The Elasticity of Partisanship in Congress: An Analysis of Legislative Bipartisanship

Transcription

1 The Elasticity of Partisanship in Congress: An Analysis of Legislative Bipartisanship Laurel Harbridge College Fellow, Department of Political Science Faculty Fellow, Institute for Policy Research Northwestern University Paper prepared for presentation at the American Political Science Association Meeting, Washington D.C., September 2-5, DRAFT In recent election cycles, Republican members of Congress distanced themselves, at least rhetorically, from an unpopular party image and president. Why should the behavior of members change in the short-term if there are no changes in ideology or membership? This project explores two complementary questions central to this puzzle. First, do members of Congress go beyond rhetoric, to actually changing their legislative behavior, in response to changes in the popularity of their party? Second, do all members change their behavior or are changes restricted to the most electorally vulnerable members? By looking at the bill cosponsorship patterns of individual members from the 98 th -109 th Congresses, I examine the likelihood that members engage in bipartisan coalitions as a function of party popularity in the electorate. The evidence suggests that members do respond to the popularity of their party in the electorate, engaging in more bipartisanship when their party is less popular. However, these effects have changed over time, with members becoming less responsive to party popularity. The results of this study have implications for a number of areas of legislative politics, including legislative behavior and preferences, electoral responsiveness, and party branding.

2 Most research on the behavior of member of Congress focuses on explanatory variables that are constant, or at least relatively constant, for each member over the short-term. Variables such as the party affiliation of a member, average presidential vote in the member s district, and divided government do not fluctuate over the course of a Congress. While these variables have proven fruitful for exploring a number of important questions in political science, including party influence versus preferences and representation, they assume that the relevant variation is between members or over long periods of time. But does legislative behavior vary in the shortterm as well as the long-term? By narrowing the time period under consideration to three month quarters, this paper explores legislative behavior in response to factors that may vary within a given Congress. In particular, this paper examines changes in legislative behavior through the lens of party popularity. Work on legislative behavior suggests that members have an underlying ideology, which may reflect their own preferences, party, and constituent influence, and that this ideology affects the behaviors of members, including how they vote and what coalitions they form. While this perspective may be instructive for summarizing behavior over the course of a Congress or a year, it may obfuscate variation in the short term. Putting ourselves in the shoes of members of Congress, let s consider when members can be expected to deviate from this ideological baseline. Assuming that individual preferences are fixed, variation can reflect changes in the relative weight of constituent and party interests, or changes in the content of the legislative agenda. This paper explores the relative weight of constituent and party interests, leveraging variation in party popularity. By looking at the bill cosponsorship patterns of individual members in the House of Representatives, an area of the legislative process that is within the purview of individual 1

3 members in contrast to roll call votes, I examine the likelihood that members engage in bipartisan coalitions as a function of party brand popularity (proxied by party advantages in party identification in the electorate). I hypothesize that members engage in more bipartisanship when the party is unpopular than when it is popular. Additionally, I hypothesize that these effects are stronger for electorally vulnerable members. This paper sheds light on a number of areas, including the influence of parties, crosspressured members, electoral responsiveness, and policy coalitions. The remainder of the paper is organized as follows. The first section reviews the literature on parties and cross-pressured members. The second section draws on findings from the literature and anecdotes from recent election cycles to make predictions about the behavior of members in response to party popularity. The third section explores the party brand question empirically. The fourth section concludes. Literature on the Determinants of Legislative Behavior Members of Congress are generally assumed to act in response to pressures from their party, their constituents, and their own preferences. Combined, the pressures from these factors result in behaviors that are summarized by a member s ideology. If we are interested in shortterm fluctuations in members behavior, at least one of two factors can be expected to be a work. Either members change how responsive they are to one or more of these groups, or the types of issues on the legislative agenda must change. That is, increased bipartisan cooperation by a member could reflect either a declining responsiveness to the party and increased attention to constituent preferences (assuming constituents desire bipartisanship) or a change in the type of issues on the agenda, with more issues appearing on the agenda that garner broad, bipartisan support. The assumption throughout this paper is that, at least when considered in broad terms, 2

4 parties are likely to desire strong showings of party unity from their members whereas constituents prefer greater bipartisanship. While some research highlights variation in desire for bipartisanship among strong partisans and others (Harbridge and Malhotra 2010), a range of research approaches have shown that many constituents desire greater moderation/bipartisan cooperation (Pew 2010; Carson et al. 2010; Canes-Wrone et al. 2002). This section provides an overview of literature on party, constituency, and ideology influences on members of Congress, as they relate to legislative behavior. Much of the research on political parties focuses on the positive electoral benefits of the party brand and the need for collective action in order to enact a legislative agenda (Cox and McCubbins 1993; Cox and McCubbins 2005). Parties provide a number of tangible resources to members including campaign funding, valuable committee assignments, and brokering deals to help legislation introduced by the member succeed (Smith 2007). Members who are loyal to the party and vote the party line are much more likely to receive these benefits than disloyal members. Parties also provide a brand or franchise name to voters in elections. However, it is typically assumed that the brand name is positive, similar to the expectations that the McDonalds brand name is positive for consumers looking for cheap, safe, fast food. Combined, the tangible resources and the brand name benefits suggest that parties may be influential in Congress. A relatively straightforward assumption is that members who respond to party pressures will tend to vote the party line and engage in other partisan behaviors, leading to high levels of partisanship and little bipartisanship in Congress. However, members may face trade-offs between the positive aspects of the party brand and party loyalty, and the potentially negative consequences of being too partisan or extreme for their district. As noted by Mayhew (1974), congressmen must achieve reelection before they can 3

