Legislative Parties and Voting Behavior in the Antebellum Congress

Legislative Parties and Voting Behavior in the Antebellum Congress September 11, 2016 Abstract Members of Congress turned to partisan organization as a solution to social choice and collective action problems despite the Founders warnings against parties. Virtually all research on partisanship and legislative behavior studies the period well after the modern twoparty system was established. We study the period from 1789 to 1861 and argue that the relationship between partisanship and legislative behavior varied in systematic ways across chambers and time. We further argue that the emergence of early parties helped manage regional conflicts. Using data on all recorded roll call votes cast in the antebellum Congress, we find support for our arguments. Partisanship was more strongly associated with voting behavior in the House than the Senate, varied in importance over time, and generally exhibited an inverse relationship with the salience of regional considerations. Our results have important implications for understanding how legislators principals, mass party organizations, and partisan institutions affected congressional behavior.

Political parties occupy a central place in scholarship on the U.S. Congress. Legislative parties are posited to exercise agenda control (e.g., Cox and McCubbins 2005), distribute particularistic goods (e.g., Balla et al. 2002; Levitt and Snyder 1995), control committee assignments (e.g., Shepsle 1978), devise rules and procedures (e.g., Binder 1997; Huber 1992), and influence presidential nominations to judicial (e.g., Moraski and Shipan 1999) and executive branch appointments (e.g., McCarty and Razaghian 1999). Few questions have received as much attention in congressional scholarship, however, as the importance of partisanship for legislative voting behavior (e.g., Brady, Cooper, and Hurley 1979; Clausen 1973; Cox and Poole 2002; Snyder and Groseclose 2000; Jenkins and Weidenmier 1999; Patterson and Caldeira 1988; Smith 2007). Despite the importance of parties for organizing legislative politics for most of the country s history, political parties were conspicuously absent from the nation s first congresses. It is wellknown that the Founders opposed political parties, with George Washington famously warning in his presidential farewell address that the common and continual mischiefs of the spirit of party are sufficient to make it the interest and duty of a wise people to discourage and restrain it. 1 Washington s sentiments notwithstanding, early members of Congress turned to partisan organization as a solution to social choice and collective action problems that initially impeded progress on a range of pressing issues (Aldrich 1995). A century and a half later, scholars of the textbook Congress observed that party membership is the single most important factor in roll-call voting. No other measurable dichotomy better predicts voting behavior (Turner and Schneier 1970, 239). In more recent congressional scholarship, the strong correlation between party membership and roll call votes has animated vigorous debates over the identification and interpretation of party effects on legislative voting behavior (e.g., Cox and McCubbins 1993, 2005; Cox and Poole 2002; Krehbiel 1993, 1998; McCarty, Poole, and Rosenthal 2001; Snyder and Groseclose 2000). 1 See https://www.gpo.gov/fdsys/pkg/gpo-cdoc-106sdoc21/pdf/gpo-cdoc-106sdoc21. pdf. 1

In this paper, we study how political party membership structured legislative behavior in the antebellum Congress. Virtually all empirical scholarship on party influence on legislative voting behavior begins with Reconstruction-era or later congresses (Cox and Poole 2002; Snyder and Groseclose 2000), well after the two-party system was well-established, or studies voting behavior on a small subset of congressional roll calls (Aldrich 1995; Jenkins and Weidenmier 1999). The antebellum period provides an important opportunity to study the connections between legislative behavior and partisanship in a context where Congress was considerably less institutionalized and in an era when party systems first emerged and subsequently underwent substantial change. We advance three primary arguments about antebellum political parties and congressional behavior. First, we argue that the importance of partisanship for legislative behavior varied across chambers and played a stronger role in the House. Members of the House and Senate served different principals and therefore had different incentives for investing in party brands and creating partisan organization in the chambers. Second, the relationship between partisanship and legislative behavior was dynamic over time in response to developments in the nature of party systems. Third, we argue that antebellum political parties helped manage regional and sectional conflicts in Congress, with regional considerations fading in importance for legislative behavior as parties played a stronger role. Our focus on the role of legislative parties in the antebellum Congress responds to recent calls for better integrating quantitative methods with studying important questions in American political development (Wawro and Katznelson 2014). We use data on recorded votes from both chambers of the 1st through 36th Congresses to study the association between partisanship and legislative behavior. Overall, our results provide strong support for our arguments. While partisanship was a statistically and substantively significant predictor of legislative voting records across the antebellum period, this relationship exhibited systematic variation between chambers and across time. In general, we find that the association between partisanship and legislative behavior was about 50 percent stronger in the 2

