Bipartisanship and Bicameralism: a New Inside View of Congressional Committees

Bipartisanship and Bicameralism: A New Inside View of Congressional Committees

Pamela Ban∗ Harvard University October 2017

Abstract Polarization in Congress has reached record highs in recent times, leading to concerns about gridlock and partisan outcomes. I show that even in a polarized environment, committees achieve consensus in the face of partisan conflict and can act as a cue for bipartisanship. Using a new text dataset of House committee hearings and committee reports from 2003-2014, I propose and construct bill-level measures of committee-stage disagreement between Republican and Democratic committee members. Results show that regardless of the level of committee-stage disagreement between the two parties, minority committee members still vote against their party on the floor, in their committee’s jurisdiction, to achieve high levels of bipartisanship among committee members. A difference-in-differences design suggests that this behavior comes from the sorting mechanism in the committee assignment process; these members were inclined to vote against their party at the same rate in the committee’s jurisdiction even when they are not on committee.

∗Ph.D. Candidate, Political Economy and Government. [email protected]. 1 Introduction

Congressional committees are central legislative units of Congress. Scholars have noted the congressional committee’s strategic policy-making position, viewing committees as gatekeep- ers of their jurisdictions, policy experts, and agenda controllers in their policy area (Shepsle and Weingast, 1987; Smith and Deering, 1984; Fenno, 1973; Weingast and Marshall, 1988; Cox and McCubbins, 1993). Given the importance of congressional committees to Congress, understanding the dynamics of committees is crucial to understanding congressional politics. With the recent focus on high levels of party conflict in Congress, of particular interest is the effect of polarization on the legislative process and its impact on congressional output. As committees are gate-keepers to the floor, polarization in the committee stage has the potential to keep bills from reaching the floor or result in split committee votes on the bill’s passage. The literature has theorized that members on a committee have incentives to cooperate with each other even in the face of internal disagreements and differing preferences. Richard Fenno coins this concept as “committee integration,” describing it as “the degree to which a committee is able to minimize conflict among its roles and subgroups, by heading off or resolving the conflicts that arise” (Fenno, 1962). The literature has argued that the committee system can resolve conflicts by exchanging support through logrolling and vote trading – more generally, in a “gains from exchange institutional arrangement” (Weingast and Marshall, 1988; Shepsle and Weingast, 1987; Fiorina, 1987). Committees may also vote together to signal quality, or because they have important information that the wider chamber may not have (Gilligan and Krehbiel, 1989). However, while committee members do tend to seek out committees with jurisdictions that align with their interests, committee members can still have differing positions and heterogenous preferences within the committee. This is especially true along party lines. Does the incentive to cooperate lead committees to resolve high levels of partisan conflict on a bill during the committee stage, resulting in bipartisan support from committee members

1 on the floor? In today’s highly polarized Congress, if we find that there is a high level of bipartisanship support among committee members during the floor stage, even though there was partisan conflict within committee during the earlier committee stage on the same bill, then the institution of the committee system may be a vehicle for bipartisanship. On the other hand, if we find that partisan conflict on a bill in the committee stage continues to exist in committee members’ voting on the bill on the floor, then we have evidence that the majority party excludes the minority party (Cox and McCubbins, 1993) and that the committee system simply passes along existing party disagreements. Are committees successful at achieving consensus in the presence of partisan conflict? To answer this, a precise measure of disagreement between party members in the committee stage is needed. While scholars have long studied partisan politics on the congressional floor, pre-floor activity in Congress has received scant attention.1 This is largely due to the data limitations surrounding congressional committee activity in the committee stage. Empirical work on Congress has mainly been restricted to activities and outcomes on the floor (relying on measures built upon roll call votes or congressional floor speech) or at the beginning of the legislative process (relying on measures built upon bill sponsorship data). Systematically measuring party disagreement in the committee stage – taking place in between these two points in time – is difficult because committee activities do not result in clean numerical data that reveal positions on bills under consideration. To tackle this, I propose a new approach that uses new text data to measure disagreement on bills inside committees. I construct two text-based measures of bill-level disagreement among committee members in the committee stage using the speech of committee members in committee hearings and the amount of formal disagreement presented by minority committee members in committee reports. To do this, I collect and use a new text dataset comprised

1The congressional literature has measured and documented the divide between parties on the ﬂoor, focusing on the growing homogeneity in policy positions within each party and the growing distance between the two parties. This growth in polarization has been measured in numerous ways, including using the percentage of party votes (Bond and Fleisher, 2000; Stonecash, Brewer and Mariani, 2003), interest group ratings (Stonecash, Brewer and Mariani, 2003), member NOMINATE scores (Poole and Rosenthal, 1997; Jacobson, 2000), and party unity scores (Bond and Fleisher, 2000; Stonecash, Brewer and Mariani, 2003).

2 of committee reports and committee hearing transcripts for the 108th-113th Congresses, linked to individual bills. I do this for five committees that have jurisdiction over important areas of domestic policy: Education and Labor, Financial Services, Energy and Commerce, Transportation and Infrastructure, and Veterans’ Affairs. I find that the amount of committee-stage disagreement over a bill does not have a statistically significant effect on that committee’s bipartisanship on the bill’s passage vote on the floor. However, there are substantial levels of bipartisanship on the floor for these five committees even in the face of committee-stage disagreement on bills. In order for these levels of bipartisanship among committee members to exist on the floor, there must be some members of the minority party voting with the majority party, or vice versa. While this is not surprising for bipartisan bills (when both parties vote in the same direction), party- line bills are an interesting case. Is there something inherent in committee service – and the benefits or incentives that come along with it – that leads certain committee members to vote against their own party? Or is it because the members on these committees were already inclined to vote against their party on legislation in their committee’s jurisdiction, perhaps due to district preferences, even when they are not on committee? Previous research has suggested three possible reasons for why committee members (minority or majority party) may vote differently on issues in that committee’s jurisdiction compared to non-committee members. Scholars have shown that committee members receive more interest group donations and lobbying efforts from interest groups in the committee’s jurisdiction than non-committee members (Romer and Snyder, 1994; Grimmer and Powell, 2016; Hojnacki and Kimball, 1998, 1999; Kollman, 1997; Hall and Wayman, 1990). Nor- matively, policymakers or observers of Congress may be concerned that this extra attention from interest groups affects committee members’ voting behavior. Furthermore, two traditional theories of committees suggest that committee members may vote differently from non-committee members of their party. First, if we think that committee service comes with more power over what goes into legislation, committee members may be in a better