5 achieve any other goals of policy influence. Reelection requires that members not stray too far from the preferences of their constituents. Numerous scholars have suggested that it is the electoral concerns of members that facilitates representation of the public (Erikson 1978; Erikson and Wright 1980). Members who are out of step with their constituents are more likely to end up out of office (Canes-Wrone et al. 2002). Where safely partisan seats exist, constituency pressures may reinforce the effect of a member s party. In other cases, however, constituency pressures may be in opposition to the pressures from party leadership. Since the public tends to be less ideological and partisan than party leaders, being out of step with constituents typically occurs when members are too partisan for the preferences of their district. Cross-pressured members those members whose constituent preferences and party preferences are in opposition may have the most to lose from toeing the party line. Many of the party theories note that members who are cross-pressured by their party and constituency may be able to break party discipline on certain votes but this tends not a central feature of their analysis. As noted by Lebo et al. (2007), party unity is a double-edged sword it increases the likelihood of legislative success but can also expose members in moderate districts to electoral defeat. Similarly, Patty notes that a fundamental tension occurs when [a] member s individual and collective interests are in conflict (2008, 640). For cross-pressured members, too much party loyalty may spell electoral defeat at any point in time. When a member s party is unpopular, party loyalty may be additionally detrimental. However, the popularity of the party may also influence the behavior of members who are not cross-pressured, as they feel the need to run against Washington as an independent thinker. Thus, having a legislative record that toes the party line is an easier sell in a campaign when the party is popular than when the party is 4

6 unpopular, but the size of this effect may vary depending on whether a member is crosspressured or not. What Matters to Members The Changing Benefits of the Party In the 2008 election, Republican members of Congress faced an uphill battle. President G.W. Bush s approval rating was dismal and the GOP image was tattered, leaving many members to focus their campaign energies on distancing themselves for Bush and the Party. An ad from a Republican Senator in Oregon highlights this effort, stating, I m Gordon Smith and I approve working across party lines (Smith 2008). His website showed no references to his party affiliation and where he referenced support from one of the presidential candidates, he focused on Obama (who later clarified that he did not endorse Smith). In both the 2006 and 2008 elections, Republican members of Congress like Smith distanced themselves, at least rhetorically, from an unpopular party image and president. Similarly, Democratic members in competitive districts are weighing the costs of tying their electoral fates to President Obama in 2010 or having him help by staying away, as his approval numbers have fallen since his inauguration (Zeleny 2010). This suggests that members viewed presidential approval as informative about the popularity of their own party. Fearing that the party label would hurt them electorally, members may have focused on strengthening their records of bipartisanship, in an effort to paint themselves in the election as independent and free thinkers rather than someone who just toed the party line. Although there is a robust debate in scholarly work about the relative importance of party and ideology (Smith 2007; Cox and McCubbins 1993, 2005; Rohde 1991; Krehbiel 1998), the question this paper seeks to address is whether the relative benefit of being associated with the party changes over the short-term and whether this effect can be seen 5

7 in the behavior of members. It does not seek to adjudicate whether ideology or party is dominant. Turning to the primary interest of the paper variation in legislative behavior over the short and long-term the popularity of a member s party may lead them to work with the party or to seek to differentiate themselves from the party. If members are rational actors seeking reelection, we can expect them to turn toward the party when the benefits of this association outweigh the costs. Focusing on this party cost-benefit story, there are two possible effects. On the one hand, when party popularity declines members may band together in an effort to rebuild the party image. The resulting behavior would be an increase in partisanship. However, this move could be electorally costly to members, particularly in competitive seats. On the other hand, a decline in popularity may lead members to turn their focus away from the party and towards their constituents, leading to an increase in bipartisanship. The anecdotal evidence of the 2008 and 2010 campaigns that were mentioned above are both supportive of the later story. The logic behind this argument is that members use bipartisan cooperation to signal to their constituents that they are more than just part of an unpopular president and party, and that they are making compromises for the good of their constituents. In contrast, when party popularity is high, members rely on a spillover from this goodwill among constituents, focusing on more partisan behaviors that may be in line with their party but too partisan for the preferences of the district. If members are concerned about the effects of the party brand in the next election, we might expect members from marginal districts to be more responsive than members in safe district. For instance, Grynaviski (2010) finds that members in centrist districts are more likely than other members to develop a personal brand rather than to rely on the party brand, as evidenced by greater contact with constituents and greater name recall by constituents. 6

