<<

Signaling Influence in the U.S. Congress

by

James P. Bassett, M.A., M.Ed.

A Dissertation

In

Political Science

Submitted to the Graduate Faculty of Texas Tech University in Partial Fulfillment of the Requirements for the Degree of

DOCTOR OF PHILOSOPHY

Approved

Dr. Timothy Nokken Chair of Committee

Dr. Frank Thames

Dr. Joel Sievert

Mark Sheridan Dean of the Graduate School

August, 2020 Copyright 2020, James P. Bassett Texas Tech University, James P. Bassett, August 2020

ACKNOWLEDGEMENTS

I would like to thank my advisor Dr. Tim Nokken for his support throughout this process. I would also like to thank Dr. Frank Thames and Dr. Joel Sievert for their help, as well as Dr. Ian Ostrander for starting me on the path. I am completely indebted to my colleagues in the political science department for their support and feedback, especially Matthew Ellison, Yonathan Hailu, Kerry Chavez, Travis Cole, and James Wright. Thank you for always being willing to read something I wrote or share something you wrote. Thanks to my parents and step-parents for keeping me focused. Thank you to my brother Dr. Will P. Bassett for always being one step ahead, and to Manda and Joe for believing I could do this. Finally, thanks to my partner and teammate Stephanie. I couldn’t have done it without you and I’m so grateful to have had you next to me.

ii Texas Tech University, James P. Bassett, August 2020

TABLE OF CONTENTS

Acknowledgements ...... ii

Abstract ...... iv

List of Tables ...... v

List of Figures ...... vi

1. Introduction ...... 1 1.1 Influence and Covoting Network Centrality ...... 3 1.2 Limitations ...... 6 1.3 Road Map ...... 7 2. Centrality as Influence ...... 10 2.1 Cosponsorship Connectedness ...... 11 2.2 Legislative Effectiveness ...... 15 2.3 Institutional Positions ...... 19 2.4 Marginalized Groups ...... 23 2.5 Context and Setting ...... 29 2.6 Legislative Influence Going Forward ...... 36 3. Centrality Over Time ...... 42 3.1 Revolt and Reform ...... 45 3.1.1 The Revolt of 1910 ...... 46 3.1.2 The Reforms of the 1970s ...... 49 3.1.3 Similarities and Differences ...... 53 3.2 Research Design ...... 56 3.3 Results ...... 60 3.4 Centrality in the 20th Century and Beyond ...... 64 4. Centrality Behind the Scenes ...... 70 4.1 Procedure and Passage in the U.S. Congress ...... 72 4.2 Research Design ...... 79 4.3 Results ...... 82 4.4 Discussion ...... 88 5. Conclusion ...... 91 5.1 Future Research ...... 92 5.2 Overarching Themes ...... 94 Bibliography ...... 97 Appendix ...... 105

iii Texas Tech University, James P. Bassett, August 2020

ABSTRACT

What does it mean to have influence in a legislature? There has been a considerable academic effort to explore many of the aspects of legislative behavior that cause some members to be more successful than others, but relatively little quantitative work has been brought to bear on exactly what influence itself entails. Using a recently developed measure called Covoting Network Centrality, I describe and test a concept known as Signaling Influence, which conceptualizes influence as the way in which influential members’ votes signal their colleagues on how they themselves should vote. I put this concept to a variety of tests, with the U.S. House of Representatives as the case. First, I examine how the measure of Covoting Network Centrality compares to similar extant measures of congressional influence. Second, I consider the impact of institutional reform on how members exert influence on one another. Finally, I explore whether members demonstrate different levels of influence on procedural votes as compared to votes on final passage of a bill. I find that centrality is effective at capturing the influence of party leaders, reformers, and ideologues across a wide variety of model specifications, time periods, and legislative contexts.

iv Texas Tech University, James P. Bassett, August 2020

LIST OF TABLES

2.1 OLS Results: Institutions ...... 22 2.2 OLS Results: Marginalized Groups ...... 26 2.3 OLS Results: Marginalized Group Intersectionality ...... 27 2.4 OLS Results: Marginalized Groups, Democrats Only ...... 28 2.5 OLS Results: Marginalized Group Intersectionality, Democrats Only . 29 2.6 OLS Results: Context and Setting (Extremity) ...... 34 2.7 OLS Results: Context and Setting (Party Difference) ...... 35 2.8 OLS Results: All Covariates ...... 39 3.1 Contingency between Committee Chairs and Insurgents, 57th-67th Congress ...... 58 3.2 OLS Results: 1910 Case ...... 61 3.3 OLS Results: 1977 Case ...... 63 4.1 Vote Counts, 93rd to 110th Congress ...... 76 4.2 Extremity versus Party Difference ...... 81 4.3 OLS Results: Hypotheses 1 and 2 ...... 83 4.4 OLS Results: Hypotheses 3 & 4 ...... 86 A1 Context and Setting, No Southern Democrats ...... 105 A2 OLS Results: Southern Democrats (Post-1994) ...... 107 A3 Hypotheses 1A and 1B: All Covariates ...... 108 A4 Hypotheses 2A and 2B: All Covariates ...... 109 A5 1977 Case, 93rd-108th Congress ...... 110 A6 1910 Case, Omitting Time Since Treatment ...... 111 A7 1977 Case, Omitting Time Since Treatment ...... 112

v Texas Tech University, James P. Bassett, August 2020

LIST OF FIGURES

2.1 Distribution of Connectedness and Centrality ...... 14 2.2 Scatterplot of Connectedness and Centrality ...... 15 2.3 Scatterplot of LES and Centrality ...... 18 2.4 Party Difference and Centrality ...... 36 3.1 Predicted Centrality Before and After 1910 Revolt ...... 62 3.2 Predicted Centrality Before and After 1977 Reforms ...... 64 3.3 Mean Centrality Over Time ...... 66 4.1 Histogram of Centrality on Procedure and Passage ...... 80 4.2 Predicted Centrality - Leadership ...... 84 4.3 Predicted Centrality - Majority Party ...... 85 4.4 Predicted Difference in Centrality - Leadership and Majority Party . 87 4.5 Predicted Difference in Centrality - Party Difference and Extremity . 87 4.6 Scatterplot of Procedure/Passage Centrality Over Time ...... 89 A.1 Scatterplot of LES and Connectedness ...... 106

vi Texas Tech University, James P. Bassett, August 2020

CHAPTER 1 INTRODUCTION

What does it mean to have influence in a legislature? There has been a considerable academic effort to explore many of the aspects of legislative behavior that cause some members to be more successful than others, but relatively little quantitative work has been brought to bear on exactly what influence itself entails. For some it has been a function of their success in passing floor amendments (Sinclair 1989; Smith and Flathman 1989) or even more broad-based measures of success in creating legislation (Volden and Wiseman 2014). Others have measured it as the ability to attract cosponsors for legislation (Fowler 2006). Scholars have investigated the individual characteristics of legislatures in a wide variety of contexts, ranging from ideology and partisanship to legislative skill and their career path. The study of these characteristics has been enormously illuminating when it comes to the study of legislative politics. In particular, the pioneering work on ideology by Poole and Rosenthal (e.g. Poole and Rosenthal 2011) has allowed a much deeper understanding of the incentives and goals that shapes the behavior of members of Congress. However, while these many quantitative measures can teach us much about individual members as they stand on their own, what they cannot tell us with as much clarity is the position that these legislators occupy relative to one another. In particular, they tend to operate on the assumption that members are essentially equal, notwithstanding inequalities in things like leadership positions. By way of example, models of behavior such as those by Krehbiel (2010) that focus on the overall configuration of members’ ideological preferences with regard to pivot points assume that members’ personal attributes are essentially interchangeable – there is

1 Texas Tech University, James P. Bassett, August 2020 no difference in the outcomes when Member A is a filibuster pivot than there would be if Member B were the pivot instead. Even quantitative models which do aim to explore these personal differences, such as Volden and Wiseman (2014) in their conception of Legislative Effectiveness, assume that legislators exist in something of a vacuum, where legislative success is essentially a result of individual effort. Unanswered in these measures is the role of relationships between members. Legislatures are a social setting, and a critical aspect of what makes members successful is their ability to cultivate influence through connections with their colleagues. This idea has been somewhat difficult to systematically explore. Some scholars have attempted to do so using survey data, which faces problems of validity and reliability (Miquel and Snyder Jr 2006; Weissert 1991a,b). More quantitative attempts have utilized bill cosponsorship Fowler (2006), finding that social network analysis of legislators is an effective way of measuring those relationships and identifying members who are the most influential. This concept has also been examined by Ringe and Wilson (2016) in the context of the European Parliament. They likewise show that members in positions which should be presumed to allow them to exert influence on their colleagues do indeed demonstrate higher levels of centrality in covoting networks, which they propose as a measure of influence that occurs from signaling or cueing. While these quantitative measures of influence have been successful in showing the face validity of the concept of influence in legislatures, they have only begun to scratch the surface of what legislative influence can tell us about legislative behavior. As a result, this dissertation aims to consider influence as it occurs in the from both directions; that is, by considering both what makes certain members influential, and what that influence means. In order to unpack these questions, I will be focusing on Ringe and Wilson (2016)’s measure of

2 Texas Tech University, James P. Bassett, August 2020 covoting network centrality (to be discussed at length in the following section). Overall, the driving strategy for this dissertation is to explore new avenues and settings to examine this concept – who is influential, how influence has changed over time, and how it changes in different legislative contexts.

1.1 Influence and Covoting Network Centrality

The main tool to be used to examine this notion of influence will be Ringe and Wilson’s (2016) measure of covoting network centrality. Developed initially for use in the European Parliament, this measure operates on the basis of social network analysis by assuming that members’ votes are related to one another – that an influential member’s decision on a has the capacity to send a signal to many influenced members on how they should cast their own votes. In this conception of influence, roll call votes are indicative of a cueing mechanism. In other words, members are not necessarily seeing a colleauge vote and making their decision based on that. Rather, as Ringe and Wilson point out, “such records include no direct reflection of the network structure that generated the votes in the first place. Voting records imply a network structure” which we are interested in measuring (p. 747, emphasis in original). In their agent-based model, Ringe and Wilson demonstrate that this model is effective at measuring the latent centrality in the network, showing that cue-givers are always more central in covoting networks than cue-receivers, and that this measurement can therefore be used to identify the legislators who are sending signals to their colleagues. In other words, legislators whose votes send a signal to others on how to vote should consistently have a higher score than those whose votes are not valued as a signal.

3 Texas Tech University, James P. Bassett, August 2020

Covoting network centrality is calculated for each member as

n 1 X C = |ρ | i n ij j=1 or the summed absolute value of the Pearson correlation between Legislator i’s voting record and every other legislator j’s voting record, standardized to be bounded by 0 and 1 by dividing the result by the total number of legislators, n. As such, it is essentially a legislator’s “average” correlation with all members of the legislature. In addition, in order to control for the wide disparity in the number of votes that members take, this number is weighted by the percentage of votes that they participated in. This measure of influence has several important advantages. First, it is relatively free of the influence of party leadership (or at least as free as a roll-call voting measure can be). Because it uses the absolute value of the correlation, consistent voting opposite to another legislator has the same effect as consistently voting with another legislator. In other words, a low centrality score does not mean that a member’s votes are unpopular in the chamber, or that they are not being successful in getting legislation passed, whether it be a result of being in the minority or anything else. Rather, it indicates that the members’ votes are irrelevant – relatively few members are taking cues from those legislators. To paraphrase Ringe and Wilson (p. 746), if a Republican legislator knows that intends to vote in favor of a particular bill, they will probably be cued to vote against it. Thus, covoting network centrality is a measure of influence that is able to work within the partisan framework, but is not necessarily constrained by party influence. It assumes that legislators receive signals from influential members of all parties, allowing it to function in many different types of legislatures whether they

4 Texas Tech University, James P. Bassett, August 2020 have many parties or only two. Second, covoting network centrality is not as prone to the dilemma faced by many measures calculated by roll call voting such as DW-NOMINATE and its many offshoots – namely, that legislative voting is a stochastic process with many influences that cannot be fully understood. These votes are assumed to capture policy opinions and ideology, but it is often difficult to separate those features from the influence of party or from vote-trading and logrolling, potentially masking the true ideal points. For the most apart, these measures assume that these effects happen reciprocally and with enough frequency that they “come out in the wash”, so to speak. Covoting network centrality, on the other hand, leans into these influences – logrolling and whipping are acts of members influencing one another, and these connections are exactly what it seeks to measure. As discussed by Minozzi and Volden (2013), the idea that members do not know how to vote and look to one another (especially the leadership) is prevalent throughout the literature on Congress. Finally, covoting network centrality is a relatively portable measure – it can be effectively calculated for any legislature (or for that matter, any setting with enough votes being cast by identifiable actors) with roll call voting records. This offers many advantages for studying influence in a wide variety of settings, and exploiting those advantages will be the main strategy employed to explore exactly what covoting network centrality can tell us about legislative influence. In particular, there are two main ways that this will be used. First, it can be applied across time. Centrality scores probably cannot be directly compared across great distances of time, but at the very least the overall distribution of scores should be comparable. Setting aside for now the possibility of direct comparison, if we were to find, for example, that in Congress A there are many legislators of medium-high

5 Texas Tech University, James P. Bassett, August 2020 influence while in Congress B there are few legislators who have very high influence, this would be an insight into how the overall configuration of influence in the chamber has changed over time. The second advantage that this portability provides is the ability to calculate centrality on subsets of votes. This can offer several important points of leverage to explore influence, some of which will be discussed in the following sections. It allows centrality to be calculated for different types of votes, such as focusing only on votes on major legislation or considering whether members have more influence on procedure than they do on final passage. It also opens up the possibility of calculating centrality dynamically over the course of a single congress. The ability to track these kinds of changes across time and context makes centrality a potent tool for exploring legislative influence.

1.2 Limitations

There are, however, some considerable limitations that must be grappled with in order to fully account for centrality as a measure of influence. One of the most pressing is the nature of the agenda in Congress. Because the agenda is controlled endogenously by the members (or at least some subset of the members) themselves, their correlation on votes is determined necessarily by the matters which come up for them to vote on (Cox and McCubbins 2005, 1993). In particular, the fact that the agenda is determined by what Cox and McCubbins call the procedural cartel of the chamber’s leadership structure – exactly the members whom we would naturally expect to be among the most influential in the first place – has potential to obfuscate exactly what the centrality data is showing. I attempt to address this in part by considering how agenda control and leadership change across eras and chambers. The variation in agenda control across

6 Texas Tech University, James P. Bassett, August 2020 these settings offers a potential point of leverage to investigate how the dynamics of agenda control affect covoting centrality. As Chapter 3 will demonstrate, comparing similar legislators from across different settings – especially across temporal settings – allows for a consideration of the institutional determinants of legislative influence. Institutions matter for legislatures, and legislative influence has to be contextualized with those institutions in order to be clearly understood. It is assumed–indeed, expected–that control over the agenda has a fundamental effect on the types of issues on which legislators are able to act (Bachrach and Baratz 1994). With this in mind, models of centrality do not seek to overturn those theories which place primacy on the leadership’s utilization of agenda power. Rather, this dissertation aims to consider how members influence one another within that context, and how the changing of that institutional context affects how members of Congress go about achieving their individual goals and ambitions. In addition, the formula being used to calculate centrality is complicated greatly by its relationship to ideological polarization. Under conditions of polarization, members vote together more closely simply as a matter of definition. As a result, while members still manage to differentiate themselves to some extent, the rising tide of polarization in this case raises all boats; as Chapter 3 will explore in detail, the average score of members in a particular Congress increases substantially as polarization increases. I attempt to begin untangling this relationship in Chapter 3 but it remains the case that it must be wrestled with during every stage of this study.

1.3 Road Map

This dissertation consists of three empirical explorations of signaling influence. In Chapter 2, I directly compare centrality to a variety of similar

7 Texas Tech University, James P. Bassett, August 2020 measures of legislative skill – namely, Legislative Effectiveness Scores by Volden and Wiseman (2014), and Cosponsorship Connectedness by Fowler (2006). These metrics ostensibly measure very similar things, but I demonstrate a variety of key differences. This chapter unpacks these similarities and differences by subjecting the metrics to something of a horse race. While it would not be precisely accurate to call these measures “competitors” per se, comparing them head-to-head offers the opportunity for a much greater understanding of all of them. I find that centrality offers several key differences from the two previous measures of congressional skill – in particular, centrality offers a unique look at the effect of party leadership on members’ ability to influence others, and on the unique role of ideological extremity in determining which members drive the agenda. In addition, this chapter finds that centrality is uniquely able to unpack the distinction between ideological extremity and ideological difference from one’s party, an important delineation that is not clearly visible through the lens of other measures. In Chapter 3, I leverage the portable nature of centrality to explore how centrality responds to major upheavals in institutional structure. I do this by examining two key periods of change in the House of Representatives: the revolt against Speaker Joseph Cannon in 1910, and the reforms of the Democratic Study Group which culminated in 1977. These periods changed the nature of congressional leadership in opposite ways, and therefore combine to create a unique and useful laboratory for studying the effect of institutions on centrality. In addition, as discussed above, the character of these reform periods offers an illuminating study on the close relationship between centrality and ideological polarization. Especially in the case of the 1970s, the empowerment of the party leadership set off a runaway causal cycle of polarization, a theme which pervades this entire disseration. Overall, in this chapter I find institutional reform increases the centrality of those members

8 Texas Tech University, James P. Bassett, August 2020 who are most influential in driving for reform in the first place. While this is not a groundbreaking finding on its own – after all, the idea that political actors create institutions to benefit themselves is practically axiomatic – it does offer a unique data point in the study of such reforms. Chapter 4 explores the differences in how influence manifests itself across institutional context by comparing members’ separately calculated scores for procedural votes and votes on final passage. There has been much scholarly work on how members’ behavior differs between these two types of votes (most crucially, Ansolabehere et al. 2001; Snyder Jr and Groseclose 2000; and Jenkins et al. 2005). The broad consensus has been that party leaders keep a tight rein on members on procedural votes, and then allow them to defect if necessary on final passage in order to appease their constituents. I offer yet more evidence in support of this hypothesis by showing that those members who are most influential see most of their influence appear on procedural votes, and are nearly indistinguishable from their colleagues on passage votes. This chapter also uses the distinction between party difference and extremity first discussed in Chapter 2 to show how, generally speaking, it is the extreme members and not the moderates who hold the most influence over their colleagues.

