sent on December 22, 2009 to urge members to sign a letter that the four members had drafted to the Centers for Medicare and Medicaid Services. The second letter was written and signed by the chairs of the Qatari-American Economic

Strategic Defense, Cultural and Educational Partnership Caucus: Representatives

Nick Rahall (D-WV), Dana Rohrabacher (R-CA), and Carolyn Maloney (D-NY).

This letter, which was sent on December 18, 2009, is urging members to join the

Qatari-American Partnership Caucus. The third letter was sent on June 5, 2009 by a single member of Congress, Representative Laura Richardson (D-CA), and asks members to become original cosponsors of the Equal Rights for Health Care Act, which was introduced on June 9, 2009 as H.R. 2744.

149 Figure A.1: Dear Colleague Example: Community Pharmacies

PRESERVE ACCESS FOR AMERICA’S SENIORS Help Community Pharmacies Combat Burdensome Requirements *** Deadline Today – December 22, 2009 *** **Supported by the National Community Pharmacists Association and the National Association of Chain Drug Stores**

Cosigners: Herseth Sandlin, McMorris Rodgers, Welch

Dear Colleague:

Beginning January 1st, pharmacies will be required to have completed a lengthy and costly accreditation process in order to maintain a supplier number under Medicare Part B to provide durable medical equipment (DME). As a result, many pharmacies have already surrendered their right to provide DME, leaving Medicare beneficiaries without the benefit of access to care for these supplies and services at their local pharmacies.

Earlier this year, the House passed a three month delay of the implementation of these requirements, which was signed into law. Unfortunately, many pharmacies are still struggling to achieve the needed accreditation, and any changes to these accreditation requirements addressed in proposed health care reform legislation will not be implemented prior to the January 1st deadline.

Without an additional delay, America’s seniors will lose access to DME supplies that they so desperately need. Particularly in rural areas, where pharmacies are long distances apart, the resulting limitation of access is particularly troubling.

We invite you to join us in sending the attached letter to CMS, requesting that CMS extend its planned date for pharmacy accreditation and delay in enforcement of these requirements for pharmacies. Please contact Ryan Mann with Congressman Space, Tessie Alexander with Congressman Terry, Cynthia Blankenship with Congressman Berry, or Brian Perkins with Congressman Moran to sign on to the letter.

Sincerely,

//s// //s// //s// //s// ZACK SPACE LEE TERRY MARION BERRY JERRY MORAN

150 Figure A.2: Dear Colleague Example: U.S.-Qatar Relations

Please Join the Qatari-American Economic Strategic Defence, Cultural and Educational Partnership Caucus

Dear Colleague:

We invite you to become a member of the QATARI-AMERICAN ECONOMIC STRATEGIC DEFENCE, CULTURAL AND EDUCATIONAL PARTNERSHIP CAUCUS.

The Caucus serves as a forum to brief Members of Congress on Qatar’s strategic defense opportunities for the U.S. and Qatar, as well as its political, economic, social, academic and cultural prospects for a greater stability in the Middle East. Years of reform and stability in Qatar and its commitment to an open, democratic society with free press, universal suffrage, and private and foreign ownership, have cemented the bonds between our two nations.

The objectives of the Caucus are to be an educational source of information within the Congress on the U.S.-Qatar relationship and to promote a better understanding of that relationship. The Caucus will also strengthen the U.S. discourse on bilateral and regional issue areas by engaging Qatar perspectives on issues of mutual interest.

Qatar and the U.S. have signed a number of agreements expanding defense, commercial and cultural ties. For example, a number of U.S. companies are currently operating under production sharing agreements for enhanced oil recovery/production and are increasing their investments. Also Qatar has established satellite branches of the following institutions: Cornell Medical, Texas A & M, VA Commonwealth and Carnegie Mellon. Also, in the ensuring course of modernization, Qatar is home to the largest pre- positioning of United States military equipment in the world.

We hope you will join us in this opportunity to work toward increasing the strong ties the United States shares with Qatar as the two countries’ relationship continues to grow. Join the Caucus by emailing [email protected] with Rep. Rahall; [email protected] with Rep. Rohrabacher; or [email protected] with Rep. Maloney.