House than in the Senate, and that partisanship was a more important factor for legislative behavior with the emergence of mass party organizations and increasingly partisan legislative institutions. We also show that regional considerations increased in importance as the importance of party declined, though we find that both factors played increasingly important roles in the years immediately prior to the Civil War. Our results suggest that partisan developments both internal and external to Congress affected patterns of legislative behavior and have important implications for studying how the development of partisan institutions affected representation and congressional operations. The Partisan Organization of Congress Modern political parties organize political life. Political parties recruit and nominate candidates for election, thus organizing the ballot choices voters confront. For voters, partisanship serves as a heuristic for distinguishing prospective officeholders and evaluating proposed policy actions. Once elected, the partisan organization of government determines what policies are chosen, how government benefits are distributed, and which applicants are hired for bureaucratic positions. Scholarship pays particular attention to how parties organize legislative institutions (e.g., Cox and McCubbins 1993, 2005; Krehbiel 1991) and influence legislative voting records (e.g., Cox and Poole 2002; Krehbiel 1993; Snyder and Groseclose 2000). Political parties can shape legislative behavior through two main channels. In the first, roll call voting records closely reflect legislators party membership through the internal politics of the chamber. As Poole and Rosenthal (1997, 35) describe it, political parties, either through the discipline of power, leaders or through successful trades, function as effective logrollers... to map complex issues into a low-dimensional space. To do so, legislative party leaders create incentives for legislators to hew to the party line (e.g., Cox and McCubbins 1993). Positive incentives might include promising favorable committee assignments, scheduling a floor vote for a member s fa- 3

vored legislation, or adding funds for a line-item distributive project. Party leaders could also use negative incentives, such as threatening to remove a legislator from a powerful committee or endorsing a legislator s opponent in a party primary. These enforcement mechanisms help party leaders ensure that legislators support their party s position. Majority party leaders can also use negative agenda control to prevent consideration of items on which party members disagree (e.g., Cox and McCubbins 2005; Gailmard and Jenkins 2007). The use of negative agenda control helps ensure that legislatures only consider policy proposals with widespread agreement among majority party members; thus, partisanship may be highly predictive of legislative behavior even when underlying preferences vary among copartisan legislators. Second, legislative voting records can also reflect partisan influence through constituencyand electorally-induced pressures. For instance, local party organizations seek to nominate candidates who share their policy views, and they can help enforce party discipline by refusing to provide electoral support or recruiting challengers for officeholders who are insufficiently loyal. Partisan constituencies prefer to support legislators who identify with their preferred party and create demand for legislators to vote in ways consistent with party priorities. In both instances, legislators desire to win and maintain office leads them to cast roll call votes that reflect the party label on which they were elected (Mayhew 1974; Schlesinger 1966). 2 A wealth of findings demonstrate strong relationships between partisanship and legislative 2 It is also possible that party membership serves as a proxy for individual legislators preferences; if like-minded legislators are also members of the same party, similarities in legislative voting behavior may reflect legislators shared preferences. A sizable literature interrogates whether party affects legislative behavior independently from preferences (e.g., Cox and Poole 2002; Krehbiel 1993; Snyder and Groseclose 2000) and we do not intend to resolve this debate here. To the extent legislators roll call records reflect electorally-induced preferences, however, an association between partisanship and legislative behavior could provide evidence of a party effect independently of the legislator s own preferences. 4

behavior. Members of Congress report seeking information about how to vote from copartisans (Matthews and Stimson 1975) and placing a great deal of emphasis on the guidance provided by party leadership (Kingdon 1989). Quantitative studies of roll call votes generally show high correlations between party membership and legislative behavior, with party membership explaining substantively large portions of roll call voting behavior in the House (Weisberg 1978) and the Senate (Bullock and Brady 1983). More recent efforts to measure party effects document significant party influence on congressional votes cast since the end of Reconstruction (e.g., Cox and Poole 2002; Snyder and Groseclose 2000), while Nokken (2000) shows that legislators compile remarkably different roll call voting records upon switching parties. Research on nonpartisan legislatures sheds further light on the effect of legislative parties. For instance, Jenkins and Weidenmier (1999) compares voting behavior in the Confederate Congress during the Civil War to the U.S. Congress, while others have compared voting patterns in the nonpartisan Nebraska legislature to those in otherwise similar states with partisan legislatures (Welch and Carlson 1973; Wright and Schaffner 2002). In the aggregate, these studies find that legislative voting behavior exhibits considerably less structure and stability in nonpartisan legislatures. Virtually all scholarship on partisanship and legislative voting behavior, however, studies this relationship well after the modern two-party system was established. By focusing almost exclusively on congressional voting behavior after the Civil War, less is known about the connections between legislative voting patterns in Congress, the development of the first three party systems in the U.S., and the rise of mass-based party organizations. With a few important exceptions (Brady, Cooper, and Hurley 1979; Cox and Poole 2002; Patterson and Caldeira 1988; Snyder and Groseclose 2000), researchers have also paid less attention to temporal and interchamber variation in the relationship between partisanship and congressional voting behavior even though the developmental processes that characterized early party-building efforts suggest that the association between party and legislative behavior during this period was far from static. We address both of these limitations in existing research and focus on partisanship and leg- 5