9 Texas Tech University, James P. Bassett, August 2020

CHAPTER 2 COVOTING NETWORK CENTRALITY AS INFLUENCE: SCHOLARLY CONTEXT AND SIMILAR METRICS

At least superficially, the description of signaling influence as measured by covoting network centrality should sound somewhat familiar to congressional scholars; after all, Ringe and Wilson are hardly the first academics to take up the question of which legislators matter and which ones do not. At its most basic level, signaling influence is measuring the relationships between legislators, and inferring that these relationships are created by the activities members undertake as part of their careers in government. These relationships – especially the informal relationships that MCs form as colleagues – are at least as, if not more, important than the formal relationships that they form by participating in the institutions of Congress (Kirkland 2011). Indeed, it is this idea of “personality and skill”, as discussed by Peabody (1976), that all of these metrics are trying to measure. However, unpacking the precise meaning of centrality is key to properly understanding the analysis in the following chapters. The goal, therefore, of this chapter will be to contextualize centrality alongisde other ostensibly related measures of legislative activity, and compare how they relate to a wide range of covariates in order to get a clearer picture of what exactly centrality shows and does not show. In particular, this chapter will focus on two specific measures of legislative ability: cosponsorship connectedness, as described by Fowler (2006), and Legislative Effectiveness, as described by Volden and Wiseman (2014). By comparing centrality with these two well-vetted and oft-cited metrics, this chapter will demonstrate how centrality is different from previous work on legislative success and how it can be used to further understand legislative influence, skill, and the

10 Texas Tech University, James P. Bassett, August 2020 relationships that undergird them.

2.1 Cosponsorship Connectedness

Cosponsorship connectedness is a measure of congressional influence put forth by Fowler (2006). Like covoting network centrality, it is primarily conceived as a social network metric, measuring the relationships between legislators. Unlike centrality, however, it focuses on patterns of bill cosponsorship in order to draw those connections. Fowler uses a wide variety of network metrics to draw conclusions about MCs, but his focus is on how often members cosponsor together, and how many cosponsors appear on each bill. Using this information, he creates the connectedness measure that this chapter will use. Fowler argues that connectedness is correlated to many important legislative activities, and shows results demonstrating that connectedness ably predicts the number of amendments an MC is able to pass, as well as predicting roll call vote choice even when the usual predictors like ideology and party are accounted for. As a result, connectedness is arguably the most thematically similar to how Ringe and Wilson (2016) conceptualize centrality; they both rest upon the logic that MCs do not exist in a vacuum, and that the social relationships that form between them have an effect on legislative activities and outcomes. Like centrality, connectedness is successful at identifying MCs who would typically be expected to be influential, such as committee chairs and party leaders. However, connectedness and centrality differ in several important ways. First and most importantly, they differ in the exact legislative activities used to approximate the social networks of legislators. Unlike centrality, connectedness is focused on bill cosponshorship. There is evidence that cosponsorship measures are a valid measure of legislative work, with ideal points calculated through cosponsorship

11 Texas Tech University, James P. Bassett, August 2020 looking quite similar to those derived from roll calls, albeit with a slightly different dimensionality (Alem´anet al. 2009). In addition, these networks differ from roll call voting networks due to the voluntary nature of cosponsorship – because they are a non-mandatory activity, legislators are free to cosponsor with legislators they choose on bills that they choose. This has both benefits and drawbacks. On one hand, the freedom to cosponsor at will means that cosponsorship represents an active and positive decision on the part of the legislator; they signed onto a bill for a particular reason. This may serve as an indicator of the social nature of the so-called social network measures. On the other hand, cosponsorship is a relatively costless activity which may be done more for credit-claiming and campaigning than for legislating. Indeed, MCs do not always vote for the bills which they have cosponsored (Harward and Moffett 2010). In addition, cosponsorship data also suffers from a selection problem in that is largely self-reinforcing. As Harward and Moffett (2010) demonstrate, there is wide variation in the number of bills that senators sponsor which can be explained by many outside factors such as electoral margins and constituency demand. This introduces the possibility that connectedness may be picking up a more general propensity of particular members to cosponsor, rather than a specific idea of legislative influence. Another important distinction is on the impact of agenda setting. While measures of roll call voting have to contend with the endogenously-determined nature of the congressional agenda and its effects on what bills can be voted on in the first place, connectedness largely avoids this by virtue of its positioning before the agenda process. Because it focuses only on bills that are introduced, rather than only those which make it as far as a floor vote, it is able to capture members’ relationships without having to deal with whether the bills in question are popular,

12 Texas Tech University, James P. Bassett, August 2020 substantive, or (perhaps most importantly) agreed upon by the rest of their caucus. Connectedness also differs from centrality mechanically. While both measures are built around the premise of social networks, connectedness is calculated using Dijkstra’s algorithm1 to find the shortest distances between legislators, weighted by the number of cosponsors on any of that legislator’s given bills. As a result, while it is still bounded by 0 and 1, the distribution and mechanics of the measure are decidedly different. First, connectedness can show directionality of relationships based on which members sponsor a bill and which members join on later. This feature means that the measure can demonstrate which members influence the others in a direct way, rather than in a way that is solely theoretical as centrality does. Second, because the connections take the form of specific bills, the character of the connections themselves can be seen by looking at the substance of the bills being cosponsored. Fowler distinguishes many different types of relationships that can be formed, dividing them into institutional, regional, issue-based, and personal categories. Centrality is unable to make these kinds of post-hoc distinctions. While it can potentially find such relationships by separating out those issue areas and calculating an entirely new score, these distinctions must be made a priori, rather than being derived from the data itself. Third, across the data range connectedness shows a more right-skewed distribution than centrality, as demonstrated in Figure 2.1. This suggests that in this measure, as compared to centrality, legislative influence is probably more concentrated among the upper echelons when considering the full sample. Figure 2.2 illustrates this relationship more clearly. Members with higher levels of connectedness than of centrality tend to be relatively few, with a much greater concentration of members at lower levels of centrality. By contrast, centrality

1A more technical explanation of the measure’s exact construction appears on pages 468-469 of Fowler (2006)

13 Texas Tech University, James P. Bassett, August 2020

Figure 2.1. Distribution of Connectedness and Centrality is distributed much more evenly. In addition, those members with high levels of connectedness are almost exclusively in the earlier congresses, while members with higher levels of centrality tend to be disproportionately concentrated during later congresses. This trend for centrality will be discussed at length in later chapters as being closely related to rapidly increasing polarization after the late 1970s. The time dynamic for connectedness is therefore likely also related to increased gridlock as polarization has risen over time, but this is pure speculation, and a full inquiry into the time dynamics of connectedness is beyond the scope of this chapter. The upshot of these distinctions is that connectedness is a metric which provides an important point of comparison for explaining and understanding centrality and legislative influence more broadly. In particular, because it is so theoretically similar in terms of measuring social relationships among MCs, the mechanical and functional differences in comparing cosponsorship and roll call voting measures of influence can tell us much about the differences between the impacts of multiple types of activities, and what they mean for members’ legislative careers.

14 Texas Tech University, James P. Bassett, August 2020

Figure 2.2. Scatterplot of Connectedness and Centrality

2.2 Legislative Effectiveness

The next measure of legislative skill this chapter will focus on is Legislative Effectiveness Scores, created by Volden and Wiseman (2014). Unlike centrality, an MC’s Legislative Effectiveness Score (henceforth LES) is measured using the outcomes of the legislative process, rather than the activities which produce it. Broadly speaking, LES can be thought of as the proportion of a member’s introduced legislation which reaches various benchmarks on the path to become law: seeing action in committee, seeing action on the floor, being passed by one chamber, and, finally, becoming law. Each of these stages are weighted, as are the substance of the bills in question, with substantively significant bills being weighted more than simple substantive bills, both of which are weighted more than commemorative bills. Thus, the end result is a metric which captures how adept an individual member is at navigating their bills through the plumbing of the capitol.2 Other studies using LES have found that MCs who share staff have similar levels of effectiveness as one 2See Chapter 2 of Volden and Wiseman (2014) for a comprehensive validation of the measure in the House of Representatives.

15 Texas Tech University, James P. Bassett, August 2020 another, and generally exhibit similar ideologies (Montgomery and Nyhan 2017). In addition, Volden and Wiseman (2018) show that the hierarchical structure of the House results in leaders being considerably more effective than the rank and file, while the Senate is comparatively more egalitarian. Critically, it must be noted that this metric is not capturing the exact same thing as either centrality or connectedness. Rather than attempting to explicitly model connections between legislators, LES is instead measuring their outputs. However, the measures do rest on the same assumption, namely that a legislator’s ability to create policy that they want is indicative of some particular trait that they possess (although Ringe and Wilson may prefer to clarify that their measure does not involve personal traits so much as structural traits of the legislator’s position within the network). Indeed, as Ringe and Wilson (2016) note, in many legislative studies in the U.S. setting, the idea of legislative influence is typically conceptualized and used interchangeably with this notion of effectiveness, with some older measures using a variety of simpler productivity metrics such as “hit rate” or “box scores” (Rivers and Rose 1985; Bratton and Haynie 1999; Cox and Terry 2008). As a result, the distinctions between centrality and LES are considerably greater than between centrality and connectedness. First, the scales differ substantially; while centrality runs from 0 to 1 and follows a comparatively normal distribution, LES is specifically normalized to take an average value of 1 in any given congress and is very right skewed overall. With this in mind, direct comparison between LES and the other metrics in this chapter should only be done using standardized coefficients. Like connectedness, the right-skewed distribution of this data implies that most members in the House are relatively ineffective and the roster of highly effective legislators is quite small. Secondly, the substantive distinction between centrality and LES is critical to

16 Texas Tech University, James P. Bassett, August 2020 understanding how the measures differ from one another. As previously discussed, LES is calculated by measuring how far a particular legislator’s bills move through the process, while centrality is calculated using the correlation between a legislator’s voting record and that of their colleagues. This distinction has several implications when comparing what these different measures of influence mean. LES is primarily a measure of productivity, while centrality is a measure of signaling. Therefore, LES focuses on members’ ability to advance their own legislation, while centrality focuses on their ability to influence the outcomes of the chamber as a whole. While both measures are contingent on the same broad structural factors which affect the policy agenda, including, crucially, the agenda-setting powers of the chamber leadership, LES tends to reward legislators who are policy entrepreneurs while centrality rewards those who influence broad policy outcomes. In the context of the Cox and McCubbins (2005, 1993) cartel theory of legislative leadership, LES can be thought of as how MCs work their way onto the agenda through negotiation with the leadership, while centrality captures how legislators work within the agenda as determined by the leadership. This substantive difference is likewise reflected in the scatterplot of LES and centrality presented in Figure 2.3.3 Unlike connectedness and centrality, LES displays no clear time trend; while cases are greatly clustered near the low end (creating the right skewness discussed above), there are low LES members and high LES members observed at essentially every point in time. This is of course a direct result of the calculation: LES is normalized so that it takes a mean value of 1 in any congress for which it is created. So while LES is not nearly as helpful for examining the overall changes in congressional influence over time, it does allow for a closer reading of individual members’ trajectories over the course of their careers that does

3Because of the difference in scale, the center line uses a slope of 18.68, the maximum observed value of LES in order to approximate the 1:1 line in Figure 2.2.

17 Texas Tech University, James P. Bassett, August 2020

Figure 2.3. Scatterplot of LES and Centrality not require nearly as strict of controls for other contextual factors. Like connectedness, LES does not correlate with centrality particularly closely. This is interesting given their similarity in what they are attempting to measure, and as the proceeding sections will discuss, the result is that all three measures have very different strengths and weaknesses in the aspects of congressional influence that they capture.4 LES serves in this chapter as an important point of reference for centrality in the context of the U.S. Congress. Like connectedness, comparing and contrasting the predictors of LES and centrality can provide important insight into exactly what is being captured by both, as well as into on the varying meaning of different aspects of the legislative process. As the following sections will demonstrate, centrality can arguably be considered to be the broadest encapsulation of the full legislative process; unlike LES and connectedness, it is the most able to capture the influence that is wielded by those in positions of institutional power, such as leadership positions. In addition, it captures the ability of members of marginalized

4For the sake of completeness, a scatterplot of LES and connectedness appears in Figure A.1 in the Appendix.

18 Texas Tech University, James P. Bassett, August 2020 groups within the legislature to still be able to influence their colleagues within the agenda. Finally, it is the most capable of capturing the effects of members’ positions within the chamber in terms of the broader political context, such as their ideological position and the presence of divided government. With this in mind, the following sections will investigate each of these aspects in turn to determine the effects that each of them have on centrality, connectedness, and LES. For the purposes of data availability as well as the hierarchical structure discussed in Volden and Wiseman (2018), the analysis will focus on the House of Representatives.

2.3 Institutional Positions

The power of institutional leadership within the House of Representatives is well documented – many of the foundational theories in U.S. Congressional studies are focused on unpacking and understanding the role that they play in how the House (and it almost always is the House) functions. Most famously, the cartel theory of Cox and McCubbins (2005, 1993) argues that the leadership exhibits nearly full control over the agenda with the aim of protecting the party reputation and maintaining control of the chamber. Somewhat similarly, the theory of conditional party government popularized by Aldrich and Rohde (2001) argues that when the parties are polarized (ideologically homogeneous within themselves and distant from one another), they will delegate increasing authority to their leaders in order to save on the transaction costs of legislation, and as polarization has increased in recent years this effect has become increasingly salient (although the actual analysis of these changes will be saved for a later chapter). Both of these theories are highly focused on the power of the House’s institutionalized leadership positions. As a result, these formal institutional positions are fundamental to the everyday functioning of the House, and many studies have demonstrated their

19 Texas Tech University, James P. Bassett, August 2020 importance. Leaders and committee chairs have been shown to be greater cue givers in comparison to the rest of the party as demonstrated by the order of voting, especially on highly technical votes where specialization matters (Box-Steffensmeier et al. 2015). Along with factors such as seniority, positions of institutional power make their holders more effective legislators in the state legislatures, at least in terms of how they are perceived by their colleagues (Miquel and Snyder Jr 2006). These institutions are further strengthened by more informal institutions like caucuses (Victor and Ringe 2009). The combination of all of these formal and informal institutional structures results in a chamber that is, for the most part, highly hierarchical (MacNeil and Baker 2013). In order to test the effect of such positions on our various measures of influence, I use several indicators of MCs’ institutional strength as predictors.5 To test the effect of institutions, I use those positions in the Lawmakers dataset as covariates: seniority, a place on power committees and the budget committee, chairmanship of a committee or subcommittee, and leadership in one of the elected positions of the House party leadership. Seniority is a logical inclusion due to the simple intuition that members who have been in Congress for longer have had more time to build relationships. However it is also included in part because of existing scholarship which finds seniority as an avenue for building influence (e.g., Miquel and Snyder Jr 2006), as well as because of its historical (though now deprecated) role in determining the rank ordering of committees even beyond their chairs. Power committees are defined as membership on the Appropriations, Rules, or Ways and Means Committee, and are included in the model because of their relationship to the leadership and the agenda-shaping role that they play.6 Finally, party leaders

5All data in this chapter comes from Volden and Wiseman (2014), except connectedness which comes from Fowler (2006) and centrality, which is calculated using roll call votes from Voteview according to the formula outlined in Ringe and Wilson (2016). 6The Budget Committee is treated as its own separate variable by Volden and Wiseman, but

20 Texas Tech University, James P. Bassett, August 2020 are included due to their control over the legislative agenda, and because of their election by the party at large which at least should be expected to be a result of some degree of influence.7 Each of the variables in this section (except for seniority) are measured as a 0-1 binary. The dependent variables throughout this chapter will be our three measures of legislative influence: covoting network centrality, connectedness, and LES. While centrality can be calculated for all congresses for which roll call votes are available (which is to say, all congresses), LES data only covers the 93rd to 110th Congress, while connectedness is available for only the 93rd to 108th; as a result, this analysis covers only these two periods, with centrality and LES covering the years 1973-2008 and connectedness 1973-2004. In all models in this chapter, the method used is OLS, with clustered errors for MCs and fixed effects for each congress to separate out the effect of time trends and compare MCs only to their contemporaneous colleagues.8 The results of this analysis appear in Table 2.1. The first result is that all three metrics are in agreement about the effect of chairing a committee or subcommittee. Holding a chair increases an MC’s level of influence, regardless of whether it is subcommittee or the main committee (p < .01 for all three measures for both). In addition, none of the metrics find an effect of being on the budget committee, with statistical significance not being attained in any of the three models. More interesting, however, is the areas where they disagree: in all the cases models which consolidated Power and Budget Committee membership into one variable showed substantively similar results as the models with the committees separated, with the coefficients not being statistically distinguishable from the coefficients for the Power Committee variable. 7Coded by Volden and Wiseman from The Almanac of American Politics, these positions include the Speaker, Majority and Minority Leader, Majority and Minority , Chief Deputy Whips from both parties, Conference/Caucus Chair and Vice Chair, Republican Conference Secretary, Republican Policy Committee Chair, and Assistant Democratic Leader. 8While the right-skewed distribution of connectedness and LES does technically violate the as- sumptions of OLS regression when used as the dependent variable, both Fowler and Volden & Wiseman use the model on their metrics, and I follow their lead throughout the empirical sections of this chapter.

21 Texas Tech University, James P. Bassett, August 2020

Table 2.1. OLS Results: Institutions Dependent variable: Centrality Connectedness LES (1) (2) (3) Seniority −0.003∗∗∗ 0.002∗∗∗ 0.043∗∗∗ (0.001) (0.0003) (0.008)

Comm. Chair 0.063∗∗∗ 0.020∗∗∗ 3.297∗∗∗ (0.007) (0.005) (0.301)

Subcomm. Chair 0.043∗∗∗ 0.018∗∗∗ 1.207∗∗∗ (0.004) (0.002) (0.083)

Power Comm. 0.020∗∗∗ −0.007∗∗ −0.113∗∗ (0.005) (0.003) (0.057)

Budget Comm. 0.008 −0.0005 −0.081 (0.006) (0.003) (0.055)

Leader 0.047∗∗∗ −0.001 0.128 (0.010) (0.005) (0.113)

Constant 0.287∗∗∗ 0.231∗∗∗ 0.313∗∗∗ (0.003) (0.004) (0.062)

Observations 7,027 6,777 7,027 R2 0.462 0.429 0.422 Adjusted R2 0.460 0.427 0.420 Residual Std. Error 0.086 (df = 7005) 0.053 (df = 6755) 1.231 (df = 7005) F Statistic 285.932∗∗∗ (df = 21; 7005) 241.842∗∗∗ (df = 21; 6755) 243.393∗∗∗ (df = 21; 7005) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 of disagreement, centrality finds one result and connectedness and LES the opposite. For seniority, each additional year a member is in Congress decreases their centrality, but increases their connectedness and LES (p < 0.01 across the board). The effect of being a member of a power committee is also in the opposite direction, with centrality being positive and significant (p < 0.01) and connectedness and LES being negative and significant (p < 0.05 for both). Finally, only centrality shows an effect of being a member of the party leadership, attaining statistical significance (p < 0.01) with a positive sign, while neither connectedness and LES attain statistical significance.

22 Texas Tech University, James P. Bassett, August 2020

On the whole, these results indicate that unlike connectedness and LES, centrality seems to capture more levels of the policymaking process to one extent or another; while those indicators which directly relate to the process of creating legislation (namely the two chair variables) are effective predictors of LES and to a lesser extent of connectedness, membership on a power committee only successfully predicts centrality, as does holding an elected party leadership position. In addition, these findings greatly underline the importance of cue-giving in driving legislative influence. Following along with the findings of Moore and Thomas (1991) and Box-Steffensmeier et al. (2015), this analysis demonstrates that the specialization that comes with committee membership is an important driver of MCs’ ability to send signals to their colleagues. Finally, as in all congressional analysis, the fact that leadership positions successfully predict centrality highlights the importance of considering the endogeneity at work when studying things that involve the agenda. Because leaders control the agenda, it follows naturally that even letting bills through to the floor may be a signal in itself of which way other members should vote. This effect is obviously not present in connectedness, where many of the bills will never be seen or heard from again, and perhaps somewhat less obviously not present in LES, where the focus on members’ own bills risks taking away from the overall influence that comes from pushing the party’s entire policy agenda. Because centrality draws information from all bills voted on in the chamber and not just the leadership’s own bills, it allows leadership to show its impact more clearly by influencing the party’s policy agenda as a whole.

2.4 Marginalized Groups

Another aspect of legislative influence that scholars have investigated is membership in a marginalized group. It remains a question without an established

23 Texas Tech University, James P. Bassett, August 2020 canonical answer whether being a woman or being a member of a racial or ethnic minority group is a harm or help for members’ ability to influence their colleagues. In a vacuum, Volden and Wiseman (2014) found that women are slightly more effective than men, while African-American MCs are less effective, with Latino MCs being about as effective as their white counterparts. Bratton and Haynie (1999), on the other hand, found that women did about as well as men, while African-American legislators were less effective than white ones. Other scholars argue that the effect may be conditional. According to Volden et al. (2013), women are more effective in the minority, where their consensus-building abilities are on display, but less effective when in the majority when the party’s gatekeepers can effectively marginalize them, a finding which matches that of Barnes (2016), who showed that women collaborate with each other as a way to overcome marginalization. Likewise, Volden and Wiseman (2014) find that Democratic African-Americans tend to do worse than non-Democratic African-American MCs. This is also supported by Orey et al. (2007), who also demonstrate the impact of intersectionality. Haynie (2002) also demonstrates at the state level that African-Americans are perceived as less effective legislators, but that this effect is moderated by the presence of a black speaker (an effect which obviously cannot be tested in the U.S. House). Finally, there is some evidence to suggest that marginalized groups may be able to actually be more effective. This can come from several factors. First, because the groups are marginalized by the public, they have to be exceptional just to be elected in the first place, which should produce more effective legislators, a phenomenon known as the “ effect” (Anzia and Berry 2011). In addition, because they are marginalized by the chamber as a whole, they are inclined to create their own institutions which can facilitate greater effectiveness.

24 Texas Tech University, James P. Bassett, August 2020

These can be done by creating their own formal or semi-formal institutions such as caucuses (Pinney and Serra 1999), or by appropriating existing ones (Barnes 2014). Because of the wide variety of hypothesized conditional relationships, attempting to disentangle the effects of majority status, individual party, and intersectionality would require too many separate models; trying to tease out what is the primary effect across so many interactions is impractical and too imprecise to be worth trying to make anything resembling a definitive determination. As a result, this section will focus only on the main effects of gender and race/ethnicity, and on the intersection between them. With this in mind, I focus mainly on two groups of models. In the first, I estimate two models, one using only variables for female, African-American, and Latino, and one with interactions to disaggregate men and women in the two racial/ethnic groups.9 In the second group, I run identical models with the sample restricted to only include Democrats. Because race (and to a slightly lesser extent gender) is so highly correlated with partisanship, especially in recent congresses, this helps to further zoom in on those areas where marginalized groups make up a greater portion of the sample. Table 2.2 shows the results of OLS regression using only the main effects. Again, the results are largely the opposite between centrality and its counterparts. Female MCs are indistinguishable from their male colleagues on centrality and connectedness, and less effective than men in LES (p < 0.01). On the other hand, this analysis shows that both African-American and Latino MCs are more central than white MCs (p < 0.01 and p < 0.05 respectively), while they show lower levels of connectedness (likewise, p < 0.01 and p < 0.05 respectively) and LES for

9Because the data continues to come from Volden and Wiseman (2014), a variable for Asian- American MCs is notably absent. Despite the popular perception that they do not face marginal- ization (see e.g. Takeda 2015), ignoring them in minority politics does remain an issue (see Aoki and Nakanishi 2001 for a lengthy discussion of this topic). That said, acknowledging this omission does not help remedy it, and as it is absent from the dataset Asian-American MCs are treated as white for the purpose of this chapter.

25 Texas Tech University, James P. Bassett, August 2020

African-American MCs with no effect on LES for Latino MCs.