8 The party brand cost-benefit hypothesis, and the marginal member corollary, can be stated as follows: Hypothesis (Party Brand Cost-Benefit): Party popularity is negatively related to legislative bipartisanship. Hypothesis (Party Brand Cost-Benefit Marginal Seats Corollary): Members in safe districts will be less responsive to party popularity than members in marginal districts. Legislative Bipartisanship and Party Popularity - Data and Results In moving from the abstract to the empirical, it s important to think about how to measure party popularity in a way that is consistent across time and allows for variation over the short term (unlike electoral returns). Although an ideal measure of party popularity would be the popularity of the congressional party, public opinion polls of congressional Democrats and Republicans are relatively infrequent. Therefore, this paper considers a number of potential measures of party popularity, ultimately leveraging advantages in party identification (PID) in the electorate as a proxy for the popularity of each of the two major parties. For Democratic members of Congress, the PID advantage in each poll is the percent of respondents identifying as Democrats minus the percent identifying as Republican. For Republican members of Congress the calculation is reversed. Although imperfect as a measure of party popularity, available polls suggest a high correlation between party identification advantages and other measures of party popularity, including congressional party approval. Using Gallup data on congressional party approval (for select months between June 1999 and June 2006 where data is available) indicates that the correlation between PID advantage (for the president s party 1 ) and approval of the president s party in Congress is 0.57 (p < 0.001). A second possible measure is the Gallup generic ballot 1 In order to make comparisons across four measures PID advantage, generic ballot advantage, presidential approval, and congressional approval, all data is calculated as the advantage/popularity of the president s party. 7

9 question. 2 Although asked more frequently than congressional party approval, this measure also suffers from substantial missing data, particularly in non-election years. Nonetheless, the correlation between PID advantage and the generic party ballot advantage is 0.93 (p < 0.001). A third option is presidential approval. Although correlated with congressional party approval of the president s party (r = 0.94, p < 0.001) and demonstrating high short-term variation, presidential approval may be caused by a number of factors, which may themselves affect legislative behavior. These include economic conditions, rally-around-the-flag events, and scandals. The correlation between presidential approval and the PID advantage of the president s party is the weakest of the possible measures (r = 0.49, p < 0.001). Given the data limitations and/or concerns about these alternate measures capturing features beyond party popularity, this paper utilizes PID advantages as the proxy for party popularity. Using the Roper Public Opinion Archive, I have collected every Gallup survey of party identification from January 1983 through December 2006 (corresponding to the 98 th to 109 th Congresses). The party identification question from Gallup is worded as follows: In politics, as of today, do you consider yourself a Republican, a Democrat, or an independent? (Asked of independents: As of today, do you lean more to the Democratic Party or the Republican Party?). Only polls with national adult samples were used and all polls in which Gallup presented the data grouping leaners with partisans were dropped. For each poll, the Democratic and Republican PID advantage was calculated by subtracting the percent of respondents who identify with the Republican (Democratic) party from the percent of respondents who identify with the Democratic (Republican) party. Once the partisan advantages were calculated for each poll, the 2 The generic ballot advantage is defined as the percent of respondents indicating a vote for the party of the president minus the percent of respondents indicating a vote for the opposing party. Data on the generic ballot was provided by Gallup, using all surveys between 1983 and As with the presidential approval data, the generic ballot data was averaged by quarter-year. 8

10 polls were aggregated by quarter-year, taking the mean of the partisan advantages of the constituent polls. Since the margin of error on each individual poll is relatively constant across time, each poll was treated equally in the aggregation. In an effort to infer causation from the results, analyses are run using a lagged value of PID advantage from the previous quarter in addition to models with simultaneous measures of PID advantages and bipartisan cooperation. Figure 1 plots the Democratic PID advantage over time, from 1983 to 2006, using three month quarterly estimates. [Figure 1 about here] To measure legislative behavior, I focus on cosponsorship. I opt for cosponsorship coalitions rather than roll call votes for a number of reasons. First, cosponsorship occurs with a frequency that allows me to break down the time period of analysis into quarters, rather than years or Congresses, allowing me to look at variation within a given Congress. Second, beyond the standard problems with roll call votes (notably that they are not a random sample of bills, and often tend to be bills which divided the parties from one another), there are reasons to believe that the degree of contentiousness of the vote could be correlated with when the vote occurs in a Congress or year. For instance, if legislation has faced a difficult battle through Congress, it may not face a vote until the end of the term. Using data collected by James Fowler (2006a, 2006b), I use the party of each bill s sponsor and cosponsors to characterize each bill as partisan or bipartisan. Bills are defined as bipartisan if at least twenty percent of the cosponsors are from the party opposite the party of the bill s sponsor. In other work, I have used alternate cut-points, such as 30% and 40%, without finding major differences in anything but the baseline percentage of bipartisan bills. Once bills are classified as partisan or bipartisan, I then examine what bills each member cosponsors. I 9