islative behavior across the first 36 congresses in U.S. history (1789-1861). While partisanship is likely to have been a significant predictor of legislative behavior during this period, we argue that its importance varied in systematic ways between the House and Senate and across time. We further argue that the development of parties during this period generally helped to manage regional conflict. Our focus and approach thus follows recent calls for better incorporating quantitative methodology with political history while accounting for temporal variation and periodicity that characterizes many institutional structures (Wawro and Katznelson 2014). Parties and the Antebellum Congress Existing research suggests that legislative parties did not play the same role in the antebellum period as they did in other periods of American history. For much of the antebellum era, Polsby (1968) argues, legislative parties lacked the will or the means to collectively organize their members. Reflecting more broadly on the role of parties in the early Congress, Silbey (1989, 130) writes that the deep and persistent republican ideological tradition affected the development of Congress in this era, with its strong antagonism... to any institutions that sought to discipline people, work out compromises, bring people into coalitions. It was a powerful tradition that opposed political parties as divisive and feared wily political leaders as dangers to republican harmony. At the same time, parties were not completely inconsequential for legislative voting behavior during this time period. For instance, Aldrich (1995) shows that parties lent structure to roll call voting patterns as early as the third Congress (1793-95). While the association between partisanship and legislative voting behavior may have indeed been weaker during the antebellum period than it was in later periods of history, in the main, we expect that partisanship was an important predictor of legislative behavior even in the country s early years. We argue, however, that antebellum parties shaped legislative behavior to a significantly greater degree in the House than in the Senate. All Senators were elected by state legislatures during the antebellum era, while members of the House were elected directly by voters. Just as 6

the direct election of Senators changed the nature of representation in the Senate (Bernhard and Sala 2006; Gailmard and Jenkins 2009), differences in selection methods had important implications for defining legislators principals and characterizing their incentives to invest in legislative parties. Because Senators were primarily agents of the state legislatures, their continued tenure in office depended more on their ability to pass legislation and secure federal benefits desired by the state legislature. But because House members were chosen directly by voters, they had significant electoral incentives to develop and maintain a party brand. 3 As such, House members had much greater incentive to delegate authority to party leadership to set legislative agendas and enforce party discipline. Consistent with this account, scholars have noted the overtly partisan character of early Federalist Speakers of the House (Furlong 1967; Jenkins 2011) and the use of the voice vote for leadership positions by the end of the 1830s allowing House parties to monitor rank-and-file voting behavior that contributed to increasingly partisan elections for Speaker, clerk, and printer (Jenkins and Stewart 2012). Harlow (1917, 250) further notes that partisan considerations affected committee composition as early as 1797 and had been fully institutionalized by 1813. In the Senate, however, chamber leadership prior to the War of 1812 was primarily committee-based rather than party-based, and Roberts and Smith (2007) further note that the Senate altogether lacked a mechanism by which floor leaders could claim control of the floor agenda. Because of the weaker connection between mass political parties and the organization of the Senate, we expect that partisanship was more important for shaping legislative behavior in the House. 3 For instance, Carson and Engstrom (2005) provide evidence on this score by studying the electoral consequences of congressional voting behavior in the case of the contentious 1824 election between John Quincy Adams and Andrew Jackson, who represented different party factions. Members of Congress who supported Adams in the House contest but who represented districts supporting Jackson experienced substantial loss of constituent support in the subsequent midterm elections. 7

The nature of political parties, however, was not static during the antebellum period. To the contrary, the country flirted with two different party systems in its first 50 years before a third developed just prior to the Civil War. The emergence (and subsequent disappearance) of major parties like the Federalists and the Whigs suggests that partisanship had greater importance for legislative behavior in some periods than in others. As legislative parties emerged in the early congresses in response to collective action problems (Aldrich 1995), we expect the association strengthened between partisanship and legislative behavior in the country s first decade. As the first party system gave way with the collapse of the Federalist Party, however, partisanship likely receded in importance in the period following the War of 1812. Related research on this period shows that party cohesion declined between the 14th and 17th congresses (e.g., Binder 1995), and Poole and Rosenthal (1997) note that a single ideological dimension fares poorly in explaining legislative voting behavior during this Era of Good Feelings (roughly 1815 through 1825), which suggests that legislative parties did not help reduce the dimensionality of contentious political debates. Jenkins and Weidenmier (1999) reports evidence from votes on the Bank of the United States to show that while partisanship structured voting behavior on the issue in 1811, voting patterns in 1816 after the collapse of the Federalists were better predicted by constituency factors. By the 1830s, however, Congress was more fully a partisan institution, by which time the Whigs and Democrats had developed electoral machines, legislative organizations, [and] partisan followers (Binder 1995, 1099). More generally, Silbey (1989, 131) characterizes the period from 1830 to 1870 as a partisan era in which party, not the individual, was the key. Thus, we expect the association between partisanship and legislative behavior to strengthen with the emergence of mass party organizations and increasingly partisan legislative institutions. Finally, we expect an important interaction between partisanship and regional factors. Mass political parties helped nationalize political issues, many of which had regional or sectional origins, and thus the importance of regional considerations were likely to decline as partisanship increased in importance. Local party organizations constructed a national party structure that 8