Table 2.2. OLS Results: Marginalized Groups

Dependent variable: Centrality Connectedness LES (1) (2) (3) Female 0.002 0.004 −0.259∗∗∗ (0.008) (0.003) (0.079)

African-American 0.071∗∗∗ −0.014∗∗∗ −0.261∗∗ (0.006) (0.005) (0.105)

Latino 0.022∗∗ −0.011∗∗ 0.044 (0.011) (0.005) (0.211)

Constant 0.287∗∗∗ 0.244∗∗∗ 1.020∗∗∗ (0.002) (0.003) (0.087)

Observations 7,920 6,778 7,920 R2 0.528 0.394 0.004 Adjusted R2 0.527 0.392 0.001 Residual Std. Error 0.089 (df = 7899) 0.055 (df = 6759) 1.611 (df = 7899) F Statistic 441.555∗∗∗ (df = 20; 7899) 244.060∗∗∗ (df = 18; 6759) 1.573∗∗ (df = 20; 7899) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

The results of the interaction models appear in Table 2.3. The most notable change is that African-American men remain more central than their white male counterparts, while African-American women are not different at a statistically significant level. Likewise, while Latinos as a whole are more central, disaggregating them by gender results in neither being statistically distinguishable from white men. For connectedness African-American MCs of both genders are slightly less influential than white men while only the African-American men are less influential according to LES (p < 0.10 for all three). White female MCs are marginally more connected and less effective than white men (p < 0.05 for both). The results of the Democrat-only models appear in Tables 2.4 and 2.5. In the main effects, female legislators are more central than their male colleagues (p < .01) when only Democratic MCs are included in the model, while their connectedness

26 Texas Tech University, James P. Bassett, August 2020

Table 2.3. OLS Results: Marginalized Group Intersectionality

Dependent variable: Centrality Connectedness LES (1) (2) (3) Female 0.002 0.008∗∗ −0.215∗∗ (0.010) (0.004) (0.093)

African-American 0.072∗∗∗ −0.011∗ −0.226∗ (0.006) (0.005) (0.129)

Latino 0.020 −0.008 0.094 (0.012) (0.005) (0.248)

Female * AA −0.003 −0.018∗ −0.178 (0.013) (0.009) (0.172)

Female * Latino 0.015 −0.019 −0.316 (0.031) (0.013) (0.296)

Constant 0.288∗∗∗ 0.244∗∗∗ 1.018∗∗∗ (0.002) (0.003) (0.087)

Observations 7,920 6,778 7,920 R2 0.528 0.395 0.004 Adjusted R2 0.527 0.393 0.001 Residual Std. Error 0.089 (df = 7897) 0.055 (df = 6757) 1.611 (df = 7897) F Statistic 401.423∗∗∗ (df = 22; 7897) 220.280∗∗∗ (df = 20; 6757) 1.521∗ (df = 22; 7897) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

and LES results remain unchanged. The rest of the results likewise are nearly identical to the results in the full model. In the intersectionality models in Table 2.5, however, the results change considerably more. White women are shown to be the main drivers of this effect, while Black women are actually less central than white men (p < .1), matching the results of connectendess. In addition, the statistical signifinance for white women on LES disappears, and connectedness now becomes statistically significant and negative for Latina women. African-American men maintain their comparatively large increase in centrality and their decreases in connectedneess and LES when compared to white men. What explains these results? One option may simply be the classic story:

27 Texas Tech University, James P. Bassett, August 2020

Table 2.4. OLS Results: Marginalized Groups, Democrats Only

Dependent variable: Centrality Connectedness LES (1) (2) (3) Female 0.033∗∗∗ 0.004 −0.201∗∗ (0.008) (0.005) (0.098)

African-American 0.071∗∗∗ −0.015∗∗∗ −0.222∗∗ (0.006) (0.005) (0.102)

Latino 0.034∗∗∗ −0.009∗ 0.117 (0.012) (0.005) (0.228)

Constant 0.302∗∗∗ 0.255∗∗∗ 1.539∗∗∗ (0.003) (0.005) (0.147)

Observations 4,432 3,816 4,432 R2 0.544 0.421 0.075 Adjusted R2 0.542 0.418 0.070 Residual Std. Error 0.083 (df = 4411) 0.057 (df = 3797) 1.680 (df = 4411) F Statistic 263.126∗∗∗ (df = 20; 4411) 153.078∗∗∗ (df = 18; 3797) 17.789∗∗∗ (df = 20; 4411) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 groups that are marginalized in the House are unable to influence their colleagues. However, these results again may highlight a critical distinction in what centrality captures when compared to connectedness and LES. As discussed in the previous section, centrality’s emphasis on floor voting and on signaling makes it less reliant on access to the formal institutions that govern so much of House procedure. In fact, it may be specifically the exclusion from those institutions that allows African-Americans specifically to be more effective signalers without necessarily being able to attract more cosponsors or successfully pass their own legislation; as members become excluded from traditional institutions, they find other ways to compensate. As Pinney and Serra (1999) demonstrate, membership in the Congressional Black Caucus leads to members being more cohesive with respect to their voting records, and while this analysis does not show consistent results for women, the underlying theory of Barnes (2016) may still be relevant. It may be,

28 Texas Tech University, James P. Bassett, August 2020 therefore, that members are able to look to African-American MCs for cues on how to vote without actually letting them into the “boys club”, so to speak, by passing or cosponsoring their own legislation.

Table 2.5. OLS Results: Marginalized Group Intersectionality, Democrats Only

Dependent variable: Centrality Connectedness LES (1) (2) (3) Female 0.036∗∗∗ 0.010∗ −0.176 (0.011) (0.006) (0.131)

African-American 0.076∗∗∗ −0.012∗∗ −0.215∗ (0.006) (0.006) (0.125)

Latino 0.029∗∗ −0.005 0.150 (0.014) (0.005) (0.267)

Female * African-American −0.023∗ −0.017∗ −0.043 (0.012) (0.010) (0.188)

Female * Latino 0.032 −0.032∗∗∗ −0.212 (0.020) (0.011) (0.301)

Constant 0.302∗∗∗ 0.254∗∗∗ 1.537∗∗∗ (0.003) (0.005) (0.146)

Observations 4,432 3,816 4,432 R2 0.545 0.422 0.075 Adjusted R2 0.543 0.419 0.070 Residual Std. Error 0.083 (df = 4409) 0.057 (df = 3795) 1.681 (df = 4409) F Statistic 240.129∗∗∗ (df = 22; 4409) 138.460∗∗∗ (df = 20; 3795) 16.185∗∗∗ (df = 22; 4409) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

2.5 Context and Setting

At the core of the theory of centrality as influence is their positioning in the social network which comprises the chamber (Ringe and Wilson 2016). As a result, the factors which determine that position should be expected to be a significant driver of legislators’ results for our various measures of legislative influence. These contextual factors, as I refer to them in this section, include the configuration of the chamber in terms of ideology and majority status, as well as features from outside the chamber itself, such as their margin of victory in the prior election and whether

29 Texas Tech University, James P. Bassett, August 2020 they operate under divided or unified government with the executive. The situation in an MC’s home state should be expected to have at least some influence on their ability to effectively govern. Because members are, at least canonically, single-minded seekers of reelection, the ability to satisfy their constituents is assumed to be a prominent component of their ability to be effective legislators (Mayhew 1974). Satisfying constituents, however, is not always easy, especially when MCs face a tradeoff between being service to their district and service to the party (Heberlig et al. 2006). There is evidence that signals from the constituency about what they want do have a marked effect on legislator behavior (Kousser et al. 2007). In addition to constituency concerns, state factors such as the party system (Matthews 1959) and level of legislative professionalism for those with state legislative experience (Volden and Wiseman 2014) have been hypothesized to affect MCs’ effectiveness. Members’ ideological considerations also play a crucial role; in particular, how ideologically extreme they are when compared to the rest of their party is exactly the kind of structural network feature that should affect how their colleagues respond to the cues they provide. There has been considerable research about how members’ ideology affects their relationship with the party, especially as the parties have moved further away from each other.10 In addition, much research has examined the question of whether the parties tend to prefer more extreme or more moderate members of their party as leaders, with mixed results; Grofman et al. (2002) argue that leaders are more extreme than their parties, while Jessee and Malhotra (2010) find that they are more extreme than the party mean, but less extreme than would normally be expected. Polser and Rhodes (1997) demonstrate

10This phenomenon of polarization after the late 1970s is a crucial part of the story of centrality, and I focus on it a great deal in later chapters. For now, Hetherington (2009) offers a comprehensive review of the literature surrounding it.

30 Texas Tech University, James P. Bassett, August 2020 an asymmetric effect on the basis of party. As a result, while the literature seems to generally agree that members’ ideological extremity compared to the party has some effect on how they are regarded by the party, it is unclear exactly what that effect is. In addition, there is evidence that divided government and majority status can affect how members send signals to their comrades. Overall, being in the majority should make members considerably more effective, at least as it has traditionally been measured. This should be fairly obvious if only on the basis of the majority leadership’s stranglehold on the agenda and the difficulty that that presents for members trying to pass a bill that the leaders do not want passed (Cox and McCubbins 2005, 1993). The evidence so far largely bears this out (Volden and Wiseman 2014; Miquel and Snyder Jr 2006). On top of this, divided government fundamentally changes the legislative process due to the changes in how bargaining works when an opposed president is involved (Groseclose and McCarty 2001; Cameron 2000). The result of this division is usually a lower level of legislative productivity (Mayhew 1991; Edwards III et al. 1997; although this finding has been questioned by others, such as Binder 1999). Crucially, however, the hypothesized effects of divided government have thus far relied on legislative productivity and gridlock. While this affects LES (specifically, gridlock presumably depresses it across the board), it must be noted that while centrality is affected by the agenda it does not require bills to be passed, only for votes to take place; while agenda control again introduces endogeneity into the picture, even those bills which do come through are able to serve as potential datapoints. As a result, it may be uniquely suited to capture the effect of divided government on members’ influence on their colleagues. This section will focus on four hypothesized contextual determinants of legislative influence: vote percentage in the prior election, ideological alignment

31 Texas Tech University, James P. Bassett, August 2020

with the party, state legislative professionalism, and an interaction between majority status and being of the same party as the president which aims to test the effect of divided government. Ideological alignment with the party will be measured in two ways: extremity (calculated as the difference between the absolute value of the MC’s first dimension DW-NOMINATE score and the absolute value of the party mean), and party difference (the absolute value of the difference between the raw values). One reason for this distinction is the presence of Southern Democrats; extremity by itself makes it difficult to distinguish the most conservative members of that faction from their very liberal counterparts and makes right-leaning moderates appear similar to left-leaning moderates, despite their distance from the party being much different. Distance from the party, on the other hand, allows the conservative democrats to correctly appear very different from the party. It also helps provide another datapoint to the “middlemen-or-extremists” question that many scholars have thus far failed to agree on–if recoding the MCs in the intra-party-mean range from positives to negatives causes the results between these models to differ, then it should help provide some insight into who is driving the findings. State legislative professionalism, on the other hand, is measured by “a weighted combination of the legislature’s salary, staff, and time in session, relative to that in Congress” (Volden and Wiseman, 2014, p. 43). This variable is coded as zero for MCs who do not have state legislative experience. While Volden and Wiseman include prior service as an interaction with professionalism, I only include professionalism due to the lack of variation in the main effects–since professionalism always takes a value of zero when experience does, it is impossible to estimate coefficients for these conditions. Simply coding them as zero allows the effect of professionalism to still be measured while providing a hard test. Finally, I include majority status, and membership in the president’s party. While being a member of

32 Texas Tech University, James P. Bassett, August 2020 the president’s party is not technically divided government, the interaction between them makes them functionally identical: if a member is in the majority and not a member of the president’s party, then by definition it must be divided government. Because of the way the variables are coded, therefore, the main effect of Majority represents the effect of being in the majority under divided government, the main effect of Same Party as President represents the effect of being in the minority under divided government, and the interaction term is the effect of being in the majority under unified government, with minority party during unified government as the reference category. The results of this analysis appear in Tables 2.6 and 2.7. First, vote percentage has no effect on centrality, while having a very substantively small effect on connectedness and LES (p < 0.01 for both). On the other hand, Ideological Extremity has a significant (p < 0.01) and substantively large effect on centrality, but no effect on the other two measures of legislative influence. Being a Southern Democrat makes members less influential across the board (p < 0.01 for centrality and connectedness, p < 0.05 for LES), while being a member of the majority under divided government makes members more central (p < 0.01) and more effective according to LES (p < 0.01). None of the other variables achieved statistical significance in these models. By contrast, in the model with party difference taking the place of extremity, the coefficient maintains its high level of significance while switching signs and becoming negative, while achieving statistical significance in the positive direction with connectedness. Vote percentage becomes significant at a very small substantive level for centrality, but the results are otherwise the same. These results indicate that members with comparatively high party difference scores but negative extremity scores (that is, they are noticeably more moderate than their party) have systematically lower levels of centrality. Figure 2.4 illustrates

33 Texas Tech University, James P. Bassett, August 2020

Table 2.6. OLS Results: Context and Setting (Extremity)

Dependent variable: Centrality Connctedness LES (1) (2) (3) Vote % 0.0001 0.0003∗∗∗ 0.010∗∗∗ (0.0001) (0.0001) (0.002)

Ideol. Extremity 0.332∗∗∗ 0.008 −0.034 (0.013) (0.007) (0.217)

Southern Dem. −0.020∗∗∗ −0.024∗∗∗ −0.282∗∗ (0.004) (0.003) (0.111)

State Leg. Prof. 0.006 0.0004 0.214 (0.007) (0.006) (0.208)

Majority 0.071∗∗∗ 0.014 1.320∗∗∗ (0.022) (0.015) (0.406)

Same Party as Pres. −0.012 −0.002 0.208 (0.022) (0.015) (0.399)

Maj. * Same Party −0.0004 −0.003 −0.444 (0.043) (0.030) (0.800)

Constant 0.250∗∗∗ 0.222∗∗∗ −0.448 (0.022) (0.016) (0.425)

Observations 6,789 6,585 6,789 R2 0.792 0.422 0.122 Adjusted R2 0.791 0.420 0.119 Residual Std. Error 0.052 (df = 6766) 0.054 (df = 6562) 1.511 (df = 6766) F Statistic 1,171.634∗∗∗ (df = 22; 6766) 217.932∗∗∗ (df = 22; 6562) 42.557∗∗∗ (df = 22; 6766) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 this: as the members get farther from the party according to this metric (that is, as their party difference score gets higher), those with negative extremity scores (indicating that they are far from the party but in the moderate, rather than the extreme, direction) are disproportionately concentrated at lower levels of centrality. This supports the idea that the party prefers more extreme MCs not just as leaders, as hypothesized by Grofman et al. (2002) and to a lesser extent by Jessee and Malhotra (2010), but as cue-givers as well. Another important finding is that neither variable has an effect on LES. This suggests that while extreme and

34 Texas Tech University, James P. Bassett, August 2020

Table 2.7. OLS Results: Context and Setting (Party Difference)

Dependent variable: Centrality Connctedness LES (1) (2) (3) Vote % 0.001∗∗∗ 0.0003∗∗∗ 0.010∗∗∗ (0.0001) (0.0001) (0.002)

Party Diff. −0.069∗∗∗ 0.022∗∗ −0.252 (0.023) (0.009) (0.273)

Southern Dem. −0.064∗∗∗ −0.026∗∗∗ −0.265∗∗ (0.006) (0.003) (0.107)

State Leg. Prof. 0.011 0.001 0.213 (0.012) (0.006) (0.208)

Majority 0.054∗∗ 0.019 1.268∗∗∗ (0.028) (0.017) (0.380)

Same Party as Pres. −0.038 0.002 0.155 (0.028) (0.017) (0.375)

Maj. * Same Party 0.048 −0.012 −0.337 (0.055) (0.033) (0.751)

Constant 0.263∗∗∗ 0.215∗∗∗ −0.371 (0.029) (0.018) (0.412)

Observations 6,789 6,585 6,789 R2 0.599 0.423 0.122 Adjusted R2 0.597 0.421 0.119 Residual Std. Error 0.073 (df = 6766) 0.054 (df = 6562) 1.510 (df = 6766) F Statistic 458.936∗∗∗ (df = 22; 6766) 218.822∗∗∗ (df = 22; 6562) 42.677∗∗∗ (df = 22; 6766) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 moderate members are equally able to get their own legislative agenda passed, it is the extremists who influence their colleagues how to vote on the agenda writ large. This finding largely comports with the earlier finding from this chapter showing a positive effect of leadership on centrality but not on LES; both leaders and extremists care about influencing the party’s policy achievements more than they necessarily care about passing the bills that they personally introduce. Finally, the lack of an effect shown by many of the variables in this model is worth noting. Given the effects demonstrated by Volden and Wiseman (2014), state

35 Texas Tech University, James P. Bassett, August 2020

Figure 2.4. Party Difference and Centrality legislative experience not being significant for any of the metrics is surprising, and may indicate that legislative effectiveness is less about learning the skill of passing laws and more about something resembling an innate ability. It is also surprising that neither minority status under divided government nor majority status under unified government make MCs more effective lawmakers or signalers, particularly because majority status under divided government does. This may be because legislating becomes increasingly a part of the bargaining process as described by Cameron (2000), and that MCs are more inclined to pass laws under divided government when they know they will run into a veto than under unified government when they stand a chance of actually becoming law. Either way, that result is beyond the scope of this chapter.

2.6 Legislative Influence Going Forward

Overall, this chapter demonstrates that centrality is a unique and useful metric for studying legislative behavior in the U.S. Congress. Across various specifications, centrality has demonstrated that it provides a window into a different aspect of the framework of legislative skill than the measures of social networks that

36 Texas Tech University, James P. Bassett, August 2020 more typically have been used in the literature. This specific phrasing is a critical part of what makes covoting network centrality a valuable tool for exploring the legislative process: not that it is necessarily a better measure, but that it is a different measure, tapping into a separate and distinct part of how legislators interact with one another. This is demonstrated by the many covariates which predict centrality in an opposite direction from connectedness and LES or which show significance when predicting centrality but not the other two. Because centrality comes from a separate stage of the legislative process – from the floor votes on the entire legislative agenda – rather than on individual pieces of legislation, it is able to capture a more complete picture of an MC’s relationship with the whole chamber. This reliance on the roll call record comes with both strengths and weaknesses. On one hand, as throughout this project, one must contend with the endogenously-determined nature of the roll call record. Control over which bills get votes in the first place is a key of majority power in the House (Lawrence et al. 2006; Cox and McCubbins 2005). As discussed by Roberts (2007), the statistical analysis of roll call votes has the potential to be misleading when considered absent of context. On the other hand, covoting network centrality has something of a unique advantage over most roll call measures: namely that it is the votes themselves that are interesting. Clinton and Lapinski (2008) argue that a lot of laws that pass do not receive roll calls, and many of the roll calls that do happen are fairly inconsequential. They draw a crucial distinction, therefore, between what they call lawmaking behavior and voting behavior. LES and connectedness are measures which aim to explain lawmaking behavior. What makes centrality interesting is that it is explicitly interested in voting behavior. The key assumption of covoting network centrality is that roll call votes imply an underlying social

37 Texas Tech University, James P. Bassett, August 2020 network, and it is this network – not the substance or outcomes of the votes – that reveals members’ signaling influence. The inconsequential votes discussed by Clinton et al. therefore are arguably not noise; in fact it is because they are inconsequential that centrality wants to measure them. When votes are relatively unimportant, it is in exactly this scenario that MCs should be expected to be less willing to invest the time to fully inform themselves and rely instead on the signals provided by their influential colleagues. Placing all of the various effects from this chapter into one large model highlights this uniqueness well. As Table 2.8 demonstrates, including all covariates preserves most of this chapter’s main findings, particularly the effects of institutional positions in committees and in the leadership and the effect of ideological extremity. While the findings for African-American drop out, this is at least in part a function of the inclusion of the Southern Democrats variable; around a quarter of the African-American MCs in this sample are counted as Southern Democrats, and models excluding that variable continue to show a coefficient for African-Americans that is significant and positive. Indeed, the Southern Democrats are a source of confoundment throughout these models and those in the previous section, in large part because they have changed so much and on so many dimensions. Leading up to and in the wake of the 1994 midterm elections, also known as the Republican Revolution, the conservative Southern Democratic faction as it was previously known more or less vanished (Black 2004, among many others). What replaced it was a faction that was much smaller (399 total observations after 1994, compared to 954 before), much more African-American (from 5.2% black to 38.5% black), much more liberal (mean 1st Dimension NOMINATE of -.165 before 1994 to -.306 after), and noticeably less different from the rest of their party (mean party difference score of .189 before 1994

38 Texas Tech University, James P. Bassett, August 2020

Table 2.8. OLS Results: All Covariates Dependent variable: Centrality Connectedness LES (1) (2) (3) Seniority −0.001∗ 0.002∗∗∗ 0.059∗∗∗ (0.0004) (0.0004) (0.009)

Comm. Chair 0.014∗∗∗ 0.017∗∗∗ 3.060∗∗∗ (0.005) (0.005) (0.288)

Subcom. Chair −0.004 0.013∗∗∗ 0.914∗∗∗ (0.003) (0.002) (0.090)

Power Comm. 0.006∗∗ −0.008∗∗∗ −0.200∗∗∗ (0.003) (0.003) (0.059)

Budget Comm. −0.002 −0.002 −0.085 (0.003) (0.003) (0.056)

Leader 0.023∗∗∗ −0.002 0.159 (0.005) (0.005) (0.120)

Female 0.008∗ 0.010∗∗∗ 0.085 (0.005) (0.003) (0.069)

African-American 0.002 −0.018∗∗∗ −0.461∗∗∗ (0.006) (0.005) (0.101)

Latino 0.012∗∗ −0.010∗∗ 0.093 (0.005) (0.005) (0.170)

Vote % 0.0001 0.0002∗∗∗ 0.003∗ (0.0001) (0.0001) (0.002)

Extremity 0.329∗∗∗ 0.016∗∗ −0.004 (0.014) (0.007) (0.172)

Southern Dem. −0.020∗∗∗ −0.022∗∗∗ −0.272∗∗∗ (0.004) (0.003) (0.066)

State Leg. Prof. 0.006 0.001 0.139 (0.006) (0.006) (0.169)

Majority 0.074∗∗∗ 0.001 0.340∗ (0.021) (0.014) (0.178)

Same Party as Pres. −0.010 −0.008 −0.113 (0.021) (0.014) (0.177)

Maj. * Same Party −0.004 0.009 0.222 (0.043) (0.028) (0.353)

Constant 0.250∗∗∗ 0.224∗∗∗ 0.074 (0.022) (0.015) (0.208)

Observations 6,789 6,585 6,789 R2 0.795 0.454 0.440 Adjusted R2 0.794 0.452 0.438 Residual Std. Error 0.052 (df = 6757) 0.052 (df = 6553) 1.207 (df = 6757) F Statistic 846.945∗∗∗ (df = 31; 6757) 175.920∗∗∗ (df = 31; 6553) 171.344∗∗∗ (df = 31; 6757) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

39 Texas Tech University, James P. Bassett, August 2020 and .154 after). 11 Beyond these changes in the coalition itself, time also saw Southern Democrats facing far different contexts. Democrats were in the majority for every Congress in the sample prior to 1994, while only being in one afterward (or none, in the case of connectedness). As the next chapter will discuss at length, polarization, which had already been on the rise prior to 1994, began a period of truly runaway increase afterward. Finally, conditions within the Democratic party were themselves a driver of all the aforementioned changes. As I discuss in Chapter 3, the reforms of the late 1970s were explicitly intended to drive the Democrats’ legislative agenda toward the priorities of the Northern wing of the party and away from the Southern wing. The party itself, therefore, changed in such a way as to alienate Southern Democrats throughout the 1980s. The upshot of this effect is that my findings for the Southern Democrats are remarkably consistent across model specification: Southern Democrats were less central (and usually less connected) than their non-Southern Democratic colleagues during the 1970s and 1980s when they were on the outs from the party writ large, and no more central after most of the conservatives were defeated or switched parties. Once the Southern wing of the party was exorcised in 1994 (or perhaps amputated, given the electoral ramifications), the party became much more uniform, and those who remained in the South came to resemble not a faction so much as a run-of-the-mill Democratic MC, a finding which is supported by a wide variety of specifications of the previously reported models. It, of course, is not perfect; all of the measures discussed in this chapter suffer from the problem of an endogenous network; as Rogowski and Sinclair (2012) argue, such networks have the ability to greatly hinder the ability of social scientists to draw causal relationships. In addition, notable in every model in this chapter is the

11Both NOMINATE-derived measures were statistically significant in a t-test with a p-value of less than .01.