3 position to extract specific benefits for themselves and their districts, versus when they are not on committee (Weingast and Marshall, 1988; Shepsle and Weingast, 1987) This distributive benefits story also relates to the “preference outlier” theory of committees.2 Second, the informational theory of committees, e.g. Gilligan and Krehbiel (1989), suggests that committee members have more information on the bill and its relevant policies compared to non-committee members – in other words, committee members may know more about the true benefits and implications of the bill than non-committee members, and so may vote differently than (co-partisan) non-committee members because of this information. The second part of this paper tests whether committee service impacts members’ frequencies of voting against their party and exploits the panel structure of my data using a within-member design. I find that on average, minority committee members tend to vote more frequently against their party than their party’s non-committee members on party-line bills coming from their committee, compared to majority committee members. This behavior is more pronounced during the “textbook Congress” era (postwar period until 1969) when committees were stronger, compared to the more recent 108th-113th time period. However, this behavior is not driven by committee membership itself. Results from a difference-in- differences design show that the “treatment” of committee membership does not have a statistically significant effect on the frequency of a minority member (or majority member) voting against their party. When looking within member-committee pairs, members tend to vote against their party at the same rate, for a given committee’s jurisdiction, whether or not they are on the committee. In other words, members’ roll call voting behaviors are “sticky” when it comes to a change in committee service, and the differences in voting behavior we see between committee and non-committee members of the same party are a result of members sorting onto committees.

2Given the possibility of committee members extracting distributive benefits in the policy area of the committee, the distributive benefits theory implies that committees will be unrepresentative of the chamber and be comprised of preference outliers. Under this view, if committee members are preference outliers in the committee’s jurisdictions, then they may vote differently than non-committee members on bills coming from the committee.

4 The broader implications of my findings relate to the existing theories on interest groups, committees, and the bicameral process of Congress. Even though committee service has been shown to come along with an increase in targeted interest group money and lobbying, this does not seem to change a member’s voting behavior, at least not enough to make the member vote in the opposite direction from his or her party. Further, any distributive benefits or information that committee members may gain compared to non-committee members also do not change this voting behavior. That does not go to say that committee members do not receive distributive benefits or more information about the bill, as argued by Weingast and Marshall (1988), Shepsle and Weingast (1987), and Gilligan and Krehbiel (1989), but anything they do receive as a result of committee membership does not change the rate of that member voting in the opposite direction of their party on the floor. More tentatively, I argue that the high levels of bipartisanship achieved among committee members is useful for the committee in a bicameral context. A committee in the House does not simply want the House to pass its bill; it also needs the Senate to pass its bill on the way to becoming law.

2 Methods and Data

To determine how successful committees are at achieving consensus in the presence of partisan conflict, a precise measure of disagreement in the committee stage is needed. However, there is a lack of numerical data on committee stage activities. Unlike the floor stage, there is no systematic access to roll call-type votes in the committee stage, making it difficult for empirical scholars to analyze committee-stage activities. However, there are two text data sources that capture committee members’ preferences in the committee stage: transcripts of committee hearings held on a bill and committee reports on bills reported to the floor. I use the speech of committee members in committee hearings and the length of the “Minority Views” section in committee reports to measure

5 intra-committee party disagreement on a bill while the bill is in the committee stage. In this section, I detail these two approaches in turn.

2.1 Committee Hearings

A congressional hearing is a meeting or session of a committee of Congress, and are held “to obtain information and opinions on proposed legislation, conduct an investigation, or eval- uate/oversee the activities of a government department of the implementation of a Federal law...may also be purely exploratory in nature, providing testimony and data about topics of current interest.” Figure 1 shows the first page of an example hearing transcript for a hearing held by the Committee on Financial Services. The U.S. Government Publishing Office (GPO) provides transcripts of congressional hearings spanning the 99th Congress to the 114th Congress. Not all hearings are available for this time period, as the availability depends on the specific committee (and are largely published anywhere from two months to two years after the hearing date), but complete sets of hearing transcripts (after optical character recognition software) are available for committees starting in the 104th Congress, with a few exceptions. Committee hearings that are held to gather information, testimony, and opinions on proposed legislation contain useful speech that reveal committee members’ views on bills under consideration by the committee. As committee members declare their views, debate their opinions, and question witnesses about the proposed legislation in hearings, we can observe their opinions and concerns about the bill under consideration. For this reason, I use committee hearings that are specifically held on proposed legislation. This eliminates from my dataset those hearings that are held in an exploratory nature (before a bill is brought to committee), to conduct an investigation, or to oversee a government agency or department.3 Currently, I match committee hearings to bills under consideration by searching for bill names in the titles of committee hearings – e.g. the title of the hearing in

3Exploratory hearings about policy matters even before a speciﬁc bill is brought to committee may also reveal preferences and positions of committee members on that policy matter that could be linked to future

6 Figure 1 – Committee Hearing Example. This is the ﬁrst page of the transcript for a hearing held by the House Committee on Financial Services on H.R. 698 in the 110th Congress.

7 Figure 1 clearly references H.R. 698. Committee hearing titles also commonly reference the bill name in words – e.g. the Committee on Energy and Commerce held a hearing entitled “An Examination of the Older Americans Act” during the 109th Congress, and a search of

the Congress.gov database shows that the Older Americans Act is H.R. 6197 (in the 109th Congress).4 From the GPO, I scrape and collect transcripts of all the published congressional hearings and select the hearings about proposed legislation through the above procedure. Committee hearings often include testimony from outside witnesses; as my goal in this paper is to use these transcripts is to extract committee members’ positions, I remove speech from witnesses so that the resulting text only includes speech made by committee members. The text is

then pre-processed using the following steps: (1) remove stop-words in the nltk python package, (2) stem using the WordNet lemmatizer, (2) remove proper nouns and technical language of procedure, (3) generate unigrams and bigrams. I focus on committee hearings for the Education and Labor (Education and the Workforce), Energy and Commerce, Financial Services, Transportation and Infrastructure, and Veterans’ Affairs committees in the 108th- 113th committees.5 To measure the positions of Republican and Democratic committee members on a bill (and hence to measure the difference between those positions) as revealed by speech during the committee stage, I use the Wordscores method developed by Michael Laver, Kenneth Benoit, and John Garry, which uses texts with exogenously defined ideological positions (“reference texts”) to estimate the ideological positions of texts of which the positions are unknown (Laver, Benoit and Garry, 2003). Wordscores assigns an ideological score to each word in the reference text, creating a dictionary – the dictionary score is a weighted average

bills brought to committee, but for the moment, I only use hearings that are explicitly for proposed bills that committees are charged with considering. 4This current procedure misses hearings that are about proposed legislation but do not explicitly reference the bill in the title – future plans include a more refined way that also searches the introduction of the hearing transcript to see if there is a mention of a bill name. 5These committees are used because they have clearly defined policy jurisdictions with the cleanest topic filtering of floor speeches.

