An Extended Historical View of Congressional Party Polarization

Transcription

1 n Extended Historical View of Congressional Party Polarization David W. Brady Hahrie Han Stanford University Department of Political Science Encina Hall Stanford, C 405 dbrady@stanford.edu hahrie@stanford.edu Working paper prepared for presentation at Princeton University, Dec. 2, 2004.

2 Introduction The question of why congressional parties have polarized so much since the late 0s is a topic that has received much scholarly attention in recent years. Existing explanations include accounts of endogenous institutional change that have contributed to increased partisanship in Congress, and accounts of exogenous political change that helped alter the configuration of congressional parties. Research focusing on endogenous institutional change argues that the House reforms of the 0s strengthened the power of party leaders, thus making it easier and profitable for parties to act as cohesive voting blocs (Rohde ; Van Houweling 200), and that changes in voting procedures on the House floor contributed to partisan voting behavior (Roberts and Smith 200). The bulk of the research focusing on the role of exogenous political change looks at changing characteristics of congressional districts, including the advent of redistricting (Carson et al. 200; Cox and Katz 2002), the increase in income inequality across districts (McCarty, Poole, and Rosenthal 2002), the change in population (Stonecash, Brewer, and Mariani 2002), and, perhaps most of all, the decline of the one-party South (Rohde, Black and Black 2002; Theriault 200). One interesting feature of this body of research is that it focuses particularly on the sources of congressional party polarization beginning in the late 0s, thus intimating that the recent period of congressional polarization is unique and should be explained as such. This paper offers an alternative perspective on this question that begins by redefining the historical time period over which polarization occurs. We start by examining trends in congressional party polarization throughout the latter half of the nineteenth century and the twentieth century. In both the House and the Senate, the historical data shows that the recent period of polarization is, in fact, a readjustment to patterns of polarization that prevailed throughout most of congressional history. In fact, instead of the recent period of polarization, the bipartisanship of the immediate post-wwii period is the unusual historical period in both the House and the Senate. The mean difference between party medians is at its lowest just after WWII (Schickler 2000) and the polarization that follows in the 0s and 0s is distinct only in comparison to the post-wwii period. 2

3 We argue that the bipartisanship of the 50s sheds light on the re-emerging polarization of the 0s and 0s. In the 50s, parties in the electorate, in Congress, and in presidential elections took similar positions on key issues such as race, national defense, and even the role of government in the immediate post-wwii era as the class-based, New Deal era partisanship faded. Partisan issue distinctions between the two major parties blurred, and did not re-emerge until the polarizing issues of the 60s and 0s (such as race) redefined partisan lines. Interestingly, this re-emergence of partisanship in the 60s became clear in the electorate and in presidential elections as voters and presidential candidates sorted themselves into parties, but did not appear in Congress until the late 0s and 0s. Our interpretation of polarization focuses on this lag in congressional party polarization. Why did congressional party polarization take so long to re-emerge? We examine the rise of cross-pressured members of Congress, who were caught between their constituencies and their parties. s the national parties began to re-polarize in the 60s, Democrats with conservative constituencies and Republicans with liberal constituencies were caught in the middle. Many of these cross-pressured members managed to stay in office by creating a personal vote, which insulated them from the re-appearance of polarizing issues in the 60s. Members of Congress could stay in office with the personal vote even if the national party no longer represented the dominant ideological preferences of the district. It was not until these outof-step legislators were replaced by members of the opposite party who better aligned party and district preference that partisan lines in Congress became redefined. The paper proceeds as follows. The first section examines the implications of looking at historical patterns of party polarization in both the House and the Senate. We show that polarization in the late twentieth century is essentially a return to historical patterns of polarization that have prevailed for most of U.S. history. Instead, it is the immediate post-wwii era that is unusual because of its high levels of bi-partisanship. Thus, we examine some of the unique historical features of this time period, including the decline of New Deal s class-based politics and the rise of the personal vote in congressional elections. The second section examines the return to polarization after the bi-partisan era of the 50s. We argue that both the House and the Senate lagged behind the electorate and presidential elections in

4 returning to historically prevalent levels of partisanship. Thus, we redefine the question of polarization to ask why congressional voting took so long to re-polarize. The third section looks closely at the role of cross-pressuring in explaining this lag in congressional party polarization. We argue that the personal vote insulated members of Congress from emerging national partisan tides in the 60s. It was not until these members retired or were electorally defeated that polarization emerged in Congress. Understanding these patterns of delayed electoral replacement is crucial to understanding the emergence of polarization in Congress in the late twentieth century. Part I: Identifying Historical Patterns of Polarization Most research on congressional party polarization begins by trying to explain the rise in partisanship in the final decades of the twentieth century. This approach ignores historical patterns of polarization in Congress that can provide important insights into understanding polarization of the modern era. This section elucidates those historical patterns, and identifies the unusual period of bipartisanship around the 50s. Then, we explore the unique features of the immediate post-wwii era to better understand the subsequent rise in partisanship in the 0s and 0s. We find that the return to polarization in both the House and the Senate lagged behind the return to polarization in the electorate and in presidential elections. This leads us to redefine the question of polarization in the twentieth century to include a broader historical view. Historical Polarization in Congress Throughout most of United States congressional history, parties in the House and Senate have been relatively polarized. Figure plots the difference between Republican and Democratic party medians, as measured by first-dimension DW-Nominate scores in both the House and the Senate. In both chambers, party polarization peaked in 5 and plummeted to its lowest level in 4 and the early This graph is similar to one found in Shickler

5 50s. 2 The return to polarization in the final decades of the twentieth century does not look as unusual when placed in this historical context. The party differences during this era are lower than they were in the late th century. Instead, what becomes clear is that the era of bi-partisanship in the immediate post- WWII era is the unusual era in merican history. INSERT FIGURE BOUT HERE The unique features of partisanship in the 50s are highlighted by an alternative measure of polarization that examines the degree of overlap between the two parties. Parties can either be polarized, with high levels of internal cohesion and low levels of intra-party overlap, or they can be convergent, with low levels of internal cohesion (wider distributions) and high levels of intra-party overlap. Figure 2 presents two hypothetical models of partisanship. In the top figure, the two parties are relatively polarized, with very little overlap between them; there are very few Republicans who overlap with the most conservative Democrats and vice versa. In contrast, the bottom figure depicts considerable overlap between the two parties. lthough the most liberal Democrats and the most conservative Republicans remain distinct from each other, partisanship in the middle of the two-party distribution is indistinct. The conservative Democrats look quite similar to the liberal Republicans, and the liberal Republicans look like the conservative Democrats. INSERT FIGURE 2 BOUT HERE We used this model to explore the degree of congressional party overlap over time. We used two different measures of ideology: first-dimension DW-Nominate scores (6-), and mericans for 2 It is worth noting that the difference between party medians in the Senate was consistently lower than the House from the late 00s to the early 50s. In addition, a relatively large decline in the difference between party medians in the Senate occurs in the early 0s, and persists through the 50s. See footnote #4 for on this topic. 5

6 Democratic ction (D) scores (4-). Both ranking systems rate the liberalism (or conservatism) of elected officials based on their roll call voting records. Using these scores, we could identify how liberal or conservative members were relative to other members of their party. We identified the cut points for the 0%, 25%, and 50% (the median value) most conservative Democrats, and the 0%, 25%, and 50% most liberal Republicans. The bottom of Figure 2 shows the hypothetical placement of the cut point for the 0% most liberal Republican on a DW-Nominate scaling. In this case, the most conservative Democrats are conservative (with higher ideological rankings) than the 0% most liberal Republicans. The shaded area shows the area of overla p. By thus by identifying these cut points, we can observe the number of members from the opposite party who fall into the overlap region. In other words, how many Democrats are conservative than the 0% most liberal Republican? nd how many Republicans are liberal than the 0% most conservative Democrats? s an example, in 4 (0 th Congress), the most conservative 0% of Democrats had DW-Nominate scores higher than 0.0. To identify the degree of overlap with Republicans, we counted the number of Republicans who had DW-Nominate scores lower than 0.0. Figure shows the degree of overlap in each Congress from 6-. For the purposes of brevity, we only show the distribution using DW-Nominate scores. INSERT FIGURE BOUT HERE These graphs demonstrate an unprecedented level of overlapping voting in both the House and the Senate in the immediate post-wwii era. Looking at the distributions of DW-Nominate scores, it becomes clear that there was almost no House party overlap prior to the 40s. In the Senate, there was some overlap in the 20s and the 0s, but it was relatively mild compared with the immediate post lthough we do not include the graphs showing D overlap, the results are the same. The only difference is that D scores do not begin until 4 and they show a greater degree of non-south overlap among Democrats in the 50s and 60s than DW-Nominate scores. 6

7 WWII era. 4 By the 40s and 50s in both the House and the Senate, the degree of partisan overlap spiked upwards. By 4, almost 45% of House Democrats were conservative than the 0% most liberal Republicans. The numbers peaked around 6, when over 55% of House Democrats were conservative than the 0% most liberal House Republicans, and almost a third of Democrats were conservative than the 25% most liberal Republicans. 0% of House Democrats were conservative than even the median Republican member of the House. mong Republicans, levels of overlap grew sharply between 4 and 55, and persisted at high levels until the early 0s when the number of members in the overlap region began to decline. In the Senate, the numbers were highest in the late 60s. In 6, % of Democrats were conservative than the 0% most liberal Republicans, and % of Republicans were liberal than the 0% most conservative Democrats. This level of overlap persisted through the late 0s, when it began to decline in both chambers, and lasted in weaker form through the 0s. Like Roberts and Smith (200), we find that for both parties in the House, polarization (or low levels of partisan overlap) re-emerges in the 0s. We performed the same analysis looking only at non-southern states to see if the partisan overlap was merely an artifact of one-party politics dominant in the South before the Civil Rights Movement. We find that although the degree of overlap decreases among Democrats, the mid-twentieth century still emerges as a unique period of high partisan overlap. mong Republicans, we also find that high levels of partisan overlap persist even when we take the Southern states out of the analysis. Two important points emerge from this examination of historical patterns of polarization in Congress. First, the patterns of polarization in the House and the Senate have been markedly similar throughout most of history. This is true using several other measures of partisanship, including party 4 Like the decline in the difference between party medians, partisan overlap appears in the Senate before it appears in the House. We are currently gathering and analyzing some data to better understand this phenomenon. We hypothesize, however, that patterns of inter-state migration may have had a similar (but earlier) impact on the Senate that redistricting had on the House. In the House, after the 60s, states often created congressional districts that were highly partisan by drawing district lines around partisan neighborhoods. In the Senate, although redistricting was not an issue, the migration of certain groups of people into a state had the similar effect of changing the constituency base of the member of Congress. Some data show that patterns of northern migration into the South around the 0s-50s could have affected the rise of Southern conservatism (Polsby 2004), and thus facilitated the growth of partisan overlap in the Senate before the House.

8 voting scores (e.g. Rohde ; Hurley and Wilson ) and party unity scores (e.g. Brady, Cooper, and Hurley ). By all measures, historical trends in the Senate are quite similar to those in the House. These parallel trends in the House and the Senate have important implications for our understanding of the sources of congressional party polarization. s previously noted, existing explanations for polarization in the House can broadly be divided into two categories: sources of endogenous institutional change, and sources of exogenous political change. The endogenous institutional changes cited as sources of polarization in the House, however for example, the 0 s reforms strengthening the power of parties, the Subcommittee Bill of Rights (), and the Committee Reform mendments (4) did not happen simultaneously in the Senate. 5 This implies that exogenous political changes had an impact on levels of polarization in Congress, influencing both chambers simultaneously. In other words, looking solely at the endogenous institutional changes in the House, or focusing on explanations like redistricting that do not occur in both chambers does not capture the full story. Second, this examination of historical patterns in the House reveals the importance of understanding the immediate post-wwii era to better understand polarization in the latter decades of the twentieth century. By understanding the sources of the unusual decline in partisanship after WWII, we can better understand the sources of a return to polarization in the 0s and 0s. Previous scholars have recognized the unique levels of bi-partisanship in the immediate post-wwii era (e.g. Brady, Cooper, and Hurley ; Schickler 2000; Jacobson 2000; 200; Poole and Rosenthal 4), but none of the authors have sought to link the decline in the immediate post-wwii era to the subsequent rise of polarization in the late 0s and 0s. Instead, most research has sought to explain the final decades of the twentieth century as the unique period. Because the 50s were the unusual period in the history of polarization in Congress, however, scholarship on the sources of polarization in the twentieth century should explain both the decline in polarization after WWII, and the subsequent return to polarization in 5 lthough the Senate experienced some rules changes in the early 0s, they did not have nearly the same effect as changes in the House (Sinclair ; ; Rohde, p. ; Shepsle ). In her analysis of committee assignments in the Senate, Sinclair () notes that, The [redistributive provisions of the 0 Reorganization ct] were modest in design and even modest in impact (2). In other words, the impact of the 0 Reorganization ct in the Senate was relatively limited, not greatly altering the composition of Senate committees.

9 the 0s and 0s. Thus, we turn next to an exploration of the unique political characteristics of this time period. Features of the Immediate Post-WWII Era To better understand the high degree of partisan overlap in the immediate post-wwii era, we discuss two important historical trends. First, during the 50s, the two-party system was characterized by broad agreement between Democrats and Republicans on key national issues, such as race, national security, and the role of government in society. Second, after WWII, congressional elections became independent, as members of Congress learned to win office independent of party. This section discusses each of those features in turn. During the immediate post-wwii era, voters demonstrated considerable overlap between the two parties for the first time, as the party system shifted away from the New Deal parties preceding the war. From 0 to the mid 40s, political parties divided clearly over economic issues and the extent of merican involvement in Europe and the world. Parties in this period were clearly defined: most voters voted for the same party at the Presidential and congressional levels (Burden and Kimball 2002), party platforms exhibited distinct positions on economic and international involvement issues (Ginsberg 2), and elected officials from each party demonstrated hardly any ideological overlap with each other. Socioeconomic class was particularly potent in defining partisanship. The People s Choice, a study of voters in Sandusky, OH in 40, found that being in a low-income group, a Catholic, or an urban resident, all predisposed a voter toward the Democrats; being in a high-income group, a Protestant, or a rural resident predisposed a voter toward the Republicans (Lazarsfeld 60). Robert R. lford (64) used Michigan Survey Research Center data and Gallup and Roper polls to analyze class voting from 6 to 60. His findings show that from 6 to the 50s, voting differences between classes were consistently high. Berelson, Lazarsfeld, and McPhee (54) found in their 4 study of Elmira that the swing to Truman which won him the election over Dewey came from those to whom class issues were salient. Judson James (6) concluded that This fifth party system not only reversed the majority/minority party role s

10 of the Republicans and Democrats, but also nearly than any previous two-party system, it had a basis in class conflict (45). The New Deal system that characterized parties before and through WWII, in other words, was undoubtedly defined in the electorate and in Congress along class lines. fter WWII, however, the differences characterizing Democrats and Republicans in the New Deal system began to disappear. The 4 election was the last election in the class-based New Deal party system. V.O. Key writes, Perhaps in the election of 6 the party division most nearly coincided with differences of income and occupation. That coincidence declined as class relevant questions faded from the forefront, and in 52 and 56, Republicans won substantia l support in the lower-income groups. (Key 5, 24) Likewise, Campbell, Converse, Miller and Stokes (60) computed an index of state polarization and found that the correlation between occupational status and the partisan vote fell over the 4-56 period. Socio-economic class, in other words, declined as the defining feature of political parties in the 50s. Interestingly, however, as class divisions between parties became less clear, a new foundation for defining the parties did not emerge. Instead, during the 50s, we see considerable overlap between the two parties on key national issues. The election of Eisenhower, a