linked interests from disparate regions and helped ameliorate regional divisions (Silbey 1981, 1991; Skowronek 1982). Elaborating on the second party system, Silbey (1981) observes that Whigs and Democrats opposed each other in most political debates, and that partisan differences extended to views on slavery. But because sectional differences on slavery did not extend to other issues, Silbey (1981) argues partisanship helped to restrain the influence of sectionalism and frustrated early attempts to unite the South around regional interests. Thus, we expect an inverse relationship between the importance of partisanship and regionalism for legislative voting behavior. Though existing research offers important insights into the predictors of especially salient roll call votes in the antebellum era and the historical development of mass political parties and legislative organization, considerably less is known about how partisanship more generally structured patterns of legislative behavior throughout this era. Our discussion above identifies several testable hypotheses about how partisanship operated differently across chambers, across time, and in conjunction with other potentially salient dimensions of legislative decision-making. We now describe the empirical approach we use to study these questions. Data We test the hypotheses outlined above using roll call voting behavior in the 1st through 36th congresses. 4 Our unit of analysis is the legislator-pair. As our key dependent variables, we use 4 These data were obtained from http://www.voteview.com. We are grateful to Keith Poole and Howard Rosenthal for making them available. We note, however, that our analysis is necessarily limited to recorded floor votes, which raises important questions about selection bias (e.g., Carrubba et al. 2006; Hug 2010). For instance, it may be possible that recorded votes were used by party leaders to enforce party discipline (Carrubba, Gabel, and Hug 2008), in which case partisan agreement may be higher than we would observe otherwise. For the purposes of our analysis, 9

legislator agreement scores generated from roll call data. 5 These scores express the number of instances in which legislators i and j voted the same way on recorded votes. The scores are normalized by the total number of recorded votes so that they represent the percentage of the time that legislators i and j voted together within a given Congress. Higher values of these entries indicate high degrees of similarity in legislative behavior. We generate agreement scores for each possible pair of legislators. 6 The distribution of these scores is displayed in Figure A.1 in the Supplementary Appendix. Pooling together the 1st through 36th Congresses yields 368,354 pairs of House members and 22,288 pairs of Senators for our main analyses. 7 We assess the importance of partisanship for legislative voting behavior by evaluating whether copartisan legislators vote together at higher rates than legislators who are members of different parties. Thus, our key explanatory variable is an indicator for whether a pair of legislators is affiliated with the same political party. We use data on legislator partisanship from McKibbin (1992) to create an indicator, Same party, which takes a value of 1 if both legislators were members of the same party and zero otherwise. 8 We note that all our results are robust to using the measure however, this potential explanation for variation in the use of recorded votes is precisely one of the main reasons we hypothesize why the importance of partisanship for legislative voting behavior may have varied over time. However, we note that further research is necessary on the use of recorded votes and potential selection issues. 5 Agreement scores have been used in several other analyses of legislative behavior, including Masket (2008) and Rogowski and Sinclair (2012). 6 We excluded members who did not serve for an entire Congress because their agreement scores with other members would be artificially low due to the smaller number of opportunities they would have had to agree with other legislators. 7 Complete information about the number of legislators and roll calls for each Congress are shown in Table A.1 in the Supplementary Appendix. 8 In our main models, we do not consider pairs of legislators who both identify as Independents 10