8 of the ideological scores assigned to the reference texts, where the weights are the relative frequencies of the words across the reference texts. This dictionary is then used to score the texts for which the researcher wishes to calculate the ideological positions. The set of un-scored committee speeches for a bill is comprised of the Democratic committee members’ speeches given in committee hearing(s) on that bill and the Republican committee members’ speeches given in committee hearing(s) on that bill. For each bill in each committee of referral, all of the speeches given by Democratic committee members are appended into a single text, and all of the speeches given by Republican committee members are appended into a single text, so that each bill will be linked to two sets of speeches on that bill, by party. In order to calculate the distance between Republicans and Democrats as revealed by speech under a supervised approach, I must use reference texts giving a known position of Republicans and Democrats. I take the floor speech of non-committee Republican and Democrats about issues in a given committee’s jurisdiction to use as reference texts. Cur- rently, I use the process below to filter the floor speeches: Bonica (2016) uses a partially labeled dirichlet allocation (PLDA) model from Ramage et al. (2009) to identify topics for every floor speech, using Congressional Research Service (CRS) tags on bills that are mentioned in the floor speeches. However, not every floor speech will have a bill mentioned, and thus not every floor speech will have an issue label assigned to it. The training set of the PLDA model is thus the set of floor speeches that do mention a bill, and can be linked to that bills’ issue labels. Bonica (2016) uses this training set and fits the PLDA model using the Stanford Topic Model Toolkit (Ramage et al., 2009). I use the identified topics from Bonica (2016) and map House floor speeches to committee jurisdictions using the top-weighted topic weight. The idea is that the comparable set of floor speeches for the Committee on Financial Services is the set of floor speeches on topics in the jurisdiction of the Committee on Financial Services. This set of floor speeches is identified for every committee of interest by using the topic labels from Bonica (2016) for topics related to that

9 committee. For instance, every floor speech whose highest weight topic label is “Education” or “Higher Education” is combined into a text that serves as the related floor speech to be used to compare with the speech from the Education and Labor Committee, every floor speech whose highest weight topic label is “Financial Services” will make up the related floor speech to be used to compare with the speech from the Financial Services committee, and so on.6 For each congressional session and committee, I append all the Republican non-committee members’ floor speeches on that committee’s jurisdiction into a single text, and append all the Democratic non-committee members’ floor speeches on that committee’s jurisdiction into a single text. The process thus results in a set of Republican floor speeches and Democratic floor speeches for each committee jurisdiction, in each congressional session. For a given bill and committee of referral, I proceed using the following Wordscores estimation procedure, which is carried out to measure the distance between Republicans and Democrats committee members – as revealed by speech – during the committee stage

for that bill: Let Fpr be the relative frequency of each phrase p used in reference text r.

Fpr Then, the probability that seeing phrase p means we are reading text r is Ppr = P . r Fpr For each phrase p, the expected position on committee dimension d when seeing phrase p P is Sp,d = r(Ppr · Ard), where Ard is the reference text’s a priori position. For the two a priori positions, I use a weighted mean of DW-NOMINATE scores of non-committee Democrats and non-committee Republicans, weighted by the length of floor speech given by a legislator on that committee jurisdiction (in other words, a legislator’s DW-NOMINATE score is given more weight in the weighted mean as the legislator talks more on the floor on the committee jurisdiction, relative to the length of other legislators’ floor speeches on the committee jurisdiction).

6Indeed, every floor speech may fall into more than one topic, with varying degrees. A speech on energy use in farms, for instance, is likely to be labeled by the PLDA model as both “Agriculture” and “Energy”. For simplification, I only consider one topic per floor speech, using the topic that received the highest weight from the model.

10 The committee’s set of committee speeches by Republicans and Democrats on the bill is then scored using these two reference texts. Compute Fp,Cparty, the relative frequency of each phrase in each committee’s party text. Then, the score of a committee’s party text on the bill is: X SCparty,d = (Fp,Cparty · Sp,d). p These committee party text scores are next transformed in order to give these scores the same dispersion as the reference text scores – in other words, to give the committee party text scores the same variance as the original reference scores (Lowe, 2008). The mean and relative positions of committee text scores are preserved, and the variance is set to equal the variance of the ﬂoor texts:

∗ SDrd SCparty,d = (SCparty,d − SCparty,d¯ ) + SCparty,d¯ SDCparty,d

where SCparty,d¯ is the average score of the two committee party texts and SD is standard deviation. Finally, I construct the variable Committee-Stage Bill Disagreement to equal the distance between Republican and Democratic committee members on the bill as a percentage of the distance between Republican and Democratic non-committee members in that committee’s jurisdiction. While there does not exist consistent vote data in the committee stage, as discussed previously, there are some bills for which the committee report vote – that is, the vote held among committee members on whether or not to report the bill to the ﬂoor – is available and published. These published report votes should not be used directly as a measure of disagreement, due to possible selection bias, but they can provide a rough validation of the text-based committee hearings measure. When correlating Committee-Stage Bill Disagree- ment with the percentage of committee members voting against reporting the bill for 22 available committee report votes, the correlation is 0.83.