11 calculate both the number and percent of bills that a member cosponsors in each three month period (quarter) that are partisan and bipartisan. Alternatively, I calculate the percent of bills that a member cosponsors where the original sponsor is from the opposite party. This measure removes any effect coming from behavior by members outside of the specific member s party. The two measures are referred to as the 20% definition and the individual definition, respectively. Unless otherwise noted, the analysis uses the individual definition. The primary difference between the measures is in the intercept, or baseline level of bipartisanship. The proportion of bipartisan bills (20% definition) that the average member cosponsors is 0.58 whereas the proportion of bills sponsored by a member of the other party and cosponsored by a member is The appendix includes a replication of the analysis using the 20% definition. The results are essentially the same with the two definitions, although the amount of variation explained is higher when using the individual definition and the weakening effect of party popularity with time is stronger with the individual definition. The analysis also includes a number of control variables for both individual level characteristics of members as well as institutional constraints on bipartisan cooperation. Individual level characteristics include the normal presidential vote in each member s district (defined as the average presidential vote for the party of the member over the previous two elections), the member s ideology (measured as the absolute value of the first dimension of DW- NOMINATE), how long the member has served in Congress, the member s age, the gender of the member, and whether the member is part of the party leadership. Institutional variables include an indicator for divided government and the size of the majority party. The normal presidential vote is also used to measure marginal and safe districts. 10

12 Finally, I control for economic conditions, as this feature may affect the content of bills proposed and thus the level of bipartisan cooperation. Since I argue that variation in bipartisan cooperation could reflect changes in the relative weight of attention to party and constituency by members, or by changes in agenda content, it is important to seek ways to control for the content of the agenda. It is widely recognized that economic issues are the basis of a primary cleavage between Democrats and Republicans (Lee 2009). Taxation, the size of government, and the resulting decisions about spending and the creation of federal safety net through social welfare policies divide the parties. If poor economic conditions lead to a greater focus on economic issues in Congress, greater partisanship by members might be expected. To control for the impact of economic conditions, via changes in the types of issues on the congressional agenda, this research uses unemployment data and also examines the relationship of unemployment to agenda content, using the coding of bill topics of the bills introduced in Congress. Economic conditions are measured using unemployment data from the U.S. Department of Labor s Bureau of Labor Statistics website. Specifically, the unemployment rate captures the non-seasonally 3 adjusted unemployment rate, where the unemployment rate is the ratio of all persons over sixteen years old who are unemployed to the number of persons over sixteen years old who are in the labor force. Like each of the other measures, a quarterly average of unemployment is used. The distribution of issues on the congressional agenda is measured using Adler and Wilkerson s Congressional Bills Project (2008). Although the major topic codes in these datasets fall into nineteen different categories, I focus on the percentage of the agenda falling into the macroeconomic category (which includes taxation and overall fiscal policy). Defining bipartisanship is instances where at least twenty percent of the cosponsors are from the 3 The non-seasonally adjusted unemployment rate is used because even seasonal unemployment may result in legislative attention to economic issues and the social safety net for these workers. 11

13 party opposite the party of the bill s original sponsor, in nine out of fifteen Congresses bipartisan coalitions occur less than half of the time on macroeconomic issues. As a point of comparison, when all bills are considered, bipartisan cooperation occurs on more than half of all cosponsored bills in twelve of the fifteen Congresses examined. Thus, bipartisanship occurs less frequently on macroeconomic legislation than on all legislation or in comparison with a number of other specific issue areas. This is not to say that these are the only partisan issues, since social welfare, civil rights, labor and education all tend to be partisan areas as well. However, this does suggest that if poor economic conditions lead to increasing congressional attention to economic legislation, higher partisanship might be expected. As a brief aside, let us consider the connection between economic conditions in the country and the content of bills introduced in Congress. As a preliminary test of this argument, I examine the proportion of congressional bills introduced in each of the nineteen major policy areas in Adler and Wilkerson s Congressional Bill Project. Again, the analysis is performed by aggregating for each quarter. Looking at the correlation between the level of unemployment and the proportion of the legislative agenda devoted to specific topics suggests that agenda content may explain why poor economic conditions are associated with low legislative bipartisanship. The correlation between the proportion of macroeconomic issues introduced in Congress and the unemployment rate is relatively high (r = 0.35). 4 This indicates that higher levels of unemployment correspond to a higher proportion of the legislative agenda focusing on macroeconomic issues. The relationship between the proportion of social welfare issues introduced in Congress and the unemployment rate is even higher (r = 0.48), suggesting that 4 The correlation between unemployment and other issue areas ranges from -0.6 to 0.48, with 10 of the 19 major topic areas showing negative correlations, and 9 of the 19 major topic areas producing correlations with a correlation (in absolute value) less than