of partisanship reported in Martis (1989), and we report these additional sets of results in the Appendix. 9 We expect that higher rates of agreement among copartisan legislators indicate greater importance of partisanship for legislative voting behavior. Because politics was highly regional in nature during much of the nation s history, we created an indicator, Same region, which indicates whether a pair of legislators both represented constituencies from the same Census region. 10 We note, however, that tensions over the issue of slavery primary led to conflict between legislators from the North and the South. Thus, we have also estimated models while specifying the Same region variable using North and South as the regional categories, with the South defined as the 11 states of the Confederacy. Our results for both shared copartisanship and shared region are robust to this alternative characterization or nonpartisans as copartisans. 9 The McKibbin and Martis measures are distinct in several ways. The McKibbin data provide a single measure of partisanship for a legislator s entire career based on the party with which the legislator identified for the longest period of time. The Martis measure reports the legislator s partisanship for each Congress, and in principle this measure could vary over time for a given legislator. The McKibbin measure has the virtue of being relatively parsimonious, while the Martis measure is useful for characterizing partisanship during each Congress (and thus is plausibly pre-treatment). In the main, though, the measures largely agree and, as a result, we obtain virtually identical patterns of results using both measures. See Table A.2 and Figures A.2 and A.3. The number of observations differs between these models due to differences in the availability of partisan indicators. 10 The states belong to each Census region are as follows: New England: CT, MA, ME, NH, RI, VT; Mid-Atlantic: NJ, NY, PA; South Atlantic: DE, FL, GA, MD, NC, SC, VA; East North Central: IL, IN, MI, OH, WI; East South Central: AL, KY, MS, TN; West North Central: IA, MO, MN; West South Central: AR, LA, TX. 11

of regionalism. 11 We also account for other characteristics that might influence shared patterns of voting behavior. To account for the influence of constituency preferences on legislator voting records, we use an indicator for whether legislators represented the Same state. 12 We include an indicator for Shared Committee that takes a value of 1 if both members of a legislator-pair served together on at least one standing committee, and zero otherwise. 13 We also included a variable for Service Overlap that characterizes the total number of years the legislators had served together in the House and Senate. 14 Legislators who served together on committees and who have served together in Congress have additional opportunities to develop relationships, identify commonalities, and rely on each other as trusted cue sources (e.g., Matthews and Stimson 1975). If shared committee membership and congressional service increased similarities in voting behavior, we expect the coefficients to be positive. 15 Summary statistics for all variables are shown in Tables 11 Results are shown in Table A.3 and Figures A.4 and A.4 in the Appendix. 12 Ideally, we would use data on district (or state) presidential vote share, yet unfortunately such data are not available for the years prior to 1840. 13 These data were obtained from: David Canon, Garrison Nelson, and Charles Stewart, Historical Congressional Standing Committees, 1st to 79th Congresses, 1789-1947: 1st through 36th Congresses, House and Senate, February 13, 1998 (available at http://web.mit.edu/17.251/www/ data_page.html#1). 14 These data were obtained from Carroll McKibbin, Roster of United States Congressional Officeholders and Biographical Characteristics of Members of the United States Congress, 1789-1996: Merged Data, 1997, ICPSR Study #7803. This variable is measured in decades. 15 We note that we also estimated models that accounted for a range of other potential sources of similarities, including attending the same college, shared military service, and originating from the same state, but the results for these additional characteristics were substantively tiny or not statistically significant, and they did not change our findings for the key variables listed above. 12

A.4 and A.5 in the Supplementary Appendix. We model legislator agreement rates as a function of the explanatory variables noted above and indicators for each Congress. For simplicity, we present results from linear regression, but we discuss alternative modeling strategies and present those results in the Supplementary Appendix. Because agreement rates between legislators i and j could also be associated with legislator i s rate of agreement with legislator k, we cluster standard errors on legislator i within each Congress. We conduct two main sets of analyses to evaluate our hypotheses about the role of partisanship for shaping legislative behavior. In the first, we estimate the models described above separately for the House and Senate over the entire time period from 1789-1861, and then estimate models for three distinct periods of party organization identified by Jenkins and Stewart (2012) in the antebellum period. Jenkins and Stewart (2012, 13) describe the first period, 1789-1811, as the least institutionalized period of organizational politics as congressional leadership positions were infrequently contested and held few important powers. In the second period, 1811-1839, competition for leadership positions was quite intense as party organizations increasingly sought to use these positions for partisan gains. Jenkins and Stewart mark the third period from 1839 through the end of the Civil War, during which period party leadership as well as activists gained new means of enforcing party discipline among rank and file legislators. We examine our hypotheses about how the importance of partisanship varied between the House and Senate and across each of these organizational epochs. We also directly evaluate how the relationship between partisanship and roll call voting behavior varied over time on a more granular level. Specifically, we estimate our models separately for each Congress and compare the coefficients for Same party and Same region across the 36 Congresses in our data. This approach is similar to that advocated by Wawro and Katznelson (2014) and allows us to more closely explore when and to what degree the role of partisanship changed for shaping legislative behavior. These coefficients will also allow us to examine our hypothesis that the increases in the importance of partisanship reduced the importance of regional 13