11 2.2 Committee Reports

While the measure Committee-Stage Bill Disagreement involves using theWordscores method, which is sensitive to the choices made in selecting the reference text, a simpler, more straightforward measure exists in another text source that captures the committee stage: committee reports. Committee reports accompany legislation that a committee reports to the ﬂoor. In each report, the majority party presents why they are reporting the legislation and why they are recommended that the bill passes. Of particular interest is the “Minority Views” section that the minority committee members (led by the ranking member on the committee) can include at the end of each committee report.7 The minority views section gives the minority party on the committee a formal opportunity to detail what parts of the bill they disagree with and why they disagree – in other words, this is the dissenting opinion of the minority committee members. Thus, this minority views section sheds light on how much disagreement the minority party had with the bill under consideration. The more grievances the minority party has, the longer we would expect the minority views section to be in order to discuss all those grievances. A straightforward measure of the amount of disagreement, then, is the length of the minority views section, which I use as an alternate, simpler measure of the amount of committee-stage disagreement between the two parties’ committee members. The idea is that the longer the dissent, the more disagreement the minority committee members had with the bill being reported by the majority committee members. One main drawback of this measure is that it assumes content is consistent in strength of dissent – however, a paragraph’s length of disagreement could be stronger or weaker disagreement, in the same space. When using the same 22 available committee report votes for a validation, though, the correlation between the committee report measure and the percentage of committee

7In some committees in some congressional sessions, this is called “Dissenting Views” instead of “Minority Views.”

12 members voting against reporting the bill is 0.71. While this use of committee reports is a more rudimentary measure of the amount of committee-stage disagreement on a bill, it can serve as a robustness check to results from the committee hearings based speech measure.

3 Results

3.1 Committee-Stage Disagreement and the Bipartisanship of Com-

mittees on the Floor

The distribution Committee-Stage Bill Disagreement as measured from committee hearings described in the preceding section is shown in Figure 2, and is shown for each committee of interest in Figure 3. From these figures, we can see that while there are many bills that have low levels of polarization while in committee, there are bills for which Democrats and Republicans are divided on in committee. From Figure 3, we can see that out of the five committees, the Committee on Veterans’ Affairs has the widest ranging levels of committee- stage disagreement on bills, with Committee-Stage Bill Disagreement ranging from 0% to 60% of the distance between the two parties on the floor. These distributions reveal varying levels of disagreement between the parties’ committee members over the bills that the committee considers. What is the effect of this committee- stage polarization? Does this polarization carry over into the floor voting behavior of the committee members on the bill? We may think that the level of polarization among committee members on a bill during the committee stage may be related to how divided those committee members would be on the passage vote of the bill on the floor. To investigate whether this is the case, I first explore the relationship between Committee-Stage Bill Disagreement and an outcome variable that captures the amount of bipartisanship on bill passage votes among committee members. Table 1 presents summary statistics on the levels of passage percentage (percentage voting “yea” together) by committee members, defined as the members of the committee where the

13 Figure 2 – Distribution of Committee-Stage Bill Disagreement

200 Count

100

0 0 20 40 60 Committee−Stage Bill Disagreement (%)

Figure 3 – Distribution of Committee-Stage Bill Disagreement, by Committee

Energy and Commerce Education and Labor Financial Services

30 40 60

20 30

Count Count Count 20

10 20 10

0 0 0 0 20 40 60 0 20 40 60 0 20 40 60 Committee−Stage Bill Disagreement (%) Committee−Stage Bill Disagreement (%) Committee−Stage Bill Disagreement (%) Transportation and Infrastructure Veterans' Affairs

60 10 Count Count

30 5

0 0 0 20 40 60 0 20 40 60 Committee−Stage Bill Disagreement (%) Committee−Stage Bill Disagreement (%)

14 bill had been referred to.8 The means of committee passage percentage (passage percentage among committee members) are from final passage roll call votes and subsetted by party.9 The party breakdown of these passage percentages is of interest because it represents the level of agreement between the two parties in passing legislation. One one extreme, if the two parties perfectly agreed on legislation (members of both parties all vote yea), then we would see 100% passage percentage for Democrats and Republicans – resulting in a party difference of 0%, or 100% bipartisanship. On other other extreme, if the two parties perfectly disagreed on legislation (members of one party all vote yea and members of the other party all vote nay), then we would see one party have 100% passage percentage and the other party have 0% passage percentage – resulting in a party difference of 100%, or 0% bipartisanship. If, however, one party perfectly agreed (all voted yea) and half of the other party voted with the majority (half voted yea), then we would see the first party have 100% passage percentage and the second party have 50% passage percentage – resulting in a party difference of 50%, or 50% bipartisanship. The mean bipartisanship of committee members’ floor passage votes are presented in the last column of Table 1, and the distribution is presented in Figure 4. Figure 4 reveals that the vast majority of bills in Congress receive unanimous support from committee members. 45% of the bills in the 108th-113th Congresses that had been referred to committee and reached a passage vote received the full support of committee members from both parties. 58% of them were passed with at least 90% committee floor bipartisanship. The remaining one-third, however, were passed with varying levels of committee floor bipartisanship, with 8% being passed with 0% committee floor bipartisanship. Table 2 shows the mean levels of floor passage and floor bipartisanship among members of the committee where the bill had been referred. From this, we can see that the level of com-

8In the case where a bill is referred to multiple committees, an observation is recorded for each committee of referral. 9Passage roll call votes are defined as roll calls on the final passage of the bill up for vote; e.g. “On Passage,” “On Agreeing to the Resolution,” “On Passage of the Bill the Objections of the President Notwithstanding,” “On Motion to Suspend the Rules and Pass,” “On Motion to Suspend the Rules and Pass as Amended,” etc. Of note is the further potential separation of passage roll call votes into passage votes of clean bills and passage votes of amended bills – I account for these differences in my regression model by taking into account the number of amendments that had been passed.

15 Table 1 – Summary Statistics: Committee Floor Bipartisanship on Bill Passage. Mean percentages of commitee members voting yea and non-committee members voting yea on roll call votes.

Majority Committee Members % Voting Yea Committee Floor Congress Party All Democratic Republican Bipartisanship 108 R 90 81 97 81 109 R 87 76 96 77 110 D 87 98 73 74 111 D 90 98 78 78 112 R 76 51 94 50 113 R 78 54 96 52

Figure 4 – Distribution of Committee Floor Bipartisanship on Bill Passage The distribution of the ﬂoor passage consensus between Republican and Democratic committee members, for the Committees on Education and Labor, Financial Services, and Energy and Commerce during the 108th-113th Congresses.