40 Texas Tech University, James P. Bassett, August 2020 very small substantive significance; the only results that showed a substantive result on the order of a full standard deviation was the effect of committee chairmanship on LES and of extremity on centrality. While I discuss the overarching theme of substantive significance more in the conclusion, for now this probably indicates that all of these measures are first and foremost, as all of the measures’ theories originally argued, indicators of some individual-level factors that cannot be fully accounted for by anything other than the personality and skill of the MC in question. To the extent that they are systematically determined by institutional or contextual factors, the evidence seems to point to a holistic process in which no one factor can be definitively identified as the one true driver of members’ ability.

41 Texas Tech University, James P. Bassett, August 2020

CHAPTER 3 CENTRALITY OVER TIME: INSTITUTIONAL CHANGE AND INFLUENCE IN CONGRESS

It is not radical to say that Congress has changed substantially during its storied history. The study of the development of Congress is as popular as it is precisely because it has undergone so much change, and done so in ways that have fundamentally reshaped the character of its deliberative process. Over time it has changed from a body dominated by outsized personalities like and John Calhoun, to one controlled with an iron fist by a -like Speaker, to one led by the various committees, before finally becoming a body that looks much like it does today, with a bounded and powerful party leadership structure. Throughout its history, changes in the rules, composition, membership, and institutionalization of the House of Representatives have created considerable variation in the ways that members navigate the chamber. With these changes, the factors that affect members’ ability to create the legislation they value and to advance their own ambitions have changed as well. This being a dissertation on the nature of congressional influence, these changes in the institutional structure of the House must occupy an important role in understanding how that influence has meant different things to different members of different eras. The goal of this chapter is to trace how institutional reforms have affected influence, both in terms of the covoting network centrality of individual MCs and in the distribution of centrality for the whole chamber. In particular, I focus primarily on two key periods of institutional upheaval: the 1910 rebellion against Speaker Joseph Cannon which significantly limited the Speaker’s ability to dominate the chamber, and the reforms of the Democratic Study Group (DSG) which culminated

42 Texas Tech University, James P. Bassett, August 2020 in 1977 and aimed to curtail the ability of senior committee chairs – largely conservative Southern Democrats – to obstruct the liberal policy goals of the DSG. Both reform periods fundamentally changed the character of policymaking and leadership in the House. The 1910 revolt created a much more decentralized party structure, with the power largely shifting to committees. The 1977 reforms had the opposite effect; by empowering subcommittee chairs, allowing bills to be referred to multiple committees, and giving the Speaker the authority to appoint the members of the Rules Committee, power (especially over the agenda) became much more concentrated in the hands of the Democratic Party leadership. In addition, as the following sections will discuss in more detail, they each mark the beginning of what can be seen as clearly defined eras of congressional influence. This chapter has two overarching themes. First, this chapter demonstrates an age-old axiom about the relationship between political actors and institutions: actors create institutions to benefit themselves. This can be seen in the creation of political parties (Aldrich 1995; Cox 1987), electoral institutions (Boix 1999), and even authoritarian institutions (Svolik 2012). Indeed, this chapter finds that for both periods of major change in the House, the most direct beneficiaries are exactly those actors who drove the changes in the first place: the insurgent Republicans who rebelled in 1910, and the Democratic leadership in 1977. While saying that actors move to benefit themselves is not in itself a groundbreaking finding, it does provide additional perspective on the drivers of major institutional change, and show that members are sophisticated enough to understand how those institutional changes will affect their own ambitions. Second, this chapter examines the relationship between the short-term decisions that actors make and the path-dependent long-term effects of those decisions. While the two reforms had specific goals in mind, they both ushered in

43 Texas Tech University, James P. Bassett, August 2020 periods where the overall configuration of covoting network centrality changed considerably. Prior to 1910, the mean level of centrality was extremely inconsistent from year to year, something that might be expected if a member’s ability to attract votes was dictated by the agenda of the Speaker and the Speaker alone. After that time, however, the mean level of centrality settled into a relatively stable range throughout the middle of the 20th Century, again something that would be expected when the chamber is largely controlled by seniority-dominated committees. This pattern continued until the reform period of the 1970s, after which the mean level of centrality trended upward, a finding consistent with increasing polarization as the party leadership consolidates control over the agenda. Both cases demonstrate that while a certain subset of MCs pushed for these reforms with the immediate aim of benefiting themselves via an expectation that they would gain the most from the proposed reforms, the effects of institutional changes that they created persisted long after their time in the House. This chapter will therefore begin with an overview of the literature surrounding these two key inflection points, and on the hypothesized effects of the institutional changes at each stage. While the literature does suggest that certain institutional positions (namely committee chairs in 1910 and subcommittee chairs in 1977) should become more influential after these changes, the main hypotheses of this chapter will focus much more on legislators as self-interested actors who seek to benefit themselves before any institutional position. Next, the empirical section of this chapter will test these hypotheses using a difference-in-difference approach, estimating models for a rough period surrounding the two inflection points: the 57th-67th Congress (1901-1922) for the Cannon revolt and the 93rd-99th Congress (1973-1986) for the DSG reforms. These models demonstrate that in both cases, the institutional changes most benefited the MCs who advocated for them. Finally, the

44 Texas Tech University, James P. Bassett, August 2020 chapter will conclude with a broader discussion of centrality throughout the 20th Century, including between and after the two primary periods of interest. In particular, this section will consider the interplay between agenda control, polarization, and centrality. It has been demonstrated by Theriault (2008, 2006) that the usage of procedural agenda power – control over which was one of the most important stakes in both reforms – is a key component in the stark increase in ideological polarization in the last decades of the 1900s and the first decades of the 2000s. Because centrality is calculated using the correlation between MCs’ voting records, it follows that high levels of polarization should cause wide swaths of members to vote together much more frequently and the average level of centrality to rise as a result. While the discussion in Chapter 2 on the link between ideological extremity and centrality sheds some light on this connection, this section will continue to attempt to untangle the relationship between the agenda, polarization, and centrality.

3.1 Revolt and Reform: Key Moments in the Development of Congres- sional Influence

There are several key inflection points where the nature of congressional deliberation changes substantially. Schickler (2001), for instance, cites four: 1890-1910, 1919-1932, 1937-1952, and 1970-1989. Of these, the literature largely focuses on the first and the last, and as a result they make the most logical choices as the focal points of this chapter. For each of these two moments, my primary interest is on the moments that are most commonly cited as the tipping point. This moment is quite clear in the case of the 1910 reforms; the revolt against Speaker Joseph Cannon is a pivotal event in the development of the U.S. House. For the latter period, however, I focus on 1977 as the critical moment. Rohde (1991) and

45 Texas Tech University, James P. Bassett, August 2020 others typically consider this to be when the preceding reforms reached their critical mass and finally allowed the DSG to start making real inroads into their agenda, and as a result the rest of this chapter will use it as the turning point and namesake for the era as a whole. With that said, the first things will remain first, and this section will start by discussing the literature surrounding the events of 1910.

3.1.1 The Revolt of 1910

The years between 1889 and 1910 in the House of Representatives were characterized by the dominance of the Speaker. The speakerships of (R-ME) and later (R-IL) were described in their own time as czar-like, with the rest of the chamber being more or less at the mercy of its leader. The Speaker’s power during this time came from several sources. First, they had nearly complete control of the Rules Committee. Aside from controlling its membership, the Speakers during this period also sat on the committee themselves (Follett 1896). In addition, the Speaker also had considerable power over the creation and dissolution of new committees, as well as essentially unilateral authority over their membership. The Speaker was also allowed membership on standing committees, and was therefore able to exert his authority on nearly all areas of policymaking. I will not attempt to craft a narrative of the 1910 revolt more dramatic than that told by Rubin (2013). There are, however, several key takeaways from that narrative of the lead up to the revolt which explain the relevance of this particular episode to the chapter overall. The most important for the purpose of this chapter are the behind-the-scenes organization of the Republican insurgency, and their lack of ideological unity. As Rubin (2013) explains in detail, there was considerable unrest amongst the rank-and-file leading up to their eventual revolt, particularly

46 Texas Tech University, James P. Bassett, August 2020 with regard for the shifts in policy that were demanded by the growing agrarian crisis in the Western regions of the country as railroad and corporate interests drove farmers into mounting debts. This dissatisfaction was crucial for creating the conditions that would eventually lead to the rebellion. The insurgent members, led by George Norris of , organized extensively outside of the formal avenues of Congress. This included the usage of their network of progressive newspapers to reward those who stuck with them. Notably for the sake of this dissertation, this created exactly the kind of informal social networks that covoting network centrality is intended to capture. In addition to their organization, the insurgent Republicans in 1910 were characterized by significant disunity on actual policy issues, particularly on tariffs (Sarasohn 1979). The tariff issue was in fact divisive enough within the Republican party so as to warrant Cannon calling the chamber back into session early and conducting votes on it even before distributing committee assignments in an attempt to keep the coalition together (Lawrence et al. 2001). While they were in agreement on the core issue of anti-Cannonism, both before and after the revolt in March of 1910 there was little else on which they agreed (Baker 1973). Although the agrarian debt crisis may have been the impetus for Republican dissatisfaction with the Speaker’s level of control, the insurgents did not organize around the issue itself, but rather against Cannon’s hard and fast refusal to make any move to alter the status quo (Schickler 2001). This division not just within the party but within the faction makes this a particularly interesting case study because it helps to disentangle influence from ideological agreement. This entanglement has been a recurring theme throughout this dissertation, especially in the period after the 20th Century when polarization and centrality become so closely linked. In addition, it makes it much more of a hard case because the post-reform period gives the insurgents little to

47 Texas Tech University, James P. Bassett, August 2020 coalesce around. Their influence, therefore, has to come from somewhere else. The House voted on the reform resolution on March 17, 1910, when Norris took advantage of a prior precedential ruling by Cannon that constitutional matters should be privileged and brought to the floor. After a lengthy bargaining process, both with the Speaker and his loyalists and with the Democratic minority, the resolution was eventually passed. When the smoke cleared, the Speaker had lost his ability to be a member of standing committees, including and especially the Rules Committee, which had also been expanded and diversified. In addition, Speakers were no longer to be the sole arbiter of committee appointments, with the aptly named Committee on Committees taking over that role. While these reforms on their own were enough to effectively neuter the Speaker’s authoritarian control over the chamber, Baker (1973) notes that they could have gone even further if not for their reliance on Democratic votes; because the Democrats intended to press anti-Cannonism as an issue in the coming midterms and anticipated controlling the Speaker’s chair afterward either way, they were reluctant to remove too much power from the Speaker. In the aftermath of this vote, Cannon asked to resign the Speakership and called for an election for a new speaker. The most important outcome of the revolt against Cannon was the reversion of power back to what were now essentially independent committees, with chairs in those committees determined almost solely by seniority (Polsby et al. 1969). Prior to this, committees were more or less made up on the fly by the Speakers, with the number of committees during this period swelling to nearly 70 by the time of the revolt (Polsby 1968). After the revolt, there was considerable consolidation of the committee structure, with the total number of standing committees falling to only 29 by 1921 (Deering 2003). The combination of seniority-based chairmanship and the right to stay on a committee once assigned severely hamstrung the leadership’s

48 Texas Tech University, James P. Bassett, August 2020 ability to exercise much control over the business that they conducted (Sinclair 2003). Indeed, once the dominance of committee government became institutionalized under the Democrats in the years following the revolt, they came to resemble something of a miniature feudal fiefdom (Peters 1990). The dominance of the House by mostly unremovable committee chairs lasted well into the 1970s, until its intransigence under the conservative coalition of Southern Democrats and Republicans forced action by yet another intra-party faction.

3.1.2 The Reforms of the 1970s

In many ways, the reforms of the 1970s were a direct consequence of the prior reforms of 1910. From the time the new Democratic majority was sworn in the following January until the late 1970s, committee government was the foundation of congressional activity. By and large, this made the chairs of those committees the primary arbiters of the agenda. Because those chairs were determined by seniority and members maintained a right to remain on committees once they had been appointed there, many of these committees, especially in the 1950s and ’60s, quickly ossified under the leadership of unassailably safe incumbents. Unfortunately for the Democratic party leadership, which by this time was increasingly made up of members from the liberal Northern wing of the party, those safely incumbent districts were largely represented by conservative Southern Democrats. Throughout much of the latter half of the 20th Century, these Southern Democrats – the Boll Weevils, as they would later come to be informally known – formed a conservative coalition with the minority Republicans to stonewall liberal legislation. Like in 1910, the response would eventually come from a formally organized faction within the same party. The Democratic Study Group (DSG) was initially formed in 1959 by a group of liberal representatives with the aim of researching and

49 Texas Tech University, James P. Bassett, August 2020 providing leadership and material to help facilitate the passage of their legislation. While they never kept official records of who their members were, Baer (2017) estimates that by the 1970s around half of the Democratic caucus was routinely receiving materials from the DSG. Frustrated by the contumacy of the committee chairs, by the late 1960s, the DSG began slowly introducing reforms with the aim of reempowering the leadership structure that had been defanged since the 1910s while maintaining enough control within the caucus to prevent the leadership from becoming as authoritarian as it had been under Cannon and Reed (Rohde 1991). Unlike in 1910, however, this process was extremely gradual, beginning in 1970 and culminating nearly a decade later in 1977. Because these reforms were implemented piecemeal and over time, this allows for additional information about the effects of different reforms. Rohde (1991) cites 1977 as the delineating moment where the post-reform period can be said to begin, and as a result I do as well. In addition, these reforms were highly ideological. The DSG was an explicitly liberal organization, and their primary frustration with the Boll Weevils was their inability to pass such legislation. While both reform periods were certainly based in a Mayhewian electoral logic, in this case the members’ electoral incentives were directly based on the Cox and McCubbins (1993) theory of building a party brand through the passage of good legislation, rather than the perception of indecision in a time of crisis that faced the progressive Republicans in 1910. By and large, these reforms came in two major avenues: reducing the power of committee chairs, and strengthening the party leadership. The first of these was done through a variety of avenues, the most important of which was the empowerment of subcommittees. On the surface this seems contradictory, since adding additional veto points and more cats to herd (so to speak) should make the pursuit of an agenda writ large even more unwieldy for leadership to manage.

50 Texas Tech University, James P. Bassett, August 2020

However, as Rohde (1991) points out, the DSG pursued this change for exactly that reason. The Subcommittee Bill of Rights in 1974 guaranteed that most legislation, once referred to committee, would first make its way to a subcommittee with appropriate jurisdiction. The effect of this was that chairs were much less able to restrict the flow of the legislation that entered their purview. The DSG also supported changes to the chair selection process, allowing chairs under certain circumstances to be determined by a vote of the caucus, with later reforms resulting in voting for committee chairs becoming the norm. Ultimately, this forced the committee structure to ultimately be accountable to the rank-and-file, and therefore to the majority of the party. With this accountability in place, chairs were greatly reduced in their ability to wield their committee power in an arbitrary and capricious manner. Likewise, strengthening the party leadership was done through several tracks, the most important of which being multiple referral, suspension of the rules, and special rules out of the Rules Committee. In 1974, multiple referral of bills to committee was introduced (Davidson and Oleszek 1977). Prior to this, bills were subject to the strict jurisdiction of a single committee. This power allowed Speakers to get around this restriction by sending bills to multiple committees, either simultaneously or in sequence, with later changes allowing the Speaker to place a time limit on the first committee in a sequential referral (Davidson et al. 1988). This led to the party leadership having considerably more influence on the scheduling of legislation (Rohde 1991). The next major reform was to the usage of the suspension of the rules. In both 1973 and 1977, the DSG supported increasing the amount of time available to consider legislation under a suspension of the rules. Because the time had been limited, it had been used up to that point primarily on commemorative bills or other matters with high levels of consensus. Expanding the

51 Texas Tech University, James P. Bassett, August 2020 amount of time allowed the leadership to consider much more contentious legislation under a suspension of the rules, a move which led to more bills failing under suspension, but changed the bargaining process significantly, usually in the leadership’s favor. The most important avenue for reform, according to (Rohde 1991), took place in the Rules Committee. Prior to this time, the Rules Committee had come to be dominated by Southern Democrats and Republicans, who used the committee to prevent most major legislation from reaching the floor. This impasse had already prompted the House leadership to expand the committee once before in 1961, when Speaker added two additional members to attempt to break the conservative coalition’s hold over it. In 1975, the Speaker was given the power to appoint the members of the Rules Committee with the approval of the caucus. This transformed the committee from one which was mostly independent under the rule of the chairman to one which was effectively an arm of the party leadership. This transformation allowed the committee to serve in a variety of roles (Oppenheimer 1977): as traffic cop, the committee began referring major legislation to the floor, rather than delaying or limiting it. As dress rehearsal, members could get a sense of how their legislation would play before bringing it to the floor. And as field commander, they could represent the leadership’s interests in committee, providing the leadership with “eyes and ears” in the committee (Rohde 1991). In particular, the traffic cop role allowed the committee incredible leverage over the terms of debate and amendment on legislation, and closed or complex rules became increasingly common, especially on substantive legislation and legislation from power committees (Oppenheimer 1981; Bach and Smith 1988). In the short term, the effect of these reforms was a greatly weakened committee system and a greatly strengthened party leadership. The period after

52 Texas Tech University, James P. Bassett, August 2020 these reforms saw massive expansions across the board in the Democratic leadership’s ability to control the agenda under the speakerships of Tip O’Neill (D-MA) and (D-TX) (Rohde 1991). Once similar caucus-based reforms were passed by Republicans, such a pattern of leadership dominance would continue into the later years under (R-GA) (Theriault and Rohde 2011) and (R-IL) (Fechner 2014; Richman 2015). Despite the presence of a Republican president for the majority of the period under study in this chapter, the Democrats’ ability to control the agenda was much enhanced, allowing them to push their legislative priorities much more effectively even over the objections of the Republicans in the minority and in the (Rohde 1991, p.109-112). In the long term, the strengthened leadership of the postreform House would lead to a huge increase in ideological polarization over the course of the following decades, a trend which has come to define much of Congressional scholarship in the 21st Century (Hetherington 2009). Theriault (2008) directly attributes this to the increased usage of procedural power to bend the legislative agenda to the will of the party leadership. In conjunction with the theory of Conditional Party Government (most frequently associated with Aldrich and Rohde 2001), which argues that leadership is delegated more responsibility when the parties are internally homogeneous, these reforms can be seen as the inciting moment for a cycle of mutually self-reinforcing growth in both leadership power and ideological polarization, with each one increasing the other in turn. It is unclear how far that trend can continue, but as of this writing it has shown no signs of abatement.