14 economic problems also lead to more attention to federally run social safety net programs, including food stamps and poverty assistance to low-income families. Returning to the discussion of how party popularity affects legislative behavior, let us turn to the data. Beginning with a multivariate analysis of member-quarter-years, Table 1 presents regression results of legislative bipartisanship as a function of PID advantage, individual, economic and Congress-level controls. The first model uses an OLS specification, where the dependent variable is the percent of cosponsored bills that are bipartisan. All other models use a quasi-binomial specification, which looks at the number of bipartisan cosponsorships relative to the number of partisan cosponsorships and allows the precision of the estimation to vary depending on the number of cosponsored bills. A quasi-binomial, rather than a binomial model, is used in order to allow for over-dispersion. The first two models combine all Congresses together. As expected in the party brand cost-benefit hypothesis, the coefficient on PID advantage is negative. The predicted probability of bipartisanship as a function of PID advantage in the electorate is presented in Figure 2. Holding all other variables at their mean or median, bipartisan cooperation declines nearly 5% across the range of PID advantages in the electorate. When the 20% definition of bipartisan cooperation is used (Appendix Figure 1), the effect is substantially larger, resulting in nearly a twenty percent drop in bipartisan cooperation. In terms of other variables, the effects are largely as expected bipartisanship increases with age, tenure, and divided government and decreases with majority party status, the normal presidential vote in the district, the extremity of ideology, unemployment, and holding leadership positions. The more surprising relationships occur for gender and the size of the majority seat share. Female members engage in less bipartisanship than their male counterparts and the majority seat share has a positive effect on bipartisan cooperation. 13

15 [Table 1 about here] [Figure 2 about here] One concern with the previous analysis is that it combines all Congresses together, when both the baseline level of bipartisanship and the average level of PID advantage may vary by Congress. Thus, the third model of Table 1 includes fixed effects for each Congress. When allowing the mean level of bipartisanship to vary by Congress, the effect of presidential approval remains negative. As noted earlier, it is possible that legislative behavior in Congress also affects party identification in the electorate. As a preliminary test for causation, I include the lagged rather than contemporaneous value of PID advantage for each member. The results, presented in Table 2, are consistent with the earlier results. Party popularity has a negative effect on legislative bipartisanship. All subsequent analyses use the lagged value of PID advantages. To better understand how party popularity affects legislative behavior, let us briefly turn to a consideration of other political factors that might mitigate or even reverse this effect. First, consider whether it is a presidential election year. The anecdotal evidence from 2008 suggests that members may have been particularly sensitive to party popularity as this time. The data do not bring strong support for this argument. Table 3, Model 1 indicates that members are only marginally more responsive to PID advantages during presidential election years than during other time (and when the 20% definition of bipartisanship is used, members are less responsive during presidential election years). [Table 3 about here] Second, consider whether members are in the majority or the minority in the House. Table 3, Model 2 indicates that minority party members are the only group of legislators who are 14

16 responding to party popularity. Figure 3 graphs the predicted probability of bipartisanship as a function of PID advantage for members of the majority and minority parties. Whereas minority party members exhibit a strong negative relationship, majority party members exhibit a slight positive relationship. [Figure 3 about here] Third, consider whether members are more or less responsive to party popularity in the electorate when there is divided government. Including an interaction term between approval and divided government suggests that members are more responsive to PID advantages when government is divided than when it is unified, in addition to the baseline propensity for bipartisan cooperation being higher during divided government (Table 3, Model 3). When the individual definition is included, the effect of PID advantage is slightly positive during unified government. Now let us turn to the marginal member corollary, looking at the interaction between party popularity and seat competitiveness. As mentioned earlier, if members are strategically changing their behavior in response to the popularity of the party, we might expect the effect of presidential approval to be greatest for members in competitive seats. Table 3, Model 4 examines this relationship. When the individual definition of bipartisanship is used, as the normal vote increases (and the district becomes safer for the member s party), the negative effect of party popularity is mitigated, although the substantive size of the interaction is small. However, when the 20% definition of bipartisanship is used, the safeness of the seat actually leads to greater legislative responsiveness to party popularity, but again, the effect is small substantively. Figure 4 presents the predicted probability of bipartisanship for members in competitive and safe seats as a function of PID advantage. Competitive seats are measured as 15

17 those where the normal presidential vote is at the 25 th percentile and safe seats are those where the normal vote is at the 75 th percentile. When the effects of all other variables are held constant across Congresses (Figure 4), members in competitive seats engage in greater bipartisanship and are somewhat more responsive to PID advantages in the electorate than members in safe seats. [Figure 4 about here] Thus far, the evidence suggests that the legislative behavior of members of Congress reflects the popularity of their party in the electorate, although the effect is stronger for minority party members, and during divided government. As expected, members in competitive districts are more responsive to the popularity of the party and they tend to engage in greater bipartisanship on average. A final interaction consideration is whether members have become more or less responsive to party popularity as the parties have polarized by conventional standards. Although bipartisan cooperation has not declined nearly as much in cosponsorship coalitions as in roll call voting as party polarization has increased (Harbridge 2009), it is worth considering whether legislative responses to party popularity have changed as congressional parties have polarized. Since polarization has increased relatively linearly over time over the period of analysis, I use time as a proxy for polarization. Measuring time by each Congress (minus 97 so that the measure runs from 1 to 12), Table 3, Model 5 presents the results of the interaction of lagged PID advantage with time. Figure 5 presents the predicted probability of bipartisan cooperation across the range of PID advantage for each Congress. When the individual measure of bipartisan cooperation is used, we see that the predicted relationship actually changes from negative to positive over the course of time. Although this change is not observed when using the 20% definition (see appendix where effect weakens over time but does not change in direction), the 16