considerations. Results Results for our first analysis of voting patterns in the U.S. Congress are shown below in Table 1. Results for the House are shown in the left panel of the table and results for the Senate are shown in the right panel of the table. The first column of each panel shows coefficients from a pooled model with all legislator-pair observations in each chamber, while the columns that follow show results for each of the three time periods discussed above. We begin by evaluating our hypothesis that partisanship was more important for voting behavior in the House than in the Senate. Across the 36 antebellum congresses, we find that partisanship was an important predictor of legislative voting behavior in both chambers. The coefficients for Same party are positive and statistically significant, showing that legislators who were members of the same party voted together more frequently than legislators who were not from the same party. However, consistent with our hypothesis, the results also show that the relationship between partisanship and legislative behavior was stronger in the House than in the Senate. The importance of shared partisanship was about 50 percent larger in magnitude in the House, where copartisans voted together about 11 to 12 percentage points more frequently than legislator-pairs from opposite parties, compared with about 7 to 8 percentage points more frequently in the Senate. This difference is significant at p <.001. We continue to find support for our hypothesis when evaluating the results for shared partisanship across each of the three time periods described above. In the period 1789-1811, the coefficient estimate for Same party is 0.098 in the House, compared with 0.076 in the Senate. The coefficients for both chambers attenuate during the second period, 1811-1839, but the coefficient in the House (0.066) is nearly twice the magnitude of the coefficient for the Senate (0.034). In the third period (1839-1861), the estimates more than double in magnitude for both chambers, yet the 14

coefficient for the House (0.143) is again larger than the coefficient for the Senate (0.111). Each of these differences is statistically significant at p <.01. The results in Table 1 also provide preliminary support for our hypothesis that the importance of partisanship for congressional behavior varied across time. In both the House and Senate, the coefficient estimates for Same party are significantly smaller in the second period (1811-1839) than they are in the first period (1789-1811). This difference is statistically signicant at p <.001 for the House and p <.05 for the Senate. The coefficients for the third periods (1839-1861), moreover, are significantly larger than they are for both the first and second periods. Each of these differences are statistically significant at p <.001. These initial comparisons suggest that partisanship has a stronger association with legislative behavior when political competition is greater, particularly when accompanied by the presence of mass-based parties and party organizations, as in the third period. Regional considerations also played an important role in structuring legislative voting behavior. In each model, the coefficients show that legislators from the same region of the country exhibited greater similarities in voting behavior than other legislators. Though these variables do not allow us to distinguish constituency influence from preferences that may be shared among legislators from similar geographic areas, they do suggest that legislators that represented similar constituencies compiled more similar voting records in Congress than legislators from different states or regions. We also find some differences in how the Same region coefficient varied across time. In the House, the coefficients for this variable are relatively stable, ranging from 0.047 to 0.051, while in the Senate the coefficient estimates vary somewhat more: 0.037 in the first period, 0.063 in the second, and 0.049 in the third. This preliminary comparison suggests that parties may have more effectively managed regional conflict in the Senate, where the Same region variable is smaller in magnitude in the time periods where the Same party variable is largest, than in the House. The results for the other covariates are also of substantive interest. The coefficient estimates 15

for Same state are positive and statistically significant in all models, indicating that legislators that represent the same state voted together at higher rates. Importantly, the magnitudes of the coefficients are larger in the Senate than in the House, which is consistent with our theoretical argument above that Senators behavior in office reflected responsiveness to their principals, state legislatures. We also find, interestingly, that the coefficients declined in magnitude over time. Overall, shared committee membership exerted a relatively minor influence on shared voting patterns. The coefficients are positive and statistically significant for the first period in both the House (0.032) and Senate (0.186), but are quite small and/or not statistically distinguishable from zero in the later periods. These results suggest that once standing committees became an institutionalized part of the House and Senate, shared committee membership was not associated with greater similarity in voting behavior. Finally, we find limited evidence that shared congressional service was associated with greater similarities in voting behavior, as virtually all of the coefficients are negative. Table 1 goes here. The results shown above are robust to a number of alternative modeling strategies and specifications. First, we obtain substantively identical patterns when estimating negative binomial regressions in place of linear regression. 16 Second, we conducted a supplementary analysis in which our dependent variable is the difference in two legislators DW-NOMINATE scores rather than their agreement rates. For these models, we would expect that shared partisanship would result in a negative relationship with the difference between DW-NOMINATE scores, which would indicate more similar voting patterns among copartisans. We find strong evidence for these patterns when analyzing both the first and second dimension of legislators DW-NOMINATE scores. The results are shown in Table A.8 in the Appendix. 16 See Tables A.6 and A.7 in the Appendix. 16