3.1.3 Similarities and Differences

These moments have several important differences: the 1910 revolt was sudden and resulted in nearly overnight change (although the full effects would take

53 Texas Tech University, James P. Bassett, August 2020 years to completely settle into place), while the 1977 reforms took place slowly over the course of several years. Indeed, while Rohde (1991) tends to cite 1977 as the tipping point, the first moves in this effort took place as far back as 1961 and began in earnest in 1970. In addition, the two reforms moved in opposite directions in terms of their relation to the Speaker and the leadership overall. The 1910 reforms were intended as a way of decentralizing power out of the hands of the Speaker, while the 1977 period was re-centralizing power back toward the party leadership. The factions who drove these changes also had very different motivations, with the insurgents in 1910 being united on very little except their opposition to Cannon, and the Democratic Study Group of the 1970s being a movement mostly centered on a desire for liberal policy goals. Finally, while the Democrats in the 1970s were extremely confident in their continuing majority status in the coming years (indeed, this would prove true well into the 1990s), the Republicans in 1910 made their move under the looming shadow of being in the minority after the upcoming midterm. As a result, while the DSG managed to accomplish everything they wanted, the insurgents had to compromise on many aspects of their reform due to having to deal with the minority party to make their institutional changes happen in the first place. However, they also notably had several key similarities. First, they both were driven by a relatively small group of legislators. Despite being joined by the entire Democratic delegation in the actual vote to strip the Speaker of power, the 1910 insurgent Republicans numbered only about 44 (Rubin 2013). The Democratic Study Group, on the other hand, generally maintained a membership comprising about half of the Democratic caucus throughout the 1960s and ’70s, and their internal records indicate that many of those members were on a comparatively informal basis of receiving research information, with a fairly small leadership structure undertaking much of the actual work (Baer 2017). Second, they were

54 Texas Tech University, James P. Bassett, August 2020

driven by explicitly factional and intra-partisan conflict. As previously discussed, the insurgents had very little in common with one another in terms of policy or ideology, but what they did have in common was opposition to the leadership of their own party – not just the Speaker per se, although he was naturally the focal point of their dissatisfaction, but with the Speaker’s allies who allowed him to maintain his iron rule. Likewise, the DSG was frustrated by the conservative southerners in their own party at least as much as they were with Republicans; their primary motivation for founding the group was resentment of the obstinacy of the senior-appointed chairs in much of the committee system in the wake of the influx of liberal Northern Democrats in the 1958 election (Rohde 1991). For each of these similarities, the existence of formal organization both inside and outside of Congress proved critical to the reformers’ ability to achieve their goals. This kind of organization and the social networks that it creates are crucial to the understanding and usage of centrality as this dissertation employs it. Finally, while both sets of reforms eventually did empower certain institutions – committee chairs in 1910 and subcommittee chairs in 1977 being the most prominent examples – in both cases the reforms were enacted primarily with the short term goal of empowering the reformers themselves. This is especially clear in 1910 due to the coalition’s ideological disunity. They had no particular interest in trying to empower any other institution in part because their only common goal was opposition to Cannon, and they were reluctant to grant too much power to any other existing institution for fear that the incoming Democratic majority would use it against them. The reversion of power to the committees, then, was more by default than by any intentional choice on the part of the insurgents. The underlying motivation of the 1977 reforms is similarly self-interested. The reforms were enacted in order to promote the policy agenda of the Democratic party leadership, and to the

55 Texas Tech University, James P. Bassett, August 2020 extent that the Rules Committee and subcommittee chairs were empowered in these reforms, they were empowered in order to be used as a tool by the party leadership. Taken together, these similarities and differences can be used to formulate the key hypotheses for the empirical section of this chapter. Because of the differences in the two eras’ reformers and goals, I propose two alternative hypotheses for each:

Hypothesis 1A: After 1910, committee chairs and power committee chairs should become more central than other MCs.

Hypothesis 1B: After 1910, members of the Republican insurgency should become more central than other MCs.

Hypothesis 2A: After 1977, subcommittee chairs should become more central than other MCs.

Hypothesis 2B: After 1977, members of the Democratic Party leadership should become more central than other MCs.

In each case, the first alternative suggests that the primary beneficiaries were those who were most directly impacted and empowered by those reforms. As I have argued, however, the main beneficiaries of these changes are more likely to be those members who were most instrumental in pushing for their respective reforms: the insurgent Republicans in 1910 and the Democratic party leadership (by way of the DSG) in 1977. In both cases, I expect that the reforming group should see their level of centrality rise in comparison to the rest of the chamber.

3.2 Research Design

In order to test these hypotheses, I estimate models using two separate samples. The first uses roll call votes by members in the 57th through 67th Congresses, while the second uses the 93rd through 99th. Covering the years from 1901 to 1923 and 1973 to 1986 respectively, these samples provide a window before and after the chapter’s key inflection points. These models utilize a

56 Texas Tech University, James P. Bassett, August 2020 difference-in-difference design, with the key cutpoints of the 61st and 95th Congresses serving as the “treatment” in each case.1 These samples were chosen, therefore, in an attempt to capture both the lead up to and aftermath of the reforms that this chapter examines. In the case of the 1910 reform, I utilize data from roughly five congresses both before and after, although there is technically a sixth treated congress at the end for the purpose of capturing an additional Republican-majority datapoint.2 For the 1977 reform, I start with the 93rd Congress due to data availability.3 Because a sample with an equal number of congresses on either side of the treatment would be relatively small, I extend the post-treatment sample to the 99th Congress in order to capture the end of Tip O’Neill’s speakership, which serves as a convenient natural ending point.4 The dependent variable, as usual in this dissertation, is covoting network centrality. I will not belabor the explanation too much here since it has already been discussed so thoroughly in the preceding chapters, but I do note once again the relationship that this variable shares with the agenda. Because so much of the variation found in this dataset is based on the changing ability of leadership to shape the agenda, there is naturally a great deal of concern with the entanglement

1In the very strictest sense I admit that this is not technically a difference-in-difference, since the untreated population is also subject to the treatment; that is, while I argue that, for example, insurgents were the “treated” group, the rest of the chamber is also subject to the institutional changes and are therefore also part of the treatment, albeit in a different way. While the design is therefore more akin to a standard interaction model, I continue to use the terminology of difference- in-difference for ease of interpretation across model specifications. 2The data in this sample is sourced from Lewis et al. (2019) for roll calls and NOMINATE scores, from Canon et al. (1998) for committee membership information, and from C-SPAN (2019) for committee chairs. 3The data for this sample comes from Volden and Wiseman (2014), except for roll calls which are sourced from Lewis et al. (2019). Because Volden and Wiseman get their data from the Library of Congress website, which only provides bill information starting in 1973, it serves as their starting point. While I do not use the bill data that censors their data process, the scale of the data collection task that would be required to expand their dataset further back in time forces me to accept the limit that their dataset creates. 4I also ran models which ran all the way through the 108th Congress, the last congress in the Volden and Wiseman (2014) dataset, and found substantively identical results as shown in Table A5 in the Appendix.

57 Texas Tech University, James P. Bassett, August 2020

Table 3.1. Contingency between Committee Chairs and Insurgents, 57th-67th Congress

Committee Chair No Yes No 3817 474 Insurgent Yes 209 32 between the treatments and the data-generating process. The core conceit of both treatments is that the party leadership gains or loses some degree of agenda power afterward, which means that a large portion of changes in centrality scores should be determined at least in part by the types of votes that come to the floor, rather than purely the influence that members have over those votes. To some extent, this does present an issue of endogeneity; of course the key groups will have more influence on roll calls if the treatment means they control which measures get to be voted on. However, the usage of the 1910 sample helps to account for this because it is a case which reduces the agenda power of the leadership, shifting instead largely to the committees. As Table 3.1 shows, there were very few insurgents serving as committee chairs, meaning that the group which I hypothesize should benefit most from this reform and the group which receives the greatest increase in agenda power have very little overlap. This helps to disentangle the two processes and demonstrates the functioning of the mechanism in an environment where the data generating process and the agenda should be much less endogenous. As a “difference-in-difference” design, these models require two main independent variables in each sample: the treatment and the treated group. In both samples, the treatment is the key moment discussed in the previous chapter. The treated groups, however, differ both between samples and between hypotheses. For the 1910 case, I use models with committee chairs and insurgents as the treated

58 Texas Tech University, James P. Bassett, August 2020

groups to test Hypotheses 1A and 1B respectively. For the 1977 reforms, I use subcommittee chairs and majority party leadership (in this sample always the

Democratic party leadership) as the treated groups for Hypotheses 2A and 2B. In the former case, the line between theory and measurement is fairly straightforward; they are exactly the groups that are hypothesized. In the latter case however, the connection is slightly murkier. While the party leadership does not translate directly to the DSG, there is evidence that there was significant overlap; Rohde (1991) specifically mentions Richard Bolling, one of the DSG’s key leaders, being one of Tip O’Neill’s early picks for the House Rules Committee. Because of the DSG’s lack of membership records, there is no good way to treat them as a distinct group, but the narrative of the reforms in this period as discussed in Rohde (1991) and Baer (2017) suggests that they were deeply ingrained within the leadership of the Democratic Party writ large. As a result, while I cannot directly measure changes in centrality for the members of the DSG itself, the party leadership for the Democratic Party should serve as a useful proxy. Finally, I run a series of controls. These controls include majority status, party, power and budget committee membership, ideological extremity, and the time trend (measured on a Congress-by-Congress basis, i.e, with fixed effects).5 I also include a separate variable which measures the time since treatment. This variable takes a value of 0 in pre-treatment congresses, then increases by 1 in each congress after treatment. While this measure is obviously highly collinear with the base time trend, I include it anyway to measure how members’ influence changes as they become more accustomed to the new institutional environment they find themselves in.6 There were also some controls which appear in some models but not others. In

5See the previous chapter for a more in-depth discussion of these variables. 6Models which exclude the time since treatment variable were substantively identical and appear in Tables A6 and A7 in the Appendix.

59 Texas Tech University, James P. Bassett, August 2020

the 1910 case, I include a variable for being a chair specifically of power committees7, and exclude the variable for the Budget Committee, since the budget at this time was broken up into a series of issue-specific Expenditures Committees. Likewise, in 1977 I include a variable for Southern Democrats and exclude the variable for party because the Democrats were in the majority throughout the entire sampled period. As always, I cluster errors on MC.

3.3 Results

The results of the difference-in-difference regressions appear in Tables 3.2 and 4.4.8. As Table 3.2 shows, before 1910 insurgents tended to be less central than Democrats or non-insurgent Republicans (p < 0.01). After the revolt against Cannon, however, committee chairs saw no statistically significant change in their centrality level (p = 0.471) while insurgents tended to be more central than non-insurgents (p < 0.01). These results show no support for Hypothesis 1A and

strong support Hypothesis 1B. This finding is reflected in Figure 3.1.9 While both groups’ centrality was trending downwards prior to the revolt against Cannon, the aftermath of that revolt saw non-insurgents centrality drop much more dramatically, while insurgents’ narrowed the gap substantially. Although these findings do not indicate a complete reversal in that insurgents demonstrated lower levels of centrality than their counterparts overall, they do show that the Republican insurgency benefited

7Appropriations, Rules, and Ways and Means 8For the sake of brevity, I only include the covariates of interest in these tables. The full results including all covariates appear in Tables A3 and A4 in the Appendix. By and large, almost all of the covariates in both cases showed very similar results to those found in Chapter 2. 9To clarify, these plots show predicted centrality based on OLS models, with predictions based on time variables (Congress fixed effects, time since treatment, and changes in party control) and the difference-in-difference cutpoint variable. So while the slopes change in response to those variables, the fact that both groups’ slopes are parallel before and after cutpoints should not be interpreted as anything except an artifact of OLS modeling.

60 Texas Tech University, James P. Bassett, August 2020

Table 3.2. OLS Results: 1910 Case

Dependent variable: Centrality (1) (2) Post-1910 −0.060∗∗∗ −0.066∗∗∗ (0.007) (0.006)

Insurgent −0.075∗∗∗ −0.110∗∗∗ (0.009) (0.013)

Comm. Chair 0.007 0.012∗∗ (0.008) (0.005)

Post-1910 * Comm. Chair 0.008 (0.010)

Post-1910 * Insurgent 0.080∗∗∗ (0.013)

Constant 2.583∗∗∗ 2.548∗∗∗ (0.097) (0.096)

Observations 4,509 4,509 R2 0.437 0.441 Adjusted R2 0.436 0.439 Residual Std. Error (df = 4497) 0.109 0.109 F Statistic (df = 11; 4497) 317.634∗∗∗ 322.301∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 considerably more in the direct aftermath from their revolt than did chairs of the various committees. Again, this seems to show strong evidence in support of

Hypothesis 1B. Table 4.4 shows the results for the 1977 models, and demonstrates a similar finding. While subcommittee chairs became slightly more central in the post-1977 period (p < 0.1), the majority leadership saw a much greater substantive increase

61 Texas Tech University, James P. Bassett, August 2020

Figure 3.1. Predicted Centrality Before and After 1910 Revolt

after the reforms (p < 0.01). In both models, the rank-and-file tended to have lower centrality scores after 1977. Perhaps most notably, the majority leadership showed no statistical significance before 1977, highlighting the importance of the DSG reforms; while they were not completely toothless prior to reform, this model demonstrates that the DSG was successful in empowering the majority leadership.

As before, this shows some very weak support for Hypothesis 2A and much stronger support for Hypothesis 2B. Again, the predicted centrality scores highlight this change. Before 1977, the majority leadership and majority rank-and-file were not statistically distinct in terms of their covoting network centrality, as shown by the overalapping error bars. However, after 1977, the gap between them widened signficantly. Indeed, like in 1910, the treated group saw a relatively small increase in centrality while the non-treated group saw their centrality drop. The predicted centrality shown here again demonstrates strong support for Hypothesis 2B. It is notable that the time trend in the 1977 case is exactly the opposite of that in 1910; in 1910 centrality was trending downward before being mitigated by the treatment and becoming more or less stagnant. By contrast, the 1977 case

62 Texas Tech University, James P. Bassett, August 2020

Table 3.3. OLS Results: 1977 Case

Dependent variable: Centrality (3) (4) Post-1977 −0.035∗∗∗ −0.034∗∗∗ (0.003) (0.003)

Subcomm. Chair −0.008∗∗ −0.003 (0.003) (0.002)

Majority Leadership 0.023∗∗∗ 0.005 (0.005) (0.009)

Post-1977 * Subcomm. Chair 0.006∗ (0.003)

Post-1977 * Maj. Leader 0.025∗∗ (0.010)

Constant −1.221∗∗∗ −1.209∗∗∗ (0.185) (0.185)

Observations 3,060 3,060 R2 0.672 0.672 Adjusted R2 0.671 0.671 Residual Std. Error (df = 3047) 0.043 0.043 F Statistic (df = 12; 3047) 519.811∗∗∗ 519.794∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 occurred during a trend that was otherwise positive. These trends will be discussed in considerably more detail in the following section, but for now I simply note that the difference in trends helps to lend credibility to the underlying mechanism; since I find similar results for the two reforms despite their occurring in such disparate environments, this helps to demonstrate that the relationship between reform and the influence of the inciting groups is not simply a product of the overall trends in

63 Texas Tech University, James P. Bassett, August 2020

Figure 3.2. Predicted Centrality Before and After 1977 Reforms their respective times.

3.4 Centrality in the 20th Century and Beyond

These results demonstrate that institutional reforms do indeed have an impact on the influence of those MCs who are most instrumental in driving that reform. MCs aim to change Congress to benefit themselves; both the insurgents in 1910 and the DSG and majority leadership in 1977 saw their levels of influence rise in comparison to their colleagues. At least at an anecdotal level, these increases in influence also had the intended effect of allowing those members to be more successful in pursuing their own agendas. While the Republican insurgents were not united on any particular issue area, its biggest leaders were animated against Cannon in large part by his opposition to progressive reforms and regulations, and in the wake of their revolt many of these progressive policy goals were successfully enacted (Harrison 2004). Likewise for the DSG, the empowerment of the majority leadership allowed for a much more liberal agenda to take hold in the House, and for a more ideologically extreme agenda overall in the following decades. The long-term results of these reforms, on the other hand, were quite

64 Texas Tech University, James P. Bassett, August 2020 different from each other. Most crucially, as discussed before, the 1910 reforms were essentially decentralizing while the 1977 reforms were essentially centralizing. This effect can be seen in Figure 3.3, which plots the mean level of centrality from the 35th Congress to the 115th. Much like the trends observed in the time-variant scatterplots in Chapter 2, the mean level of centrality is somewhat perplexing to evaluate in terms of its exact meaning. On the aggregate, more members showing up as more central within covoting networks is presumably an artifact of many different things; because it is derived from the correlations in roll call votes, to some extent or another it probably captures aspects of polarization, agenda control, and the power of the party leadership. Of course, all of these factors are deeply entangled with one another, and so while the mean level of centrality cannot distinguish between them or infer causal relationships within that bundle, it can serve as a proxy for the amalgam of polarization/agenda control/leadership power, particularly in the context of the institutional change that creates both the polarizing/agenda-controlling/leadership-empowering process and the covoting network that it implies. That this increase in centrality is the result of an amalgamation of processes is precisely why we are able to see such a stark upward trend over time in spite of scholarship that predicts that rank-and-file members have been increasingly frozen out of the process, which would seem to cause them to be less central (Curry and Lee 2019). Because polarization/agenda control/powerful leadership necessarily creates an increased correlation in members’ votes, the mean level of centrality increases along with it simply as a matter of definition. It is in large part for this exact reason that I focus on congressional fixed effects, rather than a linear time variable; because fixed effects are estimated independently of one another, it allows members to be compared to each other within the distribution of their own congress, rather than to an MC from a past or future period with a much

65 Texas Tech University, James P. Bassett, August 2020

Figure 3.3. Mean Centrality Over Time Note: Highlighted portions represent the periods modeled in this chapter.

different context of polarization/agenda control/powerful leadership. Prior to 1910, the mean level of centrality was prone to bouncing wildly from one Congress to the next. While this is not precisely a trend in itself, this sort of observation is exactly what might be expected when the agenda is at the sole discretion of the Speaker. Aside from vacillating along with the holder of the speakership itself, because of the speaker’s ironclad control of the agenda it would likewise make sense that the ability for members to influence one another would depend on the Speaker’s agenda. In either case, the Speaker’s complete dominance of the era led to on average a much higher mean level of centrality, consistent with the popular characterization of this era as one of considerable polarization that is at least prima facie comparable with the last decades of the 20th Century and first decades of the 21st (Theriault 2008). After the revolt against Cannon, however, the mean level of centrality fell into a relatively stable pattern at the lower end of the spectrum. Again, this is consistent with the expectations of a decentralized chamber where most members are at low or middling levels of influence and the committee chairs mostly control the agenda in their own areas but no one is really able to steer

66 Texas Tech University, James P. Bassett, August 2020 the entire ship. This pattern likewise continued throughout the middle of the 20th Century until the 1977 reforms initiated a period of near-constant increase in the mean level of centrality. In the wake of these reforms, the mean level shot up, again consistent with increasing polarization/agenda control/leadership power, a pattern which has continued all the way until the 115th Congress with a mean centrality score of .778, the highest in congressional history as of this writing. With these patterns and results in mind, the similarities and differences discussed in the previous sections are crucial in understanding why these two moments of reform had such distinct long-term effects: while both sets of reformers aimed to benefit themselves at the expense of their competing intra-party factions, they differed in their methods and intentions. The 1910 insurgents did not have a unified goal, and the organization that proved so valuable for ousting Cannon had no remaining purpose to keep itself together once he was gone. As a result, while the MCs themselves benefited in the short term, in the long term power in the House did not coalesce around anyone new, reverting instead to the only institution that remained: the committees. This comports with the argument put forth by Schickler (2001) which states that reform-minded groups are generally ideologically diverse and not long-lasting once the reforms are accomplished. The DSG, by contrast, executed their plan with an explicit ideological goal in mind. This is a type of reform much more in line with most theories of institutional change. Cox and McCubbins (1993, 2005) for example argue that the reforms of the 1970s were a natural result of the changes in the composition of the Democratic caucus, while DiSalvo (2012) argues that factional conflict changes the ideological coalition of the entire party. Indeed, disaggregating the causal process can show just such a path: intra-party conflict drives the membership toward institutional change (Rohde 1991), which results in the leadership increasingly utilizing its newfound

67 Texas Tech University, James P. Bassett, August 2020 agenda power and polarizing the rest of the party (Theriault 2008), which creates an increasingly homogeneous party that is inclined to delegate more and more power to the leadership (Aldrich and Rohde 2001). With this process in mind, once one faction has enough procedural power to kick it off this can be seen as a self-reinforcing mutually causal relationship. Where these schools of thought agree is that institutional change creates a new structure of incentives for members. The remaking of the congressional infrastructure fundamentally affects MCs’ access to the party resources that they need in order to achieve their electoral, careerist, and policy goals (Fenno 1973). This can mean actual campaign resources, which the leadership gains increasing access to as their power increases (Heberlig et al. 2006), but also the creation of a strong party brand through the passage of legislation (Cox and McCubbins 1993, 2005). While there is no part of institutional change that forces members to behave differently, it remains that because they face such wildly different constraints and incentives, the behavior of both the leadership and the rank-and-file (or in the case of the inter-reform period, the committee chairs) is structured around the institutional environment. The impact of reform in the long term comes in how those key incentives are filtered through to the membership. While members are all subject to the incentive of winning reelection by the creation of a strong party brand through the passage of legislation, the institutional structure changes every aspect of what exactly that means: what constitutes a strong brand, what defines good legislation, perhaps even what constitutes a real majority when a major faction within your own party is dedicated to working with the opposition on a substantial part of the agenda. The eventual resorting of congressional parties along the newly-defined partisan lines in the wake of 1977 demonstrates this reshuffling of incentives quite clearly.