18 results suggest that party polarization (which is linearly related to time) may have changed the way that members of Congress relate to their party. [Figure 5 about here] An alternate approach to examining whether member responses to party popularity have changed over time, as well as to further explore the short-term effect of party popularity, Figure 6 presents the predicted probability of bipartisan cooperation as a function of lagged PID advantage, including an interaction term between PID advantage and each Congress. This model specification allows the average level of bipartisan cooperation and the effect of PID advantage to vary by Congress. The model includes all individual level covariates but drops Congress-level covariates that do not vary within a Congress or across members. Using the individual definition, the negative relationship between PID advantage and bipartisan cooperation exists in seven of the twelve Congresses, although the magnitude of the relationship varies substantially within those Congresses. Using the 20% definition of bipartisan cooperation, the negative relationship appears in nine of the twelve Congresses, with strong negative relationships in the 98 th to 103 rd Congresses, followed by more mixed results in the 104 th to 109 th Congresses. Again, this points to a change in the way that members have responded to party popularity over time, which may reflect effects of party polarization. [Figure 6 about here] As a final set of robustness checks, fixed effects are included either by district or by member (Table 4). In the case of fixed effects by member, the data is split into two periods the 98 th 103 rd Congresses and the 104 th -109 th Congresses in order to examine the change in the relationship between party popularity and legislative bipartisanship over time. When fixed effects by district are included, lagged PID advantage has a significant and negative effect on 17

19 legislative bipartisanship. However, when fixed effects by member are included, lagged PID advantage is negative only in the first half of the data. In the second half of the data, the relationship is positive, consistent with the findings when interacting PID advantage with time. [Table 4 about here] Conclusion Combined, these results suggest that legislative behavior does reflect party popularity and that the effects are strongest for minority party members, during divided government, and among members in competitive seats. However, contrary to the motivating claims of Republicans distancing themselves from the party in 2008, the evidence suggests that the relationship between party popularity and legislative behavior is not stronger during elections and that the relationship has weakened over time. Whether it is growth in the number of safe districts (which has corresponded with increases in polarization), the growth in resources that parties have for their members, or some other element of polarization that explains the change over time remains to be explained. However, thinking about how polarization affects the relationship between members and their parties may help us better understand the response of members to party popularity more generally. In contrast to existing work that has focused on long-term changes in legislative behavior, this paper finds that members behavior changes in the short-term as well, even within a Congress. Although there are a number of factors that may influence behavior, and these factors can push members in competing directions, the evidence presented in this paper supports the party brand cost-benefit story, with members in competitive seats being most responsive to the level of party popularity. Although ideology is certainly a strong predictor of legislative 18

20 behavior, especially between members, variation in bipartisan cooperation over time also reflects responses to the popularity of the party in the electorate, even within members careers. Returning to the broader contributions of this work, these findings highlight the fact that policy coalitions are not fixed, but rather, reflect variation in the strength of competing pressures on members over time as well as changes in the legislative agenda. While parties can be influential in congressional behavior, this work also highlights the limits of the party brand. Finally, this work suggests that low approval of congressional parties may help facilitate electoral responsiveness, drawing members to form coalitions that reflect constituency pressures rather than party interests. 19

21 References Adler, E. Scott, and John Wilkerson "Congressional Bills Project: " NSF and Canes-Wrone, Brandice, David W. Brady, and John F. Cogan "Out of Step, Out of Office: Electoral Accountability and House Members' Voting." American Political Science Review 96 (1): Carson, Jamie L., Gregory Koger, Matthew J. Lebo, and Everett Young "The Electoral Costs of Party Loyalty in Congress." American Journal of Political Science 54 (3): Cox, Gary, and Mathew D. McCubbins Legislative Leviathan. Berkeley: University of California Press Setting the Agenda: Responsible Party Government in the U.S. House of Representatives. New York: Cambridge University Press. Erikson, Robert S "Constituency Opinion and Congressional Behavior: A Reexamination of the Miller-Stokes Representation Data." American Journal of Political Science 22 (3): Erikson, Robert S., and Gerald C. Jr. Wright "Policy Representation of Constituency Interests." Political Behavior 2 (1): Fowler, James. 2006a. "Connecting the Congress: A Study of Cosponsorship Networks." Political Analysis 14: b. "Legislative Cosponsorship Networks in the U.S. House and Senate." Social Networks 28: Grynaviski, Jeffrey D Partisan Bonds. New Yotk: Cambridge Univeristy Press. 20