As we noted above, our analyses cannot rule out the possibility that the Shared partisanship variable reflects legislators who generally share similar ideological dispositions (see, e.g., Krehbiel 1993). While precisely estimating the causal effect of partisanship remains an important challenge in this body of scholarship, we attempted to make headway on this question by replicating our analyses in Table 1 for those pairs of legislators who served together for more than one Congress. We estimated the models above while also including the lagged value of the difference in DW-NOMINATE scores between the pairs. To the degree that DW-NOMINATE scores reflect legislator preferences, the coefficient estimate for the difference in DW-NOMINATE scores should reflect the influence of shared preferences while the coefficient estimate for Shared partisanship characterizes the importance of party ties. 17 Our analysis focused on the House due to the relatively small number of legislators who served in successive Senates. We find that the coefficients for the difference in DW-NOMINATE scores are negative and statistically significant, indicating that legislators who voted more dissimilarly in a previous Congress exhibited lower agreement rates in the current Congress. However, we continue to find positive and statistically significant coefficients for the Shared partisanship variable. 18 These results suggest that our findings about the importance of partisanship reflect the importance of legislative parties over and above shared preferences. Across both chambers, we find strong evidence that partisanship played an important role in structuring legislative behavior in the antebellum Congress. Consistent with our hypothesis, the magnitude of this relationship was stronger in the House than in the Senate. Our theoretical discussion suggested this finding may be explained by the relatively stronger partisan institutions that developed in the House, especially related to chamber leadership positions, while Senators 17 We acknowledge that this solution is imperfect, however, because lagged DW-NOMINATE scores also likely reflect lagged party effects and because norms of congressional tenure changed during this period. 18 Results are shown in Table A.9 in the Appendix. 17

had fewer incentives to invest in partisan brands and discipline. However, the results also reveal some interesting variation in the strength of partisanship across the three periods we studied. We now examine our hypotheses related to temporal variation in the relationship between partisanship and roll call voting behavior, and the interactions between partisanship and regionalism, in greater detail. Partisanship and Legislative Voting Behavior across Time We estimate the models shown above separately for each Congress and chamber. Figure 1 below shows the coefficients for Same party when estimated separately for each of the 36 Congresses that met between 1789 and 1861. House coefficients are shown in the top plot and the bottom plot shows coefficients for the Senate. The plotted points are the coefficient estimates and the vertical lines are the 95 percent confidence intervals. The horizontal dashed line at zero indicates the null hypothesis of no association between shared party membership and legislator agreement rates. The plot for the House reveals substantial variation across time in the relationship between partisanship and legislative voting behavior, and in ways largely consistent with our hypothesis. In general, partisanship was a significant although inconsistent predictor of legislative behavior during the first party system; however, with the disappearance of the Federalists, partisanship had virtually no association with roll call voting patterns. With the emergence of mass party organizations by the mid-1830s, partisanship increased in importance and remained a statistically and substantively important prediction of voting patterns up to the Civil War, though its magnitude varied somewhat. On a more granular level, and consistent with the analysis of individual roll call votes found in Aldrich (1995, chapter 3), the coefficients for shared partisanship are fairly precisely estimated zeroes for the First and Second Congresses. By the Third Congress, however, the coefficient is positive and statistically significant, and increases to a local peak in the Sixth Congress (1799-18

1801) before declining in each of the next three Congresses. The coefficient then increases again between the 10th and 13th Houses, declines substantially in the 14th, and then is indistinguishable from zero for each of the 15th through the 19th Houses. Our results show that partisanship was not an important predictor of roll call voting patterns during the Era of Good Feelings and are consistent with accounts that emphasize the absence of legislative parties and the lack of structure in legislative voting behavior during this period. The coefficients then increase in magnitude between the 20th and 26th Houses, during the Jacksonian era as the seeds of modern political parties were sown. The coefficients decline briefly between the 31st and 32nd Houses as the Whig Party disintegrates, and then monotonically increase through the final Congress of the antebellum period. The results for the Senate are broadly consistent with those for the House. Note, first, that the confidence intervals are much larger for the Senate coefficient estimates due to its considerably fewer members. Though all of the coefficient estimates during the first 15 Congresses are positive, and are occasionally statistically significant, they mostly hover between 0.05 and 0.10 and provide less evidence of temporal variation than we find in the House. Similar to the House, however, the coefficient estimates during the Era of Good Feeling (here, the 16th through the 22nd Congresses) are mostly statistically insignificant and very small in magnitude. The coefficients are positive and statistically significant beginning with the 23rd Congress and continuing for the rest of the time series. Like the House results, though, we find some variation during this period, as the coefficient peak in magnitude in the 27th Congress and then decline through the 31st Congress. The coefficients then increase slightly over the last several Congresses before the Civil War, but the increase is considerably smaller in magnitude than it is in the House. The figure reveals that broad trends in the importance of partisanship for congressional voting behavior applied similarly to both chambers. Legislative behavior was positively associated with partisanship as the first party system emerged, though this relationship was inconsistent and sometimes statistically insignificant. Moreover, partisanship was not an important predictor 19