68 Texas Tech University, James P. Bassett, August 2020

These schools also agree that institutional change is generally quite incremental. Even in the case of the exception of 1910, while the Speaker’s power was reduced practically overnight, the reversion to committee government was comparatively gradual. The upshot of this incrementalism is that institutional changes years or even decades in the past have long-lasting reverberations. Taking a long arc view of congressional influence in the 20th Century, these changes follow a clearly path-dependent track, wherein the Republican insurgency strips the Speaker of power with no leadership structure to fill the void, the committees and their chairs eventually step in, and when the committees become obstructionist under the stewardship of the conservative coalition of Republicans and Southern Democrats (the only part of this telling that is truly exogenous), the growing liberal wing of the Democratic party takes steps to move the power back to themselves, eventually setting up the conditions for the increasing runaway elite polarization that has come to define the 21st Century in Congress. While this type of for-want-of-a-nail historicizing is of course extremely reductive, it does serve to highlight the key themes of this chapter that the long term effects of MCs’ short term moves to benefit themselves are unpredictable, regardless of the size and scope of the individual moves themselves.

69 Texas Tech University, James P. Bassett, August 2020

CHAPTER 4 CENTRALITY BEHIND THE SCENES: PROCEDURE AND PASSAGE

On October 12th, 2011, the House of Representatives held a series of votes on a trade agreement between the United States and Colombia. As is typical, the chamber held roll calls on a series of procedural motions prior to their vote on final passage. Of note, however, were the changes in vote totals between votes. On a to recommit the bill to committee, which if successful would effective kill the bill, the parties voted with near-perfect discipline: 235 Republicans for saving the bill against 188 Democrats trying to vanquish it to committee. On final passage, however, 32 Democrats seemingly had a change of heart, joining with the Republicans to pass the bill 263-167 (Lewis et al. 2019). What causes members to change their votes between different roll calls on the same bill? As several studies have noted (e.g., Ansolabehere et al. 2001; Jenkins et al. 2005), there is considerable variation in how members vote on procedural votes as compared to their votes for final passage. Generally, this distinction is attributed to the electoral connection: if members of Congress perceive that a vote will not be popular with their constituents, then members of the party leadership may allow them to vote against the bill for final passage as a sop to their constituents on the condition that they vote to advance the bill on the various procedural motions that bring it to the floor. Clearly, such procedural votes are a key arena for the expression of party influence. Parties, however, are not autonomous, self-determining entities; they are composed of individuals. As discussed in the previous chapter, party influence is deeply interwoven with the influence of its members. How, then, does the signaling

70 Texas Tech University, James P. Bassett, August 2020 influence of members of Congress (henceforth MCs) manifest itself differently across these different contexts? This chapter aims to investigate the differences in who matters when it comes to final passage of a bill as opposed to on procedure. We should expect, for example, that members of the party leadership should have higher levels of centrality on procedural votes than on passage specifically because of the effect proposed by Snyder Jr and Groseclose (2000) – leaders being much more forceful in attempting to control the actions of the rank and file on procedural motions, then loosening the reins somewhat on the final passage vote for MCs who need to appease their districts. As usual, this chapter begins with an overview of the literature surrounding procedure and passage, and the differences between them. This literature generally explains these distinctions through the lens of party discipline; as a result, this chapter will attempt to zoom in, refocusing the discussion to center on the intersection of party discipline and individual behavior. Party discipline in this context can therefore be conceptualized as the result of individual efforts to enforce whipped votes where they matter and allow defections where they do not. Centrality is useful in this endeavor in part because it allows a focus on how all members use their power of influence, not just party leaders. While party leaders are expected to be among the most influential, it also allows us to investigate the influence that the rank-and-file exert over one another. Considering the literature surrounding the asymmetric effects of party influence on extremists and centrists (e.g. Minozzi and Volden 2013), this approach provides us with a rare opportunity to measure the influence of the party using the entire social network of Congress. Next, the chapter will move into its empirical section. Using roll calls from 1973-2008 separated by procedure and passage, I demonstrate significant influence among party leaders and majority party members in general on procedure, with

71 Texas Tech University, James P. Bassett, August 2020 relatively little influence detected on passage. In addition, as the chamber has become more polarized over the years, this effect has only grown stronger. As discussed in the previous chapter, polarization is extremely entangled in the covoting centrality of its members. This chapter’s contribution, therefore, is in its strong concurrence with the findings of Theriault (2008) on the relationship between polarization and an increasing reliance on legislating through procedure. This chapter shows that as the chamber has become more polarized, legislators’ influence is increasingly relevant when it comes to procedural votes, but, critically, almost entirely stagnant on final passage votes. This finding allows considerable insight into the relationship between polarization and influence, especially in the context of this dissertation’s other findings. Finally, the chapter concludes with a brief discussion of the results.

4.1 Procedure and Passage in the U.S. Congress

No discussion of the process of congressional legislation can begin without the requisite quote from Rep. (D-MI)1, but its place in the lore is well-deserved: the importance of procedure is clearly established within the congressional literature. No bill springs into the world fully formed, and as Krehbiel and Woon (2005) discuss at length, there is a considerable amount of procedural work that must be done before a final passage vote can be held. In their discussion of party cohesion, Froman and Ripley (1965), hypothesized a procedure-substance axis for congressional activity, formulating it into a seven-part continuum with election of the speaker and adoption of rules as the most procedural and amendments as the most substantive. The difference in levels of cohesion is typically attributed to the electoral connection (Mayhew 1974). Because procedural

1”If you let me write the procedure, and I let you write substance, I’ll screw you every time.”, quoted in Evans (1999).

72 Texas Tech University, James P. Bassett, August 2020 votes are less visible to their constituents, MCs can vote the party’s wishes on them without fear (or at least with less fear) of their constituents voting them out in retribution (Arnold 1990). One of the most important theories of congressional procedure has been the cartel theory, developed and popularized by Cox and McCubbins (1993, 2005). According to this theory, party leadership forms a “procedural cartel”, utilizing negative agenda control to prevent votes which might split the party and cause them to be rolled, or have a policy passed by the minority party with the help of a minority segment of the majority party. The use of negative agenda control in this way allows leaders to effectively move the status quo in their preferred direction, away from the floor median, by restricting votes to those issues which have both broad party support and specifically the support of the leadership (Monroe and Robinson 2008). Indeed, Lawrence et al. (2006) argue that majority power seems to explicitly come from their ability to control the agenda. However, despite being a theory of negative agenda control, reality dictates they must also pass controversial legislation themselves. Evidence suggests that negative agenda control is deeply intwined with positive agenda control (Finocchiaro and Rohde 2008). In order to promote a strong party brand of accomplishment with the goal of being reelected into the majority by their voters, cartel theory and its many offshoots dictate that members must pass legislation that is amendable to the party at large (Carson et al. 2014). Even this, however, is not as straightforward as it may seem, as voters have a tendency to punish MCs for excessive partisanship on divisive votes (Carson et al. 2010). As a result, the leadership may allow for MCs to vote against the party on final passage as long as they agree to toe the party line on the associated procedures (Ansolabehere et al. 2001; Jenkins et al. 2005). As a result, in the wake of challenges to conventional wisdom of party

73 Texas Tech University, James P. Bassett, August 2020 influence (most notably by Krehbiel 1993, 2010), procedure has become a key testing ground for those looking for the influence of parties on congressional behavior. Binder et al. (1999), for instance, found strong party effects on procedure using Krehbiel’s own standard of the “waffling analysis”, which looks at those who changed their votes in the manner of the anecdote at the beginning of this chapter. Many studies, including Cox and Poole (2002), Ansolabehere et al. (2001) and Jenkins et al. (2005) find strong party effects on procedure and very little on final passage. The leadership that creates these party effects can take many forms. In some cases, it can be outright vote-buying through the use of campaign contributions (Jenkins and Monroe 2012). This can also take the form of particularistic benefits, in which party leaders pay off more moderate members of the party in order to guarantee their support in moving the policy agenda closer to the party (Carroll and Kim 2010). In other scenarios, it can be through the use of the whip system, in which the designated whips act as a go-between for the leaders and the rank-and-file, collecting information on where the members sit on a particular piece of legislation and building consensus through various means of arm-twisting (Evans and Grandy 2009). This can, for example, be policy concessions in the form of amendments (Lynch et al. 2016), or outright persuasion in the name of the broader party strategy (Roberts and Smith 2003). In still other situations, leaders are able to signal their positions to the party en masse. This can mean something explicit, such as a document like the “leader’s floor update” (Carson et al. 2014), or something more general (Minozzi and Volden 2013). Notably in this case, leaders are able to signal to the entire party at once; rather than focusing all their efforts on one particular member, they can simply make their position known to all MCs. As Carson et al. (2014) demonstrate, votes that are signaled in this way attract greater

74 Texas Tech University, James P. Bassett, August 2020 party support, especially among ideological extremists. The use of party influence in this way is not without cost. The increased importance of such procedural matters has been cited by Theriault (2008) as being a key cause in the runaway ideological polarization observed since the late 1970s and early 1980s. These changes have coincided with changes over time in the relationship between procedure and passage. Indeed, party effects on procedure have grown over the course of this period, while their observable effects on final passage have remained relatively static (Jessee and Theriault 2012). The divisiveness of these procedures may have other effects as well, including masking the variability that might otherwise occur on final passage (Theriault 2006; Froman and Ripley 1965). This illustrates a key concern with the usage of roll call data, especially in such a disaggregated state. Recurring throughout this chapter – indeed, throughout this dissertation – is the issue of the process through which roll calls are generated. Despite the tendency of scholars to treat the roll call record as exogenously generated, the fact remains that roll calls require a positive procedure from the chamber: a member must request a vote and have one-fifth of the chamber agree. While often treated as a matter of course, this does leave a non-trivial number of votes which are recorded viva voce, or through voice vote. This results in the roll call record being largely populated with votes that members want to be seen taking a stand on (Lynch and Madonna 2013). On one hand, this somewhat contradicts the idea of the obscure and invisible procedure vote that members can hide from. As Clinton and Lapinski (2008) note, a relatively small proportion of the laws which are eventually passed do so via roll call. On the other hand, this is somewhat mitigated by the relationship between procedure and policy; as Clinton (2007) describes, the two are deeply entangled. In addition, the bar for requesting roll calls is relatively low, and easily meetable by the minority on their own, allowing them to

75 Texas Tech University, James P. Bassett, August 2020

Table 4.1. Vote Counts, 93rd to 110th Congress

Congress Procedure Passage 93 105 348 94 138 356 95 165 298 96 138 223 97 77 135 98 109 151 99 124 126 100 108 149 101 101 158 102 134 163 103 155 143 104 189 178 105 178 152 106 142 185 107 158 125 108 150 154 109 155 142 110 343 170 force the majority onto the record. Given the observation by Curry (2015) that procedural votes tend to be highly polarized even when the underlying issue is not, the result of this endogenously-defined process of creating roll calls into the record is that the process must be considered throughout the process of studying them, especially with respect to the time dimension as the proportion of roll call votes has shifted increasingly toward procedure over time (Table 4.1). As I discuss in the next section, this is an important reason why the time dimension plays such an important role in determining the shape and context of the roll call record. Throughout this narrative, the emphasis is almost entirely on majority parties. The House is a majoritarian body, and as such, the benefits to being in the majority are significant. As seen in Chapter 2, majority status is a strong predictor not just of overall covoting network centrality, but of LES as well (Volden and

76 Texas Tech University, James P. Bassett, August 2020

Wiseman 2014). Cox and McCubbins (1993) describe their narrative as being focused on the majority as early as page 2, and the theory is centered around the desire for leaders of the procedural cartel to make sure their party maintains their status. The reason for this is simple: being in the majority means a near-monopoly on the ability to pass meaningful legislation. The minority party simply does not have access to most of the most useful resources that are needed to create policy. As a result, for the most part, studies that aim to find party influence give fairly little attention to the minority. Since their minority status tends to absolve them of blame, there is very little reason to exercise much party power on dissenting members (Cox 2001). With this in mind, any search for influence should likely be focused on the majority. In any case, the narrative surrounding the difference in legislative setting between procedure and passage is fairly straightforward, and as such the hypothesized relationship between that difference and covoting network centrality as signaling influence is straightforward as well.

Hypothesis 1: Party leaders are more central than the rank-and-file on procedural votes, but no more on final passage votes.

Hypothesis 2: Majority party MCs are more central than minority party MCs on procedural votes, but no more central on final passage votes.

Hypothesis 3a: The difference between centrality on procedure and centrality on passage is greater for party leaders.

Hypothesis 3b: The difference between centrality on procedure and centrality on passage is greater for majority party members.

Hypothesis 4a: The difference between centrality on procedure and centrality on passage is greater for ideologically extreme members.

Hypothesis 4b: The difference between centrality on procedure and centrality on passage is lower for members who are far from the party mean.

Hypotheses 1 focuses on party leadership. It predicts that, as suggested by

77 Texas Tech University, James P. Bassett, August 2020 the existing literature, the influence of party leaders will be felt much more strongly on matters of procedure than on final passage. This comes as a direct result of the resources that party leaders have to control the agenda. As argued by, for example, Curry (2015), leaders have considerable advantages over the rank-and-file in information, and they use that information to pull the policy agenda in their preferred direction by way of strategic disbursement of information to their colleagues. Hypothesis 2, on the other hand, looks at majority status. It predicts a higher level of centrality for majority party members than on minority party members on procedure, with no difference on passage. This follows from the literature surrounding the value of being in the majority, and the relative lack of party influence that can be found there (Cox 2001). As in the previous hypothesis, Hypothesis 2 also predicts no significance for passage votes. This similarly follows from findings such as Lawrence et al. (2006) which demonstrate that majority members typically have win rates in the vicinity of 100% on final passage of legislation, an outcome consistent with strict agenda control by the party leadership. Hypothesis 3 attempts to synthesize the previous two hypotheses in order to explore the difference in procedure and passage directly. It specifies that not only are party leaders and majority members more central on procedure, the gap between their procedural influence and passage influence will be greater as well. In other words, not only are they more influential on procedure, they are more more influential on procedure. Finally, building off of Hypothesis 3 and the findings of Chapter 2, Hypothesis 4 once again explores the effects of ideological distance from the party as conceptualized through party difference versus extremity. The next section will explain the mechanical difference between these measures in more detail, but the essential finding from Chapter 2 is that members who are further out (i.e., further left for Democrats and further right for Republicans) tend to display

78 Texas Tech University, James P. Bassett, August 2020

higher levels of centrality, while members who are moderate tend to display lower levels. Because Carson et al. (2014) and Minozzi and Volden (2013) find such strong effects on ideological extremists, this hypothesis intends to further investigate this relationship. The next section will discuss the research design that will be used to evaluate these hypotheses.

4.2 Research Design

The sample for this chapter comes from the 93rd to 110th Congress, covering a time period from 1973 to 2008.2 This chapter uses OLS regression, with Procedure Centrality and Passage Centrality as the dependent variables. As usual, covoting network centrality for any given MC is calculated as the average absolute value of the correlation between their roll call voting record and the record of each of their colleagues3 In order to test Hypothesis 3a and 3b, I also use a measure which subtracts Passage Centrality from Procedure Centrality. Since Procedure Centrality is typically higher (only about 5% of observations had higher centrality on procedure than on passage), this helps to demonstrate the magnitude of the difference that is predicted by the other two models. Unlike previous chapters, these votes are disaggregated according to procedure and passage, and separate centrality scores are calculated for each. As a result, each legislator has two scores.4 Following the lead of Snyder Jr and Groseclose (2000), votes on rules, motions to end debate, and motions to recommit are coded as procedure votes, while passage votes are those which are on final passage of a bill, final passage of a conference report, or on overriding a presidential veto. All told, across 18 Congresses, the centrality scores for this chapter are

2Once again, this data comes from Volden and Wiseman (2014). 3A more comprehensive description appears in Chapters 1 and 2. 4Roll calls are sourced from Voteview, and the vote types used to categorize procedure and passage are based on the PIPC Roll Call Database.

79 Texas Tech University, James P. Bassett, August 2020

Figure 4.1. Histogram of Centrality on Procedure and Passage

calculated from 4384 procedure votes and 5442 passage votes. On a purely descriptive level, centrality as calculated on procedure is much more evenly distributed than on passage, the curve of which is more traditionally bell-shaped and centered around .200 or so, as shown in Figure 4.1.

The main covariates for this chapter are leadership for Hypotheses 1 and 3a

and majority status for Hypotheses 2 and 3b. Leadership is coded as members who serve in any elected leadership position, either in the majority or the minority.5 This measure is agnostic to majority or minority status; in order to distinguish the effects of leadership itself from the effects of majority status, leaders of both the majority and minority party count equally for this Hypothesis. Majority status is coded as a simple binary. Notably in this sample, 13 of the 18 congresses measured had Democratic majorities. While this does not completely negate the value of using party as a variable, it is a sufficient lack of variance to drop as a measure. In addition, the models used to test Hypothesis 4 also include measures of ideological extremity and party difference. These are related measures which are

5Coded by Volden and Wiseman from The Almanac of American Politics, these positions include the Speaker, Majority and Minority Leader, Majority and Minority Whip, Chief Deputy Whips from both parties, Conference/Caucus Chair and Vice Chair, Republican Conference Secretary, Republican Policy Committee Chair, and Assistant Democratic Leader.

80 Texas Tech University, James P. Bassett, August 2020

Table 4.2. Extremity versus Party Difference

MC NOMINATE Party Mean Extremity Party Difference Legislator 1 -.750 -.500 .250 .250 Legislator 2 -.250 -.500 -.250 .250 Legislator 3 .250 -.500 -.250 .750 intended to distinguish members who are ideologically extreme from those who are more moderate from the party. These measures are calculated for Legislator i as:

Extremityi = |NOMINATEi | − |PartyMeanNOMINATE| (4.1)

PartyDifferencei = |NOMINATEi − PartyMeanNOMINATE|

The difference is subtle but important. Consider, for example, a hypothetical Legislator 1, Legislator 2, and Legislator 3. Legislator 1 has a first-dimension DW-NOMINATE score of -.750, while Legislator 2 has a score of -.250 and Legislator 3, a Southern Democrat, has a score of .250. All three are Democrats, whose party mean NOMINATE score is -.5. The difference between their results is shown in Table 4.2. The crucial distinction is that Legislators 1 and 2 look identical in Party Difference because they are equally far from the party mean. Extremity clarifies that Legislator 1 is further to the left than the mean Democrat while Legislator 2 is more moderate. Legislator 3, by contrast, looks exactly as Extreme as Legislator 2 due to their similar distance from 0, while correctly looking much farther from the party in Party Difference than the other two. These distinctions mean that while both measures are valuable, they tell us different stories about the MCs in question. These models also use several controls. For the most part, these controls mirror the institutional factors found to have the most impact in Chapter 2:

81 Texas Tech University, James P. Bassett, August 2020 committee chair, subcommittee chair, power committee membership and budget committee membership. As always, models are estimated using fixed effects for Congress to account for the increase over time documented in Chapter 3, and with clustered standard errors on MC to account for correlation within the same members from session to session. Especially in light of the findings in Chapter 3 which explore the increase in the mean level of centrality over the course of this chapter’s sample, this helps to control for the overall trajectory of members’ centrality over the time period of this sample.

4.3 Results

The results of my analysis appear in Table 4.3. I find strong support for both Hypotheses 1 and 2: Leader and Majority are both significant and positive (p < 0.01) for centrality on procedure, while on passage they are either not significant (p = 0.39) or significant and negative (p < 0.01) for leadership and majority party, respectively. Overall, on procedure being a leader was associated with a .095 point increase over non-leadership, while on passage there was no statistical difference between them. Likewise, majority status was associated with an increase of .102 in centrality over minority members on procedure, while on passage majority members were actually less central than their minority counterparts. The substance of these findings is illustrated in Figures 4.2 and 4.3.6 On procedure, leaders remain more central than the rank-and-file even as centrality increases markedly over time, while on passage, the predicted levels between leaders and non-leaders overlap almost perfectly. Similarly, in the case of majority status in Figure 4.3, majority members are more central than minority MCs on procedure by

6Predictions are made with all covariates except leadership set equal to 0, indicating a rank-and- file member. As in Chapter 3, these plots are OLS predictions using Congress fixed effects, so the parallel movement of the lines for the two groups should not be interpreted as indicative of anything except an OLS prediction.