22 Harbridge, Laurel "Bipartisanship in a Polarized Congress." Stanford University. Harbridge, Laurel, and Neil Malhotra "Electoral Incentives and Partisan Conflict in Congress: Evidence from Survey Experiments." Krehbiel, Keith Pivotal Politics. Chicago: University of Chicago Press. Lebo, Matthew J., Adam J. McGlynn, and Gregory Koger "Strategic Party Government: Party Influence in Congress, " American Journal of Political Science 51 (3): Lee, Frances E Beyond Ideology: Politics, Principles, and Partisanship in the U.S. Senate. Chicago: University of Chicago Press. Mayhew, David R Congress: The Electoral Connection. New Haven: Yale University Press. Patty, John W "Equilibrium Party Government." American Journal of Political Science 52 (3): Pew Willingness to Compromise a Plus in Midterms. The Pew Research Center for People and the Press 2010 [cited May ]. Available from Rohde, David W Parties and Leaders in the Post-Reform House. Chicago: University of Chicago Press. Smith, Gordon Truth 2008 [cited June ]. Available from Smith, Steven S Party Influence in Congress. New York: Cambridge University Press. Zeleny, Jeff "To Help Democrats in the Fall, Obama May Stay Away." The New York Times, August 2. 21

23 Table 1: Legislative Bipartisanship (Individual Measure, Member-Quarter-Year) Model 1 (OLS) Model 2 Model 3 Intercept 54.6 *** ** 1.26 *** (1.39) (0.055) (0.0546) Majority Party *** *** *** (0.176) (0.0072) ( ) Normal Presidential Vote *** *** *** ( ) ( ) ( ) Female *** *** *** (0.258) ( ) ( ) Age *** *** *** ( ) ( ) ( ) Tenure *** *** *** (0.0233) ( ) ( ) Absolute Value DW-NOMINATE -35 *** *** *** (0.546) (0.0219) (0.022) Unemployment Rate *** *** (0.0794) ( ) ( ) Leadership *** *** *** (0.736) (0.0385) (0.0385) PID Advantage *** *** *** (0.0125) ( ) ( ) Divided Government 0.99 *** *** (0.177) ( ) House Majority Seat Share 9.63 *** *** (2.68) (0.107) Congress Fixed Effects No No Yes N R Adjusted R Log Likelihood -1.66x10 5 Standard errors in parentheses. ^p < 0.1, * p < 0.05, ** p < 0.01, *** p <

24 Table 2: Legislative Bipartisanship (Individual Measure, Member-Quarter-Year) as a Function of Lagged Party Popularity Model 1 (OLS) Model 2 Model 3 Intercept 54.2 *** *** 1.27 *** (1.39) (0.0549) (0.0546) Majority Party *** *** *** (0.177) ( ) ( ) Normal Presidential Vote *** *** *** ( ) ( ) ( ) Female *** *** *** (0.258) ( ) ( ) Age *** *** *** ( ) ( ) ( ) Tenure *** *** *** (0.0232) ( ) ( ) Absolute Value DW-NOMINATE *** *** *** (0.546) (0.0219) (0.0219) Unemployment Rate *** *** (0.0793) ( ) ( ) Leadership *** *** *** (0.735) (0.0385) (0.0384) PID Advantage (Lagged) *** *** *** (0.012) ( ) ( ) Divided Government *** *** (0.177) ( ) House Majority Seat Share 9.89 *** *** (2.67) (0.107) Congress Fixed Effects No No Yes N R Adjusted R Log Likelihood -1.66x10 5 Standard errors in parentheses. ^p < 0.1, * p < 0.05, ** p < 0.01, *** p <

25 Table 3: Legislative Bipartisanship (Individual Measure, Member-Quarter-Year) as a Function of Lagged Party Popularity, Including Interaction Terms Model 1 Model 2 Model 3 Model 4 Model 5 Intercept ** *** *** ** *** (0.111) (0.0634) (0.0547) (0.0549) (0.116) Majority Party *** *** *** *** *** (0.0227) ( ) ( ) ( ) ( ) Normal Presidential Vote *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) Female *** *** *** *** *** (0.0136) ( ) (0.009) ( ) ( ) Age *** *** *** *** ( ) ( ) ( ) ( ) ( ) Tenure *** *** *** *** *** (0.0014) ( ) ( ) ( ) ( ) Absolute Value DW-NOMINATE *** *** *** *** *** (0.0336) (0.0219) (0.0219) (0.0219) (0.0219) Unemployment Rate *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) Leadership *** *** *** *** *** (0.0559) (0.0384) (0.0383) (0.0385) (0.0384) PID Advantage (Lagged) *** *** *** *** *** ( ) ( ) (0.0015) ( ) ( ) Presidential Election Year *** (0.0205) Divided Government *** *** *** *** *** (0.0235) ( ) (0.0069) ( ) (0.0108) House Majority Seat Share 1.02 *** *** *** *** (0.19) (0.114) (0.107) (0.108) (0.17) PID Advantage x Presidential Election Year ^ ( ) PID Advantage x Majority *** ( ) PID Advantage x Divided Government *** ( ) PID Advantage x Normal Vote *** (4.55x10-5 ) Time *** (0.0025) PID Advantage x Time *** ( ) N Log Likelihood Standard errors in parentheses. < 0.1, * p < 0.05, ** p < 0.01, *** p <