400 Count

200

0 0 25 50 75 100 Committee Bipartisanship

16 mittee floor bipartisanship varies by committee. Interestingly, committees with jurisdiction over military, veterans, and national security have the highest levels of floor bipartisanship among committee members (Veterans’ Affairs with an average bipartisanship of 97%, Home- land Security at 90%, and Armed Services at 88%). Committees tasked with appropriating money and the budget have among the lowest levels of committee floor bipartisanship – Appropriations at 55% and Budget at 39%. Perhaps unsurprisingly, the Rules Committee, a powerful committee influenced by party leaders, has the lowest floor bipartisanship of 27%, a substantially lower level of cross-party consensus compared to all other committees.

Table 2 – Summary Statistics: Committee Bipartisanship on Bill Passage, by Com- mittee of Referral. Mean percentages of commitee members voting yea, by committee of referral and party (108th-113th Congresses).

Committee Members % Voting Yea Committee Floor Committee All Democratic Republican Bipartisanship All Committees 86 81 87 71 Veterans’ Aﬀairs 98 99 97 97 Homeland Security 95 94 94 90 Small Business 93 96 88 90 Armed Services 92 90 92 88 Government Reform 93 91 93 85 Science 89 91 84 78 Transportation 89 86 89 78 Education and Labor 88 85 88 74 Financial Services 85 87 81 71 Agriculture 80 72 84 66 Judiciary 82 70 90 63 Energy and Commerce 82 74 86 63 Intelligence 82 81 77 62 Natural Resources 80 80 75 59 Oversight 85 69 87 58 Appropriations 79 67 86 55 Standards of Oﬃcial Conduct 75 50 100 50 Ways and Means 78 66 81 49 Budget 68 60 70 39 Rules 72 42 78 23

17 Table 3 – Committee-Stage Bill Disagreement and Committee Floor Bipartisanship on Bill Passage. There is no significant relationship between the level of polarization on a bill during the committee stage and the difference between the percentages of each party’s committee members voting yea on that bill’s passage roll call on the floor.

Dependent Variable: Committee Floor Bipartisanship (1) (2) (3) Committee-Stage Bill Disagreement 0.040 0.051 0.054 (0.067) (0.046) (0.068) Num. Amendments Introduced −0.445∗∗ (0.202) Amended 23.3∗∗ (5.48) Num. Committees Referred 0.641 (3.24) Rules Suspended 37.5∗∗ (7.10) N 113 113 113 Committee Fixed Eﬀects Yes Yes Yes Congress Fixed Eﬀects Yes Yes Yes Note: Robust standard errors in parentheses. ∗p = 0.05,∗∗ p = 0.01

18 Table 3 reports OLS regression results that uses the Committee-Stage Bill Disagreement measure described in the preceding section with a dataset of passage roll call votes. The sample size – 113 bills – represents all bills from the Education and Labor, Financial Services, Energy and Commerce, Veterans’ Affairs, and Transportation and Infrastructure committees that had a committee hearing and reached a floor passage vote during the period 2003-2014 (108th through the 113th Congresses).10 I address potential selection effects after discussing the findings. The main independent variable, Committee-Stage Bill Disagreement, captures the distance between Republican and Democratic committee members on the bill under consideration during the committee stage (relative to the distance between Republicans and Democrats on the floor in that committee jurisdiction) in percentage point units. The dependent variable, Committee Floor Bipartisanship, represents the degree to which committees members from the two parties vote together in support of the bill’s passage, and is in percentage point units:

Committee Floor Bipartisanship = 100 − |Committee Democratic % Yea − Committee Republican % Yea|

Thus, the coefficient on Committee-Stage Bill Disagreement in Column (1) of Table 3 represents the change in committee members’ floor bipartisanship on a bill’s floor passage vote for a one percentage point increase in committee members’ Committee-Stage Bill Dis- agreement. That is, it represents the total effect of committee-stage bill disagreement on committee members’ floor bipartisanship on the bill. This coefficient is very small, and not statistically significant. From this, we can infer that there is no statistically significant relationship between the amount of committee-stage bill disagreement and committee members’ bipartisanship in floor passage votes.

10Hearings on the bills are required to construct the Committee-Stage Bill Disagreement measure (and is the limiting factor for the sample size, since not every bill gets a committee hearing), and the passage vote is needed for the dependent variable of Committee Floor Bipartisanship.

19 Another quantity of interest may be the effect of committee-stage bill disagreement con- trolling for variables that may mediate the effect of committee-stage bill disagreement on committee members’ floor bipartisanship.11 Number Amendments Introduced controls for the number of amendments that were introduced for the bill under question. An increased number of attempted amendments reflects desires to change the content of the bill (both minority and majority members may attempt an amendment). Amended is a dummy variable that equals 1 if the bill was amended and equals 0 otherwise. Number Committees Referred captures the number of committees that the bill had originally been referred to – a bill referred to multiple committees means that the bill spanned multiple policy jurisdiction. Finally, Rules Suspended is a dummy variable that equals 1 if the passage was voted on under suspended rules (which does not allow any changes or amendments to be debated on or voted on prior) and equals 0 otherwise. Suspension of the rules is used for quick passage on non-controversial bills. In specifications that control for these variables, there is still no statistically significant relationship between the amount of committee disagreement on a bill (Committee-Stage Bill Disagreement) and the degree to which committee members from opposing parties vote yea together on the passage vote on that bill (Committee Floor Bipartisanship). This lack of statistical significance is also seen when using committee reports to measure the committee-stage disagreement, as detailed in the previous section. Figure 5 is a coefficient plot of the OLS regression where the independent variable is Committee-Stage Bill Disagreement and the dependent variable is Committee Floor Bipartisanship, either measured using committee hearings (left, the same estimate as Column 1 in Table 3) or using committee reports (right). Regardless of the text data source used, there appears to be no statistically significant relationship between the amount of committee disagreement on a bill and the committee floor bipartisanship on the bill’s passage vote.

11For simplicity, I still use OLS, though other methods, such as the controlled direct eﬀect from Acharya, Blackwell and Sen (2016), may be better suited, particularly if there was an original total eﬀect.

20 Figure 5 – Committee-Stage Bill Disagreement and Committee Floor Bipartisanship on Bill Passage. There is no significant relationship between the level of disagreement on a bill during the committee stage and the difference between the percentages of each party’s committee members voting yea on that bill’s passage roll call on the floor, regardless of whether committee-stage bill disagreement is measured using committee hearings or committee reports.