of legislative behavior during the period between roughly 1815 and 1825, but after that as the forebears of today s two-party system emerged voting behavior in both the House and Senate was significantly shaped by partisanship. Partisanship significantly and suddenly declined in importance as the second party system collapsed shortly after 1850, but then quickly re-emerged as a strong predictor as the Republican Party organized after the demise of the Whigs. As we hypothesized, the relationship between partisanship and legislative voting behavior was not constant in the antebellum Congress, but instead it varied in ways that largely reflected systematic developments in the American party system both within and outside of Congress. Figure 1 about here. Figure 2 shows variation in the coefficients for the Same region variable across both chambers and allows us to examine our hypothesis about the interaction of partisan and regional considerations in the antebellum Congress. The patterns are broadly similar across both chambers. Regional considerations were strongest in the early Congresses, prior to the emergence of political parties, but then declined in importance as the Democratic-Republican and Federalist parties helped organize political conflict. The magnitudes of the coefficients then increase and are relatively large in magnitude during the Era of Good Feelings from roughly the 15th through the 20th Congresses. The Same region coefficients decline starting in the mid-1820s as the Democrats and Whigs begin to organize mass political parties, and are small in magnitude and often statistically insignificant during much of the late 1820s through the mid-1840s. As Figure 1 shows, during this period the Shared party coefficients were generally quite large in magnitude. However, the Same region coefficients increase throughout most of the 1850s, just as regional considerations propelled the nation to civil war beginning in 1861. Figure 2 about here. The findings in Figure 2 provide some support our hypothesis that political parties helped manage regional and sectional conflict and are consistent with other accounts that emphasize that 20

regional considerations often played a greater role in shaping congressional voting patterns in the absence of well-organized political parties (Poole and Rosenthal 1997). More generally, the figure indicates a substitution effect between partisanship and region across much of the antebellum period, with an important exception. The coefficients for both partisanship and region increase in magnitude during the last congresses of the antebellum era. By the mid-1850s, regional issues came to dominate national politics, and Democrats and the newly-formed Republican Party could no longer ignore them. The increasing salience of regional issues for legislative debates could have undermined the ability of national parties to absorb regional differences. This was likely to be particularly acute among Democrats, who often differed on major policy issues based on their regional affiliation. Were Some Party Ties Stronger than Others? In a final set of analyses, we explore temporal variation across parties. Several accounts emphasize the varying strengths of party commitments in the antebellum period. For instance, Jenkins (2011) notes the hard-line partisan nature of Federalist Speakers of the House, while Democratic-Republican Speakers governed the House with less partisan fervor. Discussing the second party system, (Aldrich 1995, 135) points out that the Whig Party was organized more around personal commitment and leadership of moderates rather than institutionalized sources of party discipline. Accounts such as these suggest that the association between partisanship and legislative behavior could have varied in strength across political parties. We estimated separate models for each Congress and allowed the coefficient for shared partisanship to vary across each of the major parties. We estimated these models for the House only due to the very small number of legislator-pairs within each party in the Senate. The results are shown in Figure 3. Two key findings stand out. First, in general, the magnitudes of the relationship are quite similar when evaluating the most significant party cleavages in each period. Between the 4th and the 14th Congresses, the coefficient estimates for shared partisanship among 21

Federalists and Democratic-Republicans generally parallel each other. Though the magnitude of the coefficients is generally larger for the Federalists, the differences are not statistically distinguishable from zero. (Confidence intervals of the coefficient estimates are not shown in the figure for visual clarity.) During the second party system, the figure also shows that partisanship was roughly equally influential for the behavior of Democrats and Whigs; any observed differences in the coefficients are not statistically distinguishable from zero. The second important finding from the figure is that pairs of legislators in which both identified as Independents or were unaffiliated with any party did not vote together at higher rates. The coefficient estimates for these legislators are almost always zero or very close to zero, and are clearly smaller in magnitude relative to the coefficients for partisans. This result indicates that legislators who were united by their absence of partisanship exhibited no propensity to vote together at higher rates, and points to the importance of party membership as a key factor that structured legislative voting behavior in the antebellum House. Though not dispositive, the absence of any systematic patterns of shared voting behavior between Independents also provides some evidence to suggest that party membership affected legislative behavior above and beyond individual legislators preferences. Figure 3 about here. In sum, our findings are generally consistent with our hypotheses and indicate that partisanship varied in systematic ways across chambers, time, and parties for the first 70 years of U.S. history. These patterns appear to reflect predictable responses to important partisan developments both inside of and external to Congress. In the country s infancy, political parties helped lend structure to individual-level and aggregate patterns of legislative behavior, though this process was interrupted by the emergence and disappearance of major political parties that coincided with the country s first three party systems. 22