82 Texas Tech University, James P. Bassett, August 2020

Table 4.3. OLS Results: Hypotheses 1 and 2

DV: Centrality on... Procedure Passage (1) (2) Leader 0.040∗∗∗ 0.007 (0.007) (0.006)

Majority 0.118∗∗∗ −0.066∗∗∗ (0.004) (0.002)

Comm. Chair 0.024∗∗∗ −0.005 (0.007) (0.004)

Subcomm. Chair −0.004 0.001 (0.004) (0.002)

Power Comm. 0.012∗∗ −0.001 (0.005) (0.003)

Budget Comm. 0.008 −0.001 (0.005) (0.003)

Constant 0.399∗∗∗ 0.211∗∗∗ (0.006) (0.002)

Observations 7,900 7,900 R2 0.741 0.434 Adjusted R2 0.740 0.432 Residual Std. Error (df = 7876) 0.101 0.056 F Statistic (df = 23; 7876) 978.608∗∗∗ 262.472∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 a margin that falls just short of a full standard deviation. By contrast, on passage it is actually the minority members who are more central, a finding which comports with the observation by Cox (2001) of more defection by the minority on such votes; since their vote is less constrained by party, members’ individual influence is more

83 Texas Tech University, James P. Bassett, August 2020

Figure 4.2. Predicted Centrality - Leadership

impactful.

Table 4.4 shows the results for Hypotheses 3a and 3b. As before, the difference between procedure and passage centrality is greater for party leaders and much greater for majority members. This again shows strong support for both parts of Hypothesis 3. Not only are leaders and majority MCs more influential on procedure than their non-leader or minority counterparts, the magnitude of the effect of that change of setting is higher for both groups as well. The difference between procedure centrality and passage centrality is .033 points greater for leaders than for the rank-and-file (p < .01) and .184 points greater for majority MCs than for minority MCs (p < .01). Notably, the substantive impact of leadership is nearly equaled by that of committee chairmanship (.029, p < .01). While not explicitly hypothesized, the same underlying logic for party leadership applies for committee chairs as well, and offers a sort of backdoor confirmation of Hypothesis 3. This finding is demonstrated substantively in Figure 4.4. As before, while the time dimension demonstrates considerable variation, leaders are marginally more central than non-leaders, while majority members are considerably more central than

84 Texas Tech University, James P. Bassett, August 2020

Figure 4.3. Predicted Centrality - Majority Party minority MCs. Turning to Hypothesis 4, as hypothesized on the basis of the findings in Chapter 2, more extreme members show a larger gap in Procedure Centrality and Passage Centrality than do more moderate ones (p < .01), while members with higher levels of party difference are less central (p < .01). Again, this can be attributed to the recoding of more moderate members from positive to negative on extremity. This finding indicates that it is the leftmost Democrats and rightmost Republicans who are the most important influencers on procedure, in agreement with Carson et al. (2014) and Minozzi and Volden (2013). Substantively, a member whose extremity score is one standard deviation above the mean would be predicted to have a difference in procedure and passage centrality scores about .098 higher than one who is a standard deviation below the mean. Members meeting the same criteria for party difference are predicted to have a procedure-passage centrality difference of about .047. The difference between them is shown in Figure 4.5; as party difference increases (which I again note is in either direction away from the party mean), the gap between members’ procedure and passage centrality becomes

85 Texas Tech University, James P. Bassett, August 2020

Table 4.4. OLS Results: Hypotheses 3 & 4

Dependent variable:

CP rocedure − CP assage (1) (2) (3) Leader 0.033∗∗∗ 0.015∗ 0.030∗∗∗ (0.008) (0.008) (0.009)

Majority 0.184∗∗∗ 0.184∗∗∗ 0.187∗∗∗ (0.004) (0.004) (0.004)

Comm. Chair 0.029∗∗∗ 0.020∗∗ 0.028∗∗∗ (0.008) (0.008) (0.008)

Subcomm. Chair −0.005 −0.007 −0.004 (0.005) (0.004) (0.004)

Power Comm. 0.013∗∗∗ 0.012∗∗∗ 0.008∗ (0.005) (0.004) (0.005)

Budget Comm. 0.008 0.0003 0.006 (0.005) (0.004) (0.005)

Extremity 0.295∗∗∗ (0.016)

Party Diff. −0.187∗∗∗ (0.033)

Constant 0.022∗∗∗ 0.020∗∗∗ 0.049∗∗∗ (0.004) (0.004) (0.006)

Observations 7,900 7,883 7,883 R2 0.712 0.773 0.727 Adjusted R2 0.711 0.773 0.726 Residual Std. Error 0.107 (df = 7876) 0.094 (df = 7858) 0.104 (df = 7858) F Statistic 847.831∗∗∗ (df = 23; 7876) 1,117.131∗∗∗ (df = 24; 7858) 872.933∗∗∗ (df = 24; 7858) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 smaller, while for extremity, where moderate members are negative, MCs become more more central on procedure than on passage. It is again worth noting that most of the substantive significance found in this chapter is relatively small. None of the findings for party leadership approach an effect size that exceeds a standard deviation for their respective dependent variables. The effect size of majority status is greater across the board, coming just short of a standard deviation in size in each model. While the effect size is fairly

86 Texas Tech University, James P. Bassett, August 2020

Figure 4.4. Predicted Difference in Centrality - Leadership and Majority Party

Figure 4.5. Predicted Difference in Centrality - Party Difference and Extremity small in most models, the finding remains strong on the basis of its consistency, and in the context of the rest of the findings that this dissertation has presented. While small, the party leadership, majority status, and extremity results that I find in this chapter are robust to a wide variety of specifications as demonstrated in previous chapters, and are similar in magnitude to the effect sizes found in Chapters 2 and 3. As a result, I consider this a strong and consistent finding despite the relatively minor substantive significance.

87 Texas Tech University, James P. Bassett, August 2020

4.4 Discussion

This chapter presents strong findings on the effects of party leadership and majority status on MCs’ signaling influence on procedural votes. Overall, I find that party leaders and majority MCs are more influential than their rank-and-file and minority colleagues on procedural votes, while being no more influential on matters of final passage. This is a result that comports very closely with the existing literature on the distinction between these two legislative contexts, and offers a new and unique data point in the ongoing study of how members of Congress react differently to the different constraints of the two settings. The findings I present here offer a crucial broadening of the literature surrounding procedure and passage, and, by proxy, party influence writ large. Most theories of party influence have tended to focus on party leaders (Carson et al. 2014), ideological extremists (Minozzi and Volden 2013), or some combination thereof (Theriault and Rohde 2011; Theriault 2013). This analysis opens up the influence of party to include the impact of the entire social network that forms within that party. I demonstrate that majority party members are more influential than minority members particularly in those arenas where party influence is deemed to be the most powerful – on procedure. In addition, I show that the loosening of party influence on passage votes (a phenomenon which as been widely discussed, in particuar by Snyder Jr and Groseclose 2000 and Ansolabehere et al. 2001) is not exclusive to leaders. Rather, passage appears to be relatively free of influence in general. This is likely a result of a combination of several effects; the offering of free defections to members who need to satisfy their party, the procedural cartel restricting votes on final passage depending on the outcomes of the procedural votes, and the lack of necessity for whipping votes when the outcome is already known. In any case, this chapter offers another entry in the procedure-passage canon.

88 Texas Tech University, James P. Bassett, August 2020

Figure 4.6. Scatterplot of Procedure/Passage Centrality Over Time

In the broader context of this dissertation, the findings of this chapter offer crucial clarification and context particularly for the increase in mean levels of centrality discussed in Chapter 3. The findings here demonstrate that the lion’s share of that increase in centrality, attributed in Chapter 3 to polarization, comes on procedural votes. This pattern is clearly identified in Figure 4.6.7 As identified along the X-axis, centrality on passage is relatively stable across time, with a slight tendency to concentrate on the higher side (relatively speaking) in later congresses. The story for procedure on the other hand, is quite clear – regardless of members’ influence on passage, their centrality on procedure clearly moves up over time, and becomes increasingly clustered as members’ votes become more and more correlated under conditions of polarization. This offers a powerful statement in support of the theory of procedural polarization offered by Theriault (2006). Over the course of this period of time when ideological polarization and overall levels of covoting network centrality were rapidly increasing, the levels of centrality on final passage votes remain almost entirely static. This suggests that, as Theriault (2006)

7Of the major outliers on passage, three are Speaker Dennis Hastert.

89 Texas Tech University, James P. Bassett, August 2020 hypothesizes, polarization occurs mainly on procedural matters, with final passage being an arena which is more free from partisan influence. Because procedural matters are so important and so opaque to voters, the reigns become much looser on final passage as MCs are allowed to cast votes that will look good back home. The findings here suggest that this has remained the case even as Congress has become increasingly polarized.

90 Texas Tech University, James P. Bassett, August 2020

CHAPTER 5 CONCLUSION

This dissertation offers a unique perspective on influence in the U.S. House of Representatives. By exploring covoting network centrality in a variety of legislative contexts, settings, and periods, I demonstrate its usefulness as a measure of influence despite the limitations that come with all roll call voting measures in the House. In Chapter 2, I considered centrality alongside existing measures of congressional relationships and influence, namely Legislative Effectiveness Scores (Volden and Wiseman 2014) and Cosponsorship Connectedness (Fowler 2006). I explored the differences between all three measures, and put them head-to-head in measuring a wide variety of covariates. I focused on three main types of covariates: institutional positions, marginalized groups, and legislative context. My final combined model demonstrated that, in particular, centrality is uniquely able to measure the influence of party leaders, ideological extremists, and majority status. Chapter 3 looked at how centrality changed on either side of major moments of institutional change. By looking at centrality before and after the rebellion against Speaker Joseph Cannon in 1910 and the reforms of the Democratic Study Group in 1977, I demonstrated that those reforms were successful at increasing the centrality of those members who pushed for them relative to their colleagues. The insurgent Republican faction saw the gap in centrality between them and the rest of the chamber shrink, and the party leadership saw the gap between them and the rank-and-file grow. This falls in line with existing theories of institutional reform. In this chapter I also explored how these reforms interacted with the ongoing story of partisan polarization. I showed how the disruptive effects of these reforms contributed to the de-polarization of the mid-20th Century, and how the

91 Texas Tech University, James P. Bassett, August 2020 re-empowerment of the party leadership during the 1970s kicked off a mutually causal cycle of strengthening leadership and polarization. This chapter, therefore, attempts to use centrality to synthesize the extant literature on Conditional Party Government and polarization, and I believe it illustrates clearly the self-reinforcing relationship between them. Finally, in Chapter 4 I examined how centrality manifests itself differently on procedural votes as compared to on votes on final passage of legislation. I showed that, as predicted by a wide body of literature, influential members see most of their influence show through on matters of procedure, with passage votes seeing relatively little influence exerted on it by comparison. This gap is especially pronounced on party leaders and majority members. In addition, I found in this chapter that ideological extremists tend to be more influential on passage than middlemen, a finding which again comports with existing scholarship.

5.1 Future Research

The findings in this dissertation present many opportunities for future work on the overarching centrality project. First, how does centrality change over time within a particular congress? While the measure of centrality as it is currently conceptualized is fixed and static within a congressional session (as are most quantitative measures of congressional activity), this question aims to unpack exactly what allows members to be influential by transforming centrality into a measure which is dynamic throughout the course of a member’s time in office. In point of fact, this would aim to calculate moving centrality scores by using a rolling sample of roll calls, rather than the population of roll calls that take place in a full congress. In order to investigate this, this study would look at this dynamic centrality

92 Texas Tech University, James P. Bassett, August 2020 in relation to events which may be expected to cause members to gain or lose influence. This could take many forms. Certainly it includes legislative activity, such as the passage or failure of bills or amendments that an MC has sponsored or proposed. It could also, depending on the availability of data, include things which occur outside the halls of the Capitol, such as changes in poll numbers, becoming embroiled in a scandal (assuming the scandal does not end in resignation), or increased media coverage from some other source. While this avenue of research is fruitful on the surface, it also presents the most daunting empirical challenges. First is the measure itself – while centrality has shown itself to be reliable on a congress-by-congress basis, there is no clear answer to how many votes are needed to make it a reliable and valid measurement of influence at any given time. This chapter, therefore, would require considerable experimentation with the measure before any of the other challenges could even begin to be considered. Second, this research design presents a major challenge purely in terms of data and methodology. Even setting aside the question of data availability (itself a concern when dealing with data that is, at the moment, on an unknown time scale), attempting to collate multiple variables which may be on very different time scales is, at least at first glance, a fairly monumental task. A second avenue for future research involves gender. Scholarship is largely in agreement that women tend to be marginalized in legislatures. Despite that women have been demonstrated to have similar attitudes as men on most issues, these attitudes cannot be translated into actions (Schwindt-Bayer 2006). Indeed, as several studies have shown, women are frequently excluded from both formal and informal channels of influence (Barnes 2016). Centrality has major potential as a measure of female marginalization. First, it allows an examination of women’s influence in the legislature as a whole, rather than restricting it to committees or

93 Texas Tech University, James P. Bassett, August 2020 other formal and informal structures. Removing this limitation which has been present in many similar studies allows for a much broader interpretation of what marginalization means: instead of a nebulous idea that women are being placed in positions that they may or may not like, centrality can give a much more concrete indication that women are not just being forced into women’s committees but are being broadly excluded from the entire policymaking process. Second, centrality allows a more systematic analysis not just of the binary question of “is marginalization present”, but more generally at the overall level of marginalization. Because centrality quantifies the idea of legislative influence, it therefore allows us to compare how influential the average woman legislator is to the average male legislator. Thus, not only can we consider marginalization as a matter of scale, we can compare how much women are marginalized in a particular legislature across time and with a variety of individual and chamber-level covariates. While Chapter 2 did not demonstrate strong statistical significance for women overall, the usefulness of centrality overall means that it is worth unpacking in a more focused study.

5.2 Overarching Themes

Broadly speaking, I find that centrality is an effective lens through which to look at the influence of party leaders. In Chapter 2, centrality shows unique results for members of party leadership when compared to more traditional measures such as legislative effectiveness and cosponshorship connectedness. Both other measures find either a negative or no influence of leadership and ideological extremity, while centrality finds positive results for both, albeit at a somewhat marginal substantive level. This finding carries through into Chapter 3, where I find that leaders became more influential over time after the reforms to empower them in the 1970s, and into Chapter 4, where I demonstrate that leaders are specifically more influential on

94 Texas Tech University, James P. Bassett, August 2020 procedural votes than on final passage. As I will discuss momentarily, the consistency of this finding across both time and context gives the impression that this metric is picking something real in the way that legislators respond to their party leaders. Throughout this project I also found that centrality is a useful tool for examining the influence of ideologically extreme members. In Chapters 2 and 4 I discussed at length the distinction between extremity and distance from ones party, and emphasized the importance of distinguishing moderates and extremists who might otherwise look similar by simply looking at the distance from the party mean. In both chapters, I found that more ideologically extreme members tend to be more influential than their moderate counterparts. Given the literature examining questions like if party leaders tend to be middlemen or extremists (e.g., Clausen and Wilcox 1987; Jessee and Malhotra 2010; Grofman et al. 2002; Heberlig et al. 2006), I consider this to be an important datapoint in that debate. Finally, I do find the substantive impact of my findings worth discussing. By and large, the findings I present in this dissertation tend to be of relatively low substantive significance across the board; in only a handful of cases does the regression coefficient even approach the standard deviation for centrality in the given sample, and models which examine issues of party leadership, for example, tend to demonstrate changes which at first glance might seem to hardly be noticeable. On one hand this may be reason for pause when it comes to drawing wide-ranging conclusions on the basis of these chapters. On the other hand, the relatively abstract nature of this measure in some ways obscures the possible range of variation. After all, even in the original Ringe and Wilson (2016) paper which proposed covoting network centrality as signaling influence, none of the effect sizes in their model exceeded .025. And so while it may be an unsatisfying answer, I feel

95 Texas Tech University, James P. Bassett, August 2020 confident that despite the relatively small coefficients that I present throughout this dissertation, the consistency that I find across model specifications indicates the presence of something real and substantive – a measure of legislators’ influence that can work across time and setting to illuminate new aspects of their behavior.

96 Texas Tech University, James P. Bassett, August 2020

BIBLIOGRAPHY

Aldrich, J. H., 1995: Why parties?: The origin and transformation of political parties in America. University of Press.

Aldrich, J. H. and D. W. Rohde, 2001: The logic of conditional party government: Revisiting the electoral connection. Congress Reconsidered, L. C. Dodd and B. Oppenheimer, Eds., CQ Press, Washington, DC, 269–292.

Alem´an,E., E. Calvo, M. P. Jones, and N. Kaplan, 2009: Comparing cosponsorship and roll-call ideal points. Legislative Studies Quarterly, 34 (1), 87–116.

Ansolabehere, S., J. M. Snyder Jr, and C. Stewart III, 2001: The effects of party and preferences on congressional roll-call voting. Legislative Studies Quarterly, 533–572.

Anzia, S. F. and C. R. Berry, 2011: The jackie (and jill) robinson effect: why do congresswomen outperform congressmen? American Journal of Political Science, 55 (3), 478–493.

Aoki, A. L. and D. T. Nakanishi, 2001: Asian pacific americans and the new minority politics. PS: Political Science and Politics, 34 (3), 605–610.

Arnold, R. D., 1990: The logic of congressional action. Yale University Press.

Bach, S. J. and S. S. Smith, 1988: Managing uncertainty in the House of Representatives: adaption and innovation in special rules. Brookings Institution Press.

Bachrach, P. and M. S. Baratz, 1994: Two faces of power. Power: Critical Concepts, 2, 85.

Baer, E., 2017: Organizing for reform: The democratic study group and the role of party factions in driving institutional change in the house of representatives.

Baker, J. D., 1973: The character of the congressional revolution of 1910. Journal of American History, 60, 679–691.

Barnes, T. D., 2014: Women’s representation and legislative committee appointments: The case of the argentine provinces. Revista Uruguaya de Ciencia Pol´ıtica, 23 (2), 135–163.

Barnes, T. D., 2016: Gendering Legislative Behavior: Institutional Constraints and Collaboration. Cambridge University Press, New York.

Binder, S. A., 1999: The dynamics of legislative gridlock, 1947–96. American Political Science Review, 93 (3), 519–533.

97 Texas Tech University, James P. Bassett, August 2020

Binder, S. A., E. D. Lawrence, and F. Maltzman, 1999: Uncovering the hidden effect of party. The Journal of Politics, 61 (3), 815–831.

Black, M., 2004: The transformation of the southern democratic party. The Journal of Politics, 66 (4), 1001–1017.

Boix, C., 1999: Setting the rules of the game: the choice of electoral systems in advanced democracies. American Political Science Review, 93 (3), 609–624.

Box-Steffensmeier, J., J. M. Ryan, and A. E. Sokhey, 2015: Examining legislative cue-taking in the us senate. Legislative Studies Quarterly, 40 (1), 13–53.

Bratton, K. A. and K. L. Haynie, 1999: Agenda setting and legislative success in state legislatures: The effects of gender and race. The Journal of Politics, 61 (3), 658–679.

C-SPAN, 2019: C-span congressional chronicle. C-SPAN.

Cameron, C. M., 2000: Veto bargaining: Presidents and the politics of negative power. Cambridge University Press.

Canon, D., G. Nelson, and C. Stewart, 1998: Historical congressional standing committees, 1st to 79th congresses, 1789-1947: House/57th to 67th congress.

Carroll, R. and H. A. Kim, 2010: Party government and the “cohesive power of public plunder”. American Journal of Political Science, 54 (1), 34–44.

Carson, J. L., M. H. Crespin, and A. J. Madonna, 2014: Procedural signaling, party loyalty, and traceability in the us house of representatives. Political Research Quarterly, 67 (4), 729–742.

Carson, J. L., G. Koger, M. J. Lebo, and E. Young, 2010: The electoral costs of party loyalty in congress. American Journal of Political Science, 54 (3), 598–616.

Clausen, A. R. and C. Wilcox, 1987: Policy partisanship in legislative leadership recruitment and behavior. Legislative Studies Quarterly, 243–263.

Clinton, J. D., 2007: Lawmaking and roll calls. The Journal of Politics, 69 (2), 457–469.

Clinton, J. D. and J. Lapinski, 2008: Laws and roll calls in the us congress, 1891–1994. Legislative Studies Quarterly, 33 (4), 511–541.

Cox, G. W., 1987: The efficient secret: The cabinet and the development of political parties in Victorian England. Cambridge University Press.

Cox, G. W., 2001: Agenda setting in the us house: A majority-party monopoly? Legislative Studies Quarterly, 185–210.

98 Texas Tech University, James P. Bassett, August 2020

Cox, G. W. and M. D. McCubbins, 1993: Legislative Leviathan. University of California Press.

Cox, G. W. and M. D. McCubbins, 2005: Setting the Agenda: Responsible Party Government in the U.S. House of Representatives. Cambridge University Press.

Cox, G. W. and K. T. Poole, 2002: On measuring partisanship in roll-call voting: The us house of representatives, 1877-1999. American Journal of Political Science, 477–489.

Cox, G. W. and W. C. Terry, 2008: Legislative productivity in the 93d–105th congresses. Legislative Studies Quarterly, 33 (4), 603–618.

Crespin, M. H. and D. Rohde, 2018: Political institutions and public choice roll-call database. Retrieved from https://ou.edu/carlalbertcenter/research/pipc-votes/.

Curry, J. M., 2015: Legislating in the Dark: Information and Power in the House of Representatives. University of Chicago Press.

Curry, J. M. and F. E. Lee, 2019: Non-party government: Bipartisan lawmaking and party power in congress. Perspectives on Politics, 17 (1), 47–65.

Davidson, R. H. and W. J. Oleszek, 1977: Congress against itself. Midland Books.