26 Table 4: Legislative Bipartisanship (Individual Measure, Member-Quarter-Year) as a Function of Lagged Party Popularity, Including Fixed Effects Model 1 Model 2 Model 3 Intercept 0.13 *** *** 2.43 *** (0.09) (0.24) (0.52) Majority Party *** (0.01) Normal Presidential Vote *** ** *** (0.0004) (0.001) (0.001) Female ** (0.01) Age *** (0.0004) Tenure 0.01 *** *** (0.001) (0.005) (0.01) Absolute Value DW-NOMINATE *** *** *** (0.03) (0.23) (0.16) Unemployment Rate *** *** *** (0.003) (0.005) (0.01) Leadership *** *** -0.16^ (0.05) (0.09) (0.09) PID Advantage (Lagged) *** *** * (0.001) (0.001) (0.001) Divided Government 0.09 *** ** (0.01) (0.02) (0.02) House Majority Seat Share 1.14 *** 3.88 *** ** (0.11) (0.46) (0.76) District Fixed Effects Yes No No Member Fixed Effects No Yes Yes Time Period 98 th -109 th Congresses 98 th -103 rd Congresses 104 th th Congresses N Standard errors in parentheses. ^p < 0.1, * p < 0.05, ** p < 0.01, *** p <

27 Figure 1: Time trend of Democratic PID advantage ( , Quarterly) 26

28 Figure 2: Predicted bipartisanship as a function of party popularity (Table 1, Model 2) 27

29 Figure 3: Predicted bipartisanship as a function of party popularity interacted with majority status (Table 3, Model 2) 28

30 Figure 4: Predicted bipartisanship as a function of party popularity interacted with seat competitiveness (Table 3, Model 4) 29

31 Figure 5: Predicted bipartisanship as a function of party popularity interacted with time (Table 3, Model 5) 30

32 Figure 6: Predicted bipartisanship as a function of party popularity interacted with Congress (full model available upon request) 31

33 Appendix Table 1: Legislative Bipartisanship (20% Measure, Member-Quarter-Year) Model 1 (OLS) Model 2 Model 3 Intercept 75.6 *** *** 2.2 *** (1.66) (0.0545) (0.0545) Majority Party *** *** *** (0.21) (0.007) ( ) Normal Presidential Vote *** *** *** (0.0102) ( ) ( ) Female *** *** *** (0.307) ( ) ( ) Age *** *** *** ( ) ( ) ( ) Tenure ^ *** ** (0.0277) ( ) ( ) Absolute Value DW-NOMINATE *** *** *** (0.649) (0.0215) (0.0214) Unemployment Rate *** *** *** (0.0944) ( ) ( ) Leadership *** *** *** (0.875) (0.035) (0.0347) PID Advantage *** *** *** (0.0148) ( ) ( ) Divided Government 2.7 *** *** (0.211) ( ) House Majority Seat Share 13.5 *** *** (3.18) (0.108) Congress Fixed Effects No No Yes N R Adjusted R Log Likelihood -1.73x10 5 Standard errors in parentheses. ^p < 0.1, * p < 0.05, ** p < 0.01, *** p <

34 Appendix Table 2: Legislative Bipartisanship (20% Measure, Member-Quarter-Year) as a Function of Lagged Party Popularity Model 1 (OLS) Model 2 Model 3 Intercept 74.7 *** *** 2.26 *** (1.65) (0.0545) (0.0545) Majority Party *** *** *** (0.211) ( ) ( ) Normal Presidential Vote *** *** *** (0.0102) ( ) ( ) Female *** *** *** (0.306) ( ) ( ) Age *** *** *** ( ) ( ) ( ) Tenure ^ *** ** (0.0276) ( ) ( ) Absolute Value DW-NOMINATE *** *** *** (0.649) (0.0215) (0.0214) Unemployment Rate *** *** *** (0.0942) ( ) ( ) Leadership *** *** *** (0.874) (0.0349) (0.0346) PID Advantage (Lagged) *** *** *** (0.0142) ( ) ( ) Divided Government 2.68 *** *** (0.21) ( ) House Majority Seat Share 14.2 *** *** (3.18) (0.107) Congress Fixed Effects No No Yes N R Adjusted R Log Likelihood -1.72x10 5 Standard errors in parentheses. ^p < 0.1, * p < 0.05, ** p < 0.01, *** p <