Sincerely,

Nick J. Rahall, II Dana Rohrabacher Carolyn Maloney Member of Congress Member of Congress Member of Congress

151 Figure A.3: Dear Colleague Example: Health Care

Support Equal Rights for Health Care!

***Deadline to become an original co-sponsor THIS MONDAY, June 8th at 12 noon***

Endorsed by the National Minority Quality Forum, Families USA, and Family Equality Council

Original Co-Sponsors: Corrine Brown (FL), Steve Cohen (TN), Donna M. Christensen (USVI), John Conyers (MI), Bob Filner (CA), Carolyn C. Kilpatrick (MI), Carolyn Maloney (NY), Eleanor Holmes Norton (DC)

Dear Colleague:

Please join me in becoming an original co-sponsor of legislation that I believe should be part of the fast-approaching healthcare debate.

As a fundamental right, all Americans should be guaranteed equal access to healthcare, and it is unfortunate that this issue has not yet been adequately addressed. I urge you to join me in reaching a solution by becoming a co-sponsor of the Equal Rights for Health Care Act- Title 42. Inspired by Title IX, the goal of this legislation is to ensure that all Americans are treated equally when obtaining healthcare treatment.

As you know, Title IX was passed in 1972 and created federal law prohibiting discrimination on the basis of sex in any federally funded education activities. Title IX has attracted much attention by leveling the playing field in athletics, but it has also had a tremendous impact on gender equality in all educational activities. Similarly, my legislation will expand upon the belief that Americans should receive equal treatment in all areas of their lives from education to healthcare. Specifically, this legislation will prohibit discrimination in Federal assisted health care services and research programs on the basis of sex, race, color, national origin, sexual orientation, gender identity, or disability status.

With the introduction of this legislation we are taking a step toward equal access to healthcare, and I urge you to become an original co-sponsor the Equal Rights for Health Care Act- Title 42. For further information or to be added as an original co-sponsor please contact Mariel Lim at x5-7924 or [email protected].

Regards,

Laura Richardson Member of Congress

152 Appendix B: Alternate ERGM Specifications

Table B.1 replicates the TERGM produced in table 2.1 but with party mixing covariates rather than majority-minority mixing covariates to examine whether the partisanship of tie formation in the network is driven by the tendency of Democratic members to be more collaborative than their Republican counterparts regardless of which party is in control of the House. As the model shows, bipartisanship still dominates the policy collaboration network, and while the coefficient for Democratic members collaborating with other Democrats, it fails to meet the threshold of statistical significance, just as the Majority-Majority term did.

153 Table B.1: Probability of Policy Collaboration with Party Mixing Covariate

Estimate 2.5% 97.5% Strategic Considerations Bipartisan 0.653∗ 0.343 1.032 Democrat-Democrat 0.127 -0.031 0.283 1st DW-NOMINATE -0.455∗ -0.602 -0.365 2nd DW-NOMINATE -0.197∗ -0.232 -0.175 Electoral Security -0.002 -0.003 0.000

Personal Relationships Triadic Closure 0.445∗ 0.377 0.529 Structural Imbalance 0.030∗ 0.026 0.041 Preferential Attachment 2.679∗ 2.132 3.539 Same Class 0.154∗ 0.138 0.192 First Term -0.101∗ -0.212 -0.011

Shared Policy Goals Same State 1.118∗ 1.034 1.199 Same Committee 0.567∗ 0.546 0.589 Caucus Leader 2.503∗ 2.154 2.884 Same Minority 0.848∗ 0.679 0.930 Female-Female 0.306∗ 0.136 0.582

Controls Delegation Size -0.004∗ -0.005 -0.004 Party Leader -0.012 -0.051 0.036 Bills Introduced -0.002∗ -0.003 -0.001 Opposite Gender -0.150∗ -0.217 -0.091 Edges -8.981∗ -10.685 -8.068 ∗ 0 outside the confidence interval

154