Davidson, R. H., W. J. Oleszek, and T. Kephart, 1988: One bill, many committees: Multiple referrals in the us house of representatives. Legislative Studies Quarterly, 3–28.

Deering, C. J., 2003: Ebb and flow in twentieth-century committee power. Congress Responds to the Twentieth Century, S. Ahuja and R. Dewhirst, Eds., The State University Press.

DiSalvo, D., 2012: Engines of change: party factions in American politics, 1868-2010. Oxford University Press.

Edwards III, G. C., A. Barrett, and J. Peake, 1997: The legislative impact of divided government. American journal of political science, 545–563.

Evans, C. L., 1999: Legislative structure: Rules, precedents, and jurisdictions. Legislative Studies Quarterly, 605–642.

Evans, C. L. and C. E. Grandy, 2009: The whip systems of congress. Congress reconsidered, CQ Press Washington, DC, Vol. 9.

Fechner, H., 2014: Managing political polarization in congress: A case study on the use of the . Utah L. Rev., 757.

Fenno, R. F., 1973: Congressmen in committees. Little, Brown.

99 Texas Tech University, James P. Bassett, August 2020

Finocchiaro, C. J. and D. W. Rohde, 2008: War for the floor: Partisan theory and agenda control in the us house of representatives. Legislative Studies Quarterly, 33 (1), 35–61.

Follett, M. P., 1896: The speaker of the House of Representatives. London and Bombay, Longmans, Green, and Company.

Fowler, J. H., 2006: Connecting the congress: A study of cosponsorship networks. Political Analysis.

Froman, L. A. and R. B. Ripley, 1965: Conditions for party leadership: The case of the house democrats. American Political Science Review, 59 (1), 52–63.

Grofman, B., W. Koetzle, and A. J. McGann, 2002: Congressional leadership 1965-96: A new look at the extremism versus centrality debate. Legislative Studies Quarterly.

Groseclose, T. and N. McCarty, 2001: The politics of blame: Bargaining before an audience. American Journal of Political Science, 100–119.

Harrison, R., 2004: Congress, Progressive Reform, and the New American State. Cambridge University Press.

Harward, B. M. and K. W. Moffett, 2010: The calculus of cosponsorship in the us senate. Legislative Studies Quarterly, 35 (1), 117–143.

Haynie, K. L., 2002: The color of their skin or the content of their behavior? race and perceptions of african american legislators. Legislative Studies Quarterly, 27 (2), 295–314.

Heberlig, E., M. Hetherington, and B. Larson, 2006: The price of leadership: Campaign money and the polarization of congressional parties. Journal of Politics.

Hetherington, M. J., 2009: Putting polarization in perspective. British Journal of Political Science, 39 (2), 413–448.

Jenkins, J. A., M. H. Crespin, and J. L. Carson, 2005: Parties as procedural coalitions in congress: An examination of differing career tracks. Legislative Studies Quarterly, 30 (3), 365–389.

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

Jessee, S. and N. Malhotra, 2010: Are congressional leaders middlepersons or extremists? yes. Legislative Studies Quarterly.

100 Texas Tech University, James P. Bassett, August 2020

Jessee, S. A. and S. M. Theriault, 2012: The two faces of congressional roll-call voting. Party Politics, 20 (6), 836–848.

Kirkland, J. H., 2011: The relational determinants of legislative outcomes: Strong and weak ties between legislators. The Journal of Politics, 73 (3), 887–898.

Kousser, T., J. B. Lewis, and S. E. Masket, 2007: Ideological adaptation? the survival instinct of threatened legislators. The Journal of Politics, 69 (3), 828–843.

Krehbiel, K., 1993: Where’s the party? British Journal of Political Science, 23 (2), 235–266.

Krehbiel, K., 2010: Pivotal politics: A theory of US lawmaking. University of Chicago Press.

Krehbiel, K. and J. Woon, 2005: Selection criteria for roll call votes.

Lawrence, E. D., F. Maltzman, and S. S. Smith, 2006: Who wins? party effects in legislative voting. Legislative Studies Quarterly, 31 (1), 33–69.

Lawrence, E. D., F. Maltzman, and P. J. Wahlbeck, 2001: The politics of speaker cannon’s committee assignments. American Journal of Political Science, 551–562.

Lewis, J. B., H. Rosenthal, A. Boche, A. Rudkin, and L. Sonnet, 2019: Voteview: Congressional roll-call votes database. URL http://voteview.com. Lynch, M. S. and A. J. Madonna, 2013: Viva voce: Implications from the disappearing voice vote, 1865–1996. Social Science Quarterly, 94 (2), 530–550.

Lynch, M. S., A. J. Madonna, and J. M. Roberts, 2016: The cost of majority-party bias: Amending activity under structured rules. Legislative Studies Quarterly, 41 (3), 633–655.

MacNeil, N. and R. A. Baker, 2013: The American Senate: an insider’s history. Oxford University Press.

Matthews, D. R., 1959: The folkways of the united states senate: Conformity to group norms and legislative effectiveness. American Political Science Review, 53 (4), 1064–1089.

Mayhew, D. R., 1974: Congress: The electoral connection, Vol. 26. Yale University Press.

Mayhew, D. R., 1991: Divided we govern. Yale University New Haven, CT.

Minozzi, W. and C. Volden, 2013: Who heeds the call of the party in congress? The Journal of Politics, 75 (3), 787–802.

101 Texas Tech University, James P. Bassett, August 2020

Miquel, G. P. I. and J. M. Snyder Jr, 2006: Legislative effectiveness and legislative careers. Legislative Studies Quarterly, 31 (3), 347–381.

Monroe, N. W. and G. Robinson, 2008: Do restrictive rules produce nonmedian outcomes? a theory with evidence from the 101st- 108th congresses. The Journal of Politics, 70 (1), 217–231.

Montgomery, J. M. and B. Nyhan, 2017: The effects of congressional staff networks in the us house of representatives. The Journal of Politics, 79 (3), 745–761.

Moore, M. K. and S. Thomas, 1991: Explaining legislative success in the us senate: The role of the majority and minority parties. Western Political Quarterly, 44 (4), 959–970.

Oppenheimer, B. I., 1977: The rules committee: New arm of leadership in a decentralized house. Congress Reconsidered, Praeger New York, Vol. 1, 96–116.

Oppenheimer, B. I., 1981: The changing relationship between house leadership and the committee on rules. Understanding Congressional Leadership, 207–225.

Orey, B. D., W. Smooth, K. S. Adams, and K. Harris-Clark, 2007: Race and gender matter: Refining models of legislative policy making in state legislatures. Journal of Women, Politics & Policy, 28 (3-4), 97–119.

Peabody, R. L., 1976: Leadership in Congress: Stability, Succession, and Change. Little Brown.

Peters, R. M., 1990: The American Speakership: The Office in Historical Perspective. The Johns Hopkins University Press.

Pinney, N. and G. Serra, 1999: The congressional black caucus and vote cohesion: Placing the caucus within house voting patterns. Political Research Quarterly, 52 (3), 583–607.

Polsby, N. W., 1968: The institutionalization of the us house of representatives. American political science review, 62 (1), 144–168.

Polsby, N. W., M. Gallaher, and B. S. Rundquist, 1969: The growth of the seniority system in the us house of representatives. American Political Science Review, 63 (3), 787–807.

Polser, B. and C. Rhodes, 1997: Pre-leadership signaling in the u.s. house. Legislative Studies Quarterly.

Poole, K. T. and H. L. Rosenthal, 2011: Ideology and congress, Vol. 1. Transaction Publishers.

102 Texas Tech University, James P. Bassett, August 2020

Richman, J., 2015: The electoral costs of party agenda setting: why the hastert rule leads to defeat. The Journal of Politics, 77 (4), 1129–1141. Ringe, N. and S. L. Wilson, 2016: Pinpointing the powerful: Covoting network centrality as a measure of political influence. Legislative Studies Quarterly. Rivers, D. and N. L. Rose, 1985: Passing the president’s program: Public opinion and presidential influence in congress. American Journal of Political Science, 183–196. Roberts, J. M., 2007: The statistical analysis of roll-call data: A cautionary tale. Legislative Studies Quarterly, 32 (3), 341–360. Roberts, J. M. and S. S. Smith, 2003: Procedural contexts, party strategy, and conditional party voting in the us house of representatives, 1971–2000. American Journal of Political Science, 47 (2), 305–317. Rogowski, J. C. and B. Sinclair, 2012: Estimating the causal effects of social interaction with endogenous networks. Political Analysis, 20 (3), 316–328. Rohde, D. W., 1991: Parties and Leaders in the Postreform House. University of Chicago Press, Chicago. Rubin, R. B., 2013: Organizing for insurgency: Intraparty organization and the development of the house insurgency, 1908–1910. Studies in American Political Development, 27 (2), 86–110. Sarasohn, D., 1979: The insurgent republicans: Insurgent image and republican reality. Social Science History, 3 (3-4), 245–261. Schickler, E., 2001: Disjointed pluralism: Institutional innovation and the development of the US Congress, Vol. 124. Princeton University Press. Schwindt-Bayer, L. A., 2006: Still supermadres? gender and the policy priorities of latin american legislators. American Journal of Political Science, 50 (3), 570–585. Sinclair, B., 1989: The transformation of the US Senate. Johns Hopkins Univ Pr. Sinclair, B., 2003: Full circle? congressional party leadership during the twentieth century. Congress Responds to the Twentieth Century, S. Ahuja and R. Dewhirst, Eds., The Ohio State University Press. Smith, S. S. and M. Flathman, 1989: Managing the senate floor: Complex unanimous consent agreements since the 1950s. Legislative Studies Quarterly, 349–374. Snyder Jr, J. M. and T. Groseclose, 2000: Estimating party influence in congressional roll-call voting. American Journal of Political Science, 193–211.

103 Texas Tech University, James P. Bassett, August 2020

Svolik, M. W., 2012: The politics of authoritarian rule. Cambridge University Press.

Takeda, O., 2015: A forgotten minority? a content analysis of asian pacific americans in introductory american government textbooks. PS: Political Science & Politics, 48 (3), 430–439.

Theriault, S. M., 2006: Procedural polarization in the us congress. annual meeting of the Midwest Political Science Association.

Theriault, S. M., 2008: Party polarization in congress. Cambridge University Press.

Theriault, S. M., 2013: The Gingrich senators: The roots of partisan warfare in Congress. Oxford University Press.

Theriault, S. M. and D. W. Rohde, 2011: The gingrich senators and party polarization in the us senate. The Journal of Politics, 73 (4), 1011–1024.

Victor, J. N. and N. Ringe, 2009: The social utility of informal institutions: Caucuses as networks in the 110th us house of representatives. American Politics Research, 37 (5), 742–766.

Volden, C. and A. E. Wiseman, 2014: Legislative Effectiveness in the United States Congress: The Lawmakers. Cambridge University Press.

Volden, C. and A. E. Wiseman, 2018: Legislative effectiveness in the united states senate. The Journal of Politics, 80 (2), 731–735.

Volden, C., A. E. Wiseman, and D. E. Wittmer, 2013: When are women more effective lawmakers than men? American Journal of Political Science, 57 (2), 326–341.

Weissert, C. S., 1991a: Issue salience and state legislative effectiveness. Legislative Studies Quarterly, 509–520.

Weissert, C. S., 1991b: Policy entrepreneurs, policy opportunists, and legislative effectiveness. American Politics Quarterly, 19 (2), 262–274.

104 Texas Tech University, James P. Bassett, August 2020

APPENDIX

Table A1. Context and Setting, No Southern Democrats

Dependent variable: Centrality Connctedness LES (1) (2) (3) Vote % 0.0001∗ 0.0004∗∗∗ 0.010∗∗∗ (0.0001) (0.0001) (0.002)

Ideol. Extremity 0.316∗∗∗ −0.001 −0.176 (0.012) (0.007) (0.208)

State Leg. Prof. 0.007 −0.003 0.130 (0.007) (0.006) (0.207)

Majority (Divided) 0.031 −0.011 0.874∗∗ (0.023) (0.017) (0.419)

Same Party as Pres. −0.051∗∗ −0.027 −0.255 (0.023) (0.017) (0.412)

Majority * Same Party 0.076 0.047 0.500 (0.046) (0.034) (0.826)

Constant 0.285∗∗∗ 0.244∗∗∗ −0.004 (0.024) (0.018) (0.437)

Observations 5,630 5,509 5,630 R2 0.802 0.424 0.133 Adjusted R2 0.801 0.421 0.130 Residual Std. Error 0.051 (df = 5608) 0.053 (df = 5487) 1.488 (df = 5608) F Statistic 1,078.458∗∗∗ (df = 21; 5608) 192.056∗∗∗ (df = 21; 5487) 40.941∗∗∗ (df = 21; 5608) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

105 Texas Tech University, James P. Bassett, August 2020

Figure A.1. Scatterplot of LES and Connectedness

106 Texas Tech University, James P. Bassett, August 2020

Table A2. OLS Results: Southern Democrats (Post-1994)

Dependent variable: Centrality Connctedness LES (1) (2) (3) Vote −0.0001 0.0002∗ 0.008∗∗∗ (0.0001) (0.0001) (0.003)

Ideol. Extremity 0.385∗∗∗ 0.035∗∗∗ 0.764∗∗ (0.017) (0.010) (0.308)

Southern Dem. −0.006 −0.021∗∗∗ −0.175 (0.005) (0.005) (0.169)

State Leg. Prof. 0.007 −0.005 −0.061 (0.008) (0.009) (0.284)

Post-1994 0.374∗∗∗ −0.085∗∗∗ −0.069 (0.005) (0.006) (0.214)

Post-1994 * Southern Dem. −0.026 −0.007 0.049 (0.007) (0.006) (0.162)

Constant 0.311∗∗∗ 0.248∗∗∗ 1.026∗∗∗ (0.006) (0.007) (0.218)

Observations 4,267 3,703 4,267 R2 0.833 0.455 0.089 Adjusted R2 0.832 0.452 0.084 Residual Std. Error 0.049 (df = 4244) 0.055 (df = 3682) 1.661 (df = 4244) F Statistic 959.779∗∗∗ (df = 22; 4244) 153.973∗∗∗ (df = 20; 3682) 18.771∗∗∗ (df = 22; 4244) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

107 Texas Tech University, James P. Bassett, August 2020

Table A3. Hypotheses 1A and 1B: All Covariates

Dependent variable: Centrality (1) (2) Post-1910 −0.066∗∗∗ −0.060∗∗∗ (0.006) (0.007)

Republican 0.016∗∗∗ 0.017∗∗∗ (0.003) (0.003)

Insurgent −0.110∗∗∗ −0.075∗∗∗ (0.013) (0.009)

Comm. Chair 0.012∗∗ 0.007 (0.005) (0.008)

Extremity 0.283∗∗∗ 0.284∗∗∗ (0.018) (0.018)

Power Comm. 0.003 0.005 (0.005) (0.005)

Power Chair −0.001 0.002 (0.021) (0.022)

Majority −0.006∗∗∗ −0.006∗∗∗ (0.001) (0.001)

Time Since Treatment 0.032∗∗∗ 0.033∗∗∗ (0.002) (0.002)

Post-1910 * Insurgent 0.080∗∗∗ (0.013)

Post-1910 * Comm. Chair 0.008 (0.010)

Constant 2.548∗∗∗ 2.583∗∗∗ (0.096) (0.097)

Observations 4,509 4,509 R2 0.441 0.437 Adjusted R2 0.439 0.436 Residual Std. Error (df = 4497) 0.109 0.109 F Statistic (df = 11; 4497) 322.301∗∗∗ 317.634∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

108 Texas Tech University, James P. Bassett, August 2020

Table A4. Hypotheses 2A and 2B: All Covariates

Dependent variable: Centrality (1) (2) Post-1977 −0.035∗∗∗ −0.034∗∗∗ (0.003) (0.003)

Majority/Democrat 0.068∗∗∗ 0.068∗∗∗ (0.003) (0.003)

Comm. Chair 0.008∗∗ 0.009∗∗ (0.004) (0.004)

Power Comm. 0.0001 0.0001 (0.003) (0.003)

Budget Comm. −0.0001 −0.00002 (0.003) (0.003)

Subcomm. Chair −0.008∗∗ −0.003 (0.003) (0.002)

Extremity 0.242∗∗∗ 0.242∗∗∗ (0.012) (0.012)

Southern Democrat −0.026∗∗∗ −0.026∗∗∗ (0.004) (0.004)

Maj. Leadership 0.023∗∗∗ 0.005 (0.005) (0.009)

Time Since Treatment 0.005∗∗ 0.005∗∗ (0.002) (0.002)

Post-1977 * Subcomm. Chair 0.006∗ (0.003)

Post-1977 * Maj. Leadership 0.025∗∗ (0.010)

Constant −1.221∗∗∗ −1.209∗∗∗ (0.185) (0.185)

Observations 3,060 3,060 R2 0.672 0.672 Adjusted R2 0.671 0.671 Residual Std. Error (df = 3047) 0.043 0.043 F Statistic (df = 12; 3047) 519.811∗∗∗ 519.794∗∗∗

∗ ∗∗ ∗∗∗ Note: 109 p<0.1; p<0.05; p<0.01 Texas Tech University, James P. Bassett, August 2020

Table A5. 1977 Case, 93rd-108th Congress

Dependent variable: Centrality (1) (2) Post-1977 −0.034∗∗∗ −0.031∗∗∗ (0.003) (0.003)

Majority/Democrat 0.076∗∗∗ 0.076∗∗∗ (0.002) (0.002)

Comm. Chair 0.010∗∗ 0.010∗∗ (0.004) (0.004)

Power Comm. 0.006∗∗ 0.006∗∗ (0.003) (0.003)

Budget Comm. −0.001 −0.001 (0.003) (0.003)

Subcomm. Chair −0.014∗∗∗ −0.005∗∗ (0.004) (0.003)

Extremity 0.342∗∗∗ 0.342∗∗∗ (0.015) (0.015)

Southern Democrat −0.015∗∗∗ −0.016∗∗∗ (0.004) (0.004)

Maj. Leadership 0.032∗∗∗ −0.007 (0.008) (0.008)

Time Since Treatment 0.004 0.004∗ (0.002) (0.002)

Post-1977 * Subcomm. Chair 0.010∗∗ (0.004)

Post-1977 * Maj. Leadership 0.042∗∗∗ (0.011)

Constant −1.140∗∗∗ −1.117∗∗∗ (0.208) (0.208)

Observations 7,885 7,885 R2 0.754 0.754 Adjusted R2 0.753 0.753 Residual Std. Error (df = 7872) 0.063 0.063 F Statistic (df = 12; 7872) 2,008.834∗∗∗ 2,008.903∗∗∗

∗ ∗∗ ∗∗∗ Note: 110 p<0.1; p<0.05; p<0.01 Texas Tech University, James P. Bassett, August 2020

Table A6. 1910 Case, Omitting Time Since Treatment

Dependent variable: Centrality (1) (2) Post-1910 −0.094∗∗∗ −0.100∗∗∗ (0.006) (0.006)

Republican 0.023∗∗∗ 0.022∗∗∗ (0.003) (0.003)

Insurgent −0.088∗∗∗ −0.125∗∗∗ (0.009) (0.014)

Comm. Chair 0.006 0.012∗∗ (0.009) (0.005)

Extremity 0.280∗∗∗ 0.278∗∗∗ (0.018) (0.018)

Power Comm. 0.006 0.004 (0.005) (0.005)

Power Chair 0.002 −0.001 (0.023) (0.022)

Majority −0.005∗∗∗ −0.006∗∗∗ (0.001) (0.001)

Post-1910 * Comm. Chair 0.011 (0.011)

Post-1910 * Insurgent 0.083∗∗∗ (0.014)

Constant 1.321∗∗∗ 1.291∗∗∗ (0.057) (0.057)

Observations 4,509 4,509 R2 0.410 0.414 Adjusted R2 0.408 0.412 Residual Std. Error (df = 4498) 0.112 0.111 F Statistic (df = 10; 4498) 312.162∗∗∗ 317.230∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

111 Texas Tech University, James P. Bassett, August 2020

Table A7. 1977 Case, Omitting Time Since Treatment

Dependent variable: Centrality (1) (2) Post-1977 −0.042∗∗∗ −0.040∗∗∗ (0.003) (0.002)

Majority/Democrat 0.068∗∗∗ 0.068∗∗∗ (0.003) (0.003)

Comm. Chair 0.008∗∗ 0.009∗∗ (0.004) (0.004)

Power Comm. 0.00003 0.00002 (0.003) (0.003)

Budget Comm. −0.0001 −0.00003 (0.003) (0.003)

Subcomm. Chair −0.008∗∗ −0.003 (0.003) (0.002)

Extremity 0.242∗∗∗ 0.242∗∗∗ (0.012) (0.012)

Southern Democrat −0.026∗∗∗ −0.026∗∗∗ (0.004) (0.004)

Maj. Leadership 0.023∗∗∗ 0.005 (0.005) (0.009)

Post-1977 * Subcomm. Chair 0.006∗ (0.003)

Post-1977 * Maj. Leadership 0.025∗∗∗ (0.010)

Constant −1.648∗∗∗ −1.648∗∗∗ (0.072) (0.072)

Observations 3,060 3,060 R2 0.672 0.672 Adjusted R2 0.670 0.670 Residual Std. Error (df = 3048) 0.043 0.043 F Statistic (df = 11; 3048) 566.523∗∗∗ 566.463∗∗∗ Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

112