0.2

0.1

●

0.0 ●

−0.1

−0.2 Committee Hearings Committee Reports Committee−Stage Disagreement

The model in Table 3, and the interpretations in the paragraph above, are based on the condition that the bill reaches a passage vote, since the dependent variable is calculated over the roll call votes on the passage of the bill. Is Committee-Stage Bill Disagreement related to the bill’s probability of reaching a passage vote? To investigate whether there are possible selection effects present in the results from Table 3, I run a logistic regressions that checks the potential relationship between committee-stage bill disagreement and whether the bill reached a passage vote on the floor. Table 4 shows that the coefficient on Committee-Stage Bill Disagreement is statistically insignificant when including committee fixed effects and congress fixed effects, suggesting that a committee’s disagreement on a bill (at least for the 661 bills for which I have speech to construct the measure) does not significantly affect the probability that the bill reaches a passage vote on the floor.

3.2 When Do Committee Members Vote Against Their Party?

The summary statistics in Tables 1 and 2 reveal that the committee bipartisanship on passage votes of bills from their committee tends to be somewhat strong for the ﬁve committees for which we have measures of committee-stage bill disagreement, ranging from 63% for Energy

21 Table 4 – Committee-Stage Bill Disagreement and Probability of Bill Reaching the Floor or a Passage Vote.

Dependent Variable: Bill Reached Floor Bill Reached Passage Vote (1) (2) (3) (4) Committee-Stage Bill Disagreement −0.02∗∗ −0.006 −0.019∗∗ −0.01 (0.004) (0.005) (0.017) (0.005) N 661 661 661 661 Committee Fixed Eﬀects No Yes No Yes Congress Fixed Eﬀects No Yes No Yes Note: Robust standard errors in parentheses. ∗p = 0.05,∗∗ p = 0.01

and Commerce to 97% for Veterans’ Affairs. The previous section has shown that there is no statistically significant relationship connecting committee-stage bill disagreement and committee bipartisanship on passage votes when the bill is on the floor, but it still remains that in order to achieve the high levels of committee bipartisanship on the floor, there must be members of the minority party voting with the majority party, or vice versa.12 This is not surprising for bipartisan bills, when both parties vote in the same direction on the floor, but on party-line bills, why would some committee members vote with the other party? Is there something inherent in committee service – and the benefits or incentives that come along with it – that leads certain committee members to vote against their party? Or are those committee members already inclined to vote against their party at the same rate, on the same kinds of bills, even if there weren’t on committee? We may think that committee service has an effect because it is well documented in the literature that committee membership comes along with an increase in targeted interest group donations, which could in turn influence the voting behavior of members when they are on a committee versus when they are off committee (Romer and Snyder, 1994; Grimmer

12In the vast majority of the cases, a bill reported by a committee receives full, or close to full, support from the committee’s majority party members In rare cases, a majority party committee member will defect and vote with the minority party, but the vast majority of cases is when the committee’s minority party members are split, with some minority party committee members voting with the majority party committee members.

22 and Powell, 2016). Likewise, interest groups have also been observed lobbying committees members more than non-committee members (Hojnacki and Kimball, 1998, 1999; Kollman, 1997; Hall and Wayman, 1990). If committee service increases the donations and lobbying of members when they are on committee versus when they are not on committee, then these “extras” may influence member votes when they are on committee versus when they are not. Additionally, two traditional theories of committees suggest why committee members may vote differently from non-committee members of their party. First, we may think that committee service comes with more power in drafting legislation. Since committees are where a bill gets debated and amended, committee members may be in a better position to extract specific benefits for themselves and their districts, versus when they are not on committee (Weingast and Marshall, 1988; Shepsle and Weingast, 1987). Furthermore, the informational theory of committees, e.g. Gilligan and Krehbiel (1989), suggests that committee members have more information on the bill and the relevant policies than non-committee members – in other words, committee members may know more about the true benefits and implications of the bill than non-committee members, and so may vote differently reflecting their information. I investigate whether committee membership changes a member’s voting behavior. I use individual roll call voting records, and for each party on the bills coming from each committee, I observe (1) the direction (yea or nay) that the majority of the party’s non- committee members vote and (2) the direction (yea or nay) that each committee member from that party votes. Let the variable Vote Against Party equal 1 if a committee member’s vote does not match the vote of the majority of their party’s non-committee members and 0 if the committee member’s vote matches the vote of the majority of their party’s non- committee members. Likewise, let Vote Against Party equal 1 if a non-committee member’s vote does not match the vote of the majority of their party’s non-committee members and 0 if the non-committee member’s vote matches the vote of the majority of their party’s non-committee members.

23 Let Vote Deviationijt be the mean of Vote Against Party for member i on bills coming from committee j during congress t. Let Committee Member ijt be 1 if member i was on committee j in congress j. Instead of comparing committee members to non-committee members, I exploit the panel structure of the data and use a difference-in-differences approach similar to Berry and Fowler (2016). Specifically, I include member-by-committee fixed effects and congress fixed effects, and estimate models of the form:

Vote Deviationijt = αij + θt + βCommittee Member ijt + ijt

For bills that had multiple committees of referral, I code its “committee” as the unique combination of all the committees of referral. This is done because for bills with multiple

committees of referral, we cannot separate out Committee Member ijt for each individual com-

mittee that the bill was referred to, since we only have a single Vote Deviationijt observation for each member on each bill. I drop bills that were referred to more than 5 committees of referral, since the larger the number of committees of referral, the more difficult it is to discern the effect of membership on a single committee on that bill.13 Table 5 presents regression results for all members in the 108th-113th Congresses (2003- 2014) on passage roll call votes.14 Panel A of Table 5 shows the results for minority members, on all bills in Columns (1) and (2) and on party-line bills in columns (3) and (4). Columns (1) and (3) show the results for passage votes on all bills and on party-line bills, with only member fixed effects, committee fixed effects, and congress fixed effects (a “naive” specification that does not include member-by-committee fixed effects). Party-line bills are bills where the majorities of both parties voted in opposite directions. Here we see that, on average, minority members of committees tend to vote slightly more frequently against their party on party-line bills coming from their committee, and majority members

13Bills that are referred to many committees are less likely to be policy-speciﬁc to an individual committee’s jurisdiction. Results are substantively similar around this cut-oﬀ. 14The similar table for passage roll call votes without suspended rules is presented in the Appendix; results are substantively similar.

24 Table 5 – Committee Membership and Voting Against the Party (Passage Votes). The impact of committee membership on a member voting against their party on passage roll call votes, during the 108th-113th Congresses (2003-2014).

Panel A. Minority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member 0.828 1.35 2.56∗∗ 3.07 (0.268) (1.69) (0.440) (2.55) N 72480 72480 41219 41219 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Panel B. Majority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member −0.51∗∗ 0.258 −0.267 1.62 (0.153) (0.913) (0.242) (1.88) N 88161 88161 50134 50134 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Note: Standard errors, clustered by member, are in parentheses.

25 of committees tend to vote slightly less frequently against their party on all bills coming from their committee. However, these estimates do not represent the effect of a “treatment” of committee membership, since they are likely biased due to selection reasons: committee members seek committee assignments that better serve their (district’s) interests, so members who seek assignment onto a certain committee may already have certain preferences that don’t perfectly align with their party on issues covered by that committee (e.g. committee members may be “preference outliers,” as argued by Weingast and Marshall (1988) and Shepsle and Weingast (1987)). Columns (2) and (4) attempt to remove selection bias by including member-by-committee fixed effects.15 Hence, the resulting estimates are computed by comparing changes in the frequency of a member voting against their party on bills coming from a specific committee, both when they are and are not members of that given committee, compared to the same outcome changes for co-partisan legislators who are not on that committee. Here we see no statistically significant effect of committee membership on the member’s frequency of voting against their party on bills coming from that committee (for either minority members or majority members). In other words, joining a committee does not seem to cause a change in the number of times a member will vote against their party. This suggests that member’s voting behaviors are relatively “sticky” when it comes to a change in committee service. Even though a member may receive more targeted interest group donations or be subject to more lobbying efforts as a committee member, or may have more opportunities to shape legislation to their liking as a committee member, these benefits of being on a committee do not seem to change the rate at which they will deviate from their party’s direction in that jurisdiction. If they are likely to deviate from their party’s direction on bills from a committee’s jurisdiction as a committee member, they would have at a similar rate even before (or after) they served on the committee.

15In other words, this includes ﬁxed eﬀects for member-committee pairs.

26 Voting against the party requires going against the party’s leaders and whips. That this effect is slightly more pronounced among minority committee members compared to majority committee members is not surprising when we think of the “veto points” that majority party leaders hold – if majority party leaders do not like a bill coming out of committee enough (if the majority committee members are in favor of a bill that is not aligned closely enough with the preferences of the majority party leaders), the speaker or the Rules Committee can kill the bill by not bringing it to the floor. Minority party leaders, however, do not have these “veto points,” and thus may have less control on minority committee members. One testable implication of this is that committee members may be more likely to vote against their party than non-committee members on bills coming from their committee during times when committees are stronger and less under the influence of party leaders. During the mid 1970s, the House implemented changes in rules and norms that transferred power from committees to party leaders and party caucuses Shepsle (1989); Rohde (1991). These changes include the Subcommittee Bill of Rights, the weakening of agenda control of committee chairs, the Speaker gaining the authority to select the majority party’s members of the Rules committee. For Democrats specifically, this also included imposing term limits on committee chairs, having a steering committee (whose membership is controlled by the Speaker) make committee assignments instead of the Ways and Means of committee, and giving the party caucus the authority to choose committee chairs, Appropriation subcommittee chairs, and ranking members. Hence, before the 1970s, in the “textbook Congress” era, committees were stronger and less under the control of party leaders compared to the recent Congresses. Is the effect of committee membership on voting against the party greater during this period? Table 6 is similar to Table 5, but for the 80th-90th Congress (the postwar period until 1969). These results show that minority committee members are almost four times more likely to vote against their party on bills coming from their committee during the 80th-90th Congresses (1947-1969) than during the 108th-113th Congresses (2003-2014).

27 This suggests that committee strength and party leaders’ inﬂuence does play a part in how likely minority committee members are to vote against their party.

Table 6 – Committee Membership and Voting Against the Party (Passage Votes) – “Textbook Congress” Era. The impact of committee membership on a member voting against their party on passage roll call votes, during the 80th-90th Congresses (1947-1969).

Panel A. Minority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member 0.980 −1.51 9.53∗∗ 6.33 (0.800) (2.53) (1.67) (5.85) N 29046 29046 17272 17272 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Panel B. Majority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member −4.90∗∗ −3.38 −7.14∗∗ −8.29∗∗ (0.73) (1.92) (1.35) (3.88) N 40878 40878 25050 25050 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Note: Standard errors, clustered by member, are in parentheses.

4 Conclusion

There may be high rates of polarization in Congress in recent times when we look at all roll call votes, but when we look at bill passage roll calls, bipartisanship is high among commit-

28 tee members, who more often than not, vote together to pass the bill on the floor. In this paper, I investigate whether committee-stage disagreement between the two parties’ committee members is related to the floor bipartisanship we see among committee members on passage votes on their bills. To do so, I proposed and constructed a new bill-level measure of committee-stage disagreement among committee members by using the speech in committee hearings and the text in committee reports. I find that on average, committee floor bipartisanship is strong on passage votes on bills from the committee. However, this committee floor bipartisanship is not related to the amount of committee-stage disagreement between the two parties’ committee members. In fact, this bipartisanship is happening because minority committee members are voting against their party (and with the majority committee members) more often than majority committee members. Furthermore, this frequency of voting against party does not significantly change when a member is on or off that committee. These members were inclined to vote against their party at the same rate in the committee’s jursidiction even when they are not on committee. This is good news for policy makers, scholars, and observers of Congress who are concerned that committee members are unduly influenced by special interest groups. Even though committee service has been shown to come along with an increase in interest group money and lobbying, this does not seem to change a member’s vote deviation behavior, at least not enough to make that member vote in the opposite direction more or less frequently. Granted, a member’s voting direction may have been favoring certain interests in the first place (or their party’s direction also favors those interests), before the member was assigned to the relevant committee, but the “extras” that come with committee do not seem to change this voting behavior any further. However, there may be other ways in which committee members may be influenced by special interest groups – it may not be the ultimate passage roll call vote that reflects this, but it may instead be certain language inserted into the bill

29 during the committee stage, or certain parts that are taken out during the committee stage. These types of changes are worth examining in future research.

30 References

Acharya, Avidit, Matthew Blackwell and Maya Sen. 2016. “Explaining Causal Findings Without Bias: Detecting and Assessing Direct Eﬀects.” American Political Science Review 110(3):512–529.

Berry, Christopher and Anthony Fowler. 2016. “Cardinals or Clerics? Congressional Com- mittees and the Distribution of Pork.” American Journal of Political Science 60(3):692– 708.

Bond, Jon R. and Richard Fleisher. 2000. Congress and the President in a Partisan Era. In Polarized Politics: Congress and the President in a Partisan Era, ed. Jon R. Bond and Richard Fleisher. Washington, DC: CQ Roll Call pp. 1–8.

Bonica, Adam. 2016. A Data-Driven Voter Guide for U.S. Elections: Adapting quantitative measures of the preferences and priorities of political elites to help voters learn about candidates. Russell.

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

Fenno, Richard F. 1962. “The House Appropriations Committee as a Political System: The Problem of Integration.” American Political Science Review pp. 310–324.

Fenno, Richard F., Jr. 1973. Congressmen in Committees. Boston: Little, Brown.

Fiorina, Morris. 1987. “Alternative Rationales for Restrictive Procedures.” Journal of Law, Economics, and Organization 3:337–345.

Gilligan, T. W. and K. Krehbiel. 1989. “Asymmetric Information and Legislative Rules with a Heterogeneous Committee.” American Journal of Political Science 33:459–490.

Grimmer, Justin and Eleanor N. Powell. 2016. “Money in Exile: Campaign Contributions and Committee Access.” Journal of Politics 78(4):974–988.

Hall, Richard L. and Frank W. Wayman. 1990. “Buying Time: Moneyed Interests and the Mobilization of Bias in Congressional Committees.” American Political Science Review 84(3):797–820.

Hojnacki, Marie and David C. Kimball. 1998. “Organized Interests and the Decision of Whom to Lobby in Congress.” American Political Science Review 92(4):775–790.

Hojnacki, Marie and David C. Kimball. 1999. “The Who and How of Organizations’ Lobbying Strategies in Committee.” Journal of Politics 61(4):999–1024.

Jacobson, Gary. 2000. Party polarization in national politics: the electoral connection. In Polarized Politics: Congress and the President in a Partisan Era, ed. Jon R. Bond and Richard Fleisher. Washington, DC: CQ Roll Call pp. 9–30.

31 Kollman, Ken. 1997. “Inviting Friends to Lobby: Interest Groups, Ideological Bias, and Congressional Committees.” American Journal of Political Science 41(2):519–544.

Laver, Michael, Kenneth Benoit and John Garry. 2003. “Extracting Policy Positions from Political Texts Using Words as Data.” American Political Science Review 97(2):311–331.

Lowe, Will. 2008. “Understanding Wordscores.” Political Analysis 16(4):356–371.

Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political-Economy History of Roll Call Voting. New York: Oxford University Press.

Ramage, Daniel, Evan Rosen, Jason Chuang, Christopher D. Manning and Daniel A. McFar- land. 2009. “Topic modeling for the social sciences.” NIPS 2009 Workshop on Applications for Topic Models: Text and Beyond 5.

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

Romer, Thomas and James M. Snyder, Jr. 1994. “An Empirical Investigation of the Dy- namics of PAC Contributions.” American Journal of Political Science 38(3):745–769.

Shepsle, Kenneth A. 1989. The Changing Textbook Congress. In Can the Government Govern?, ed. John E. Chubb and Paul E. Peterson. Washington, DC: Brookings Institution Press.

Shepsle, Kenneth A. and Barry R. Weingast. 1987. “The Institutional Foundations of Com- mittee Power.” American Political Science Association 81(1):85–104.

Smith, Steven and Christopher Deering. 1984. Committees in Congress. Washington, DC: CQ Press.

Stonecash, Jeﬀrey M., Mark D. Brewer and Mack Mariani. 2003. Diverging Parties: Social Change, Realignment, and Party Polarization. Boulder, CO: Westview Press.

Weingast, Barry and William Marshall. 1988. “The Industrial Organization of Congress; or, Why Legislatures, Like Firms, Are not Organized as Markets.” Journal of Political Economy 96:132–163.

32 A. 1 Appendix

Table A.1 – Committee-Stage Hearings and Committee Floor Bipartisanship on Pas- sage Votes.

Dependent Variable: Committee Floor Bipartisanship (1) (2) Hearing Held −7.88∗ (-7.88) Length of Hearing −0.003∗∗ (0.001) N 1054 113 Committee Fixed Eﬀects Yes Yes Congress Fixed Eﬀects Yes Yes Note: Robust standard errors in parentheses. ∗p = 0.05,∗∗ p = 0.01

33 Table A.2 – Committee Membership and Voting Against the Party (Passage Votes Without Suspended Rules). The impact of committee membership on a member voting against their party on passage roll call votes (excluding passage under rule suspension), during the 108th-113th Congresses (2003-2014).

Panel A. Minority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member 1.30∗∗ 0.551 2.43∗∗ 1.76 (0.422) (2.31) (0.459) (2.68) N 45681 45681 36439 36439 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Panel B. Majority Party Vote Deviation on Committee’s Bills All Party-Line Passage Votes Passage Votes Committee Member −0.931∗∗ −0.720 −0.737∗∗ −0.562 (0.225) (1.23) (0.241) (1.49) N 55239 55239 44414 44414 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Note: Standard errors, clustered by member, are in parentheses.

34 Table A.3 – Committee Membership and Voting Against the Party (Amendment Votes). The impact of committee membership on a member voting against their party on amendment roll call votes, during the 108th-113th Congresses (2003-2014).

Panel A. Minority Party Vote Deviation on Committee’s Bills All Party-Line Amendment Votes Amendment Votes Committee Member 1.78 0.443 2.01∗∗ −1.18 (0.376) (2.74) (0.423) (2.86) N 30069 30069 26860 26860 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Panel B. Majority Party Vote Deviation on Committee’s Bills All Party-Line Amendment Votes Amendment Votes Committee Member −0.097 −0.924 −0.304 −1.59 (0.215) (1.48) (0.266) (2.00) N 36501 36501 32476 32476 Member Fixed Effects Yes No Yes No Committee Fixed Effects Yes No Yes No Member-by-Committee Fixed Effects No Yes No Yes Congress Fixed Effects Yes Yes Yes Yes

Note: Standard errors, clustered by member, are in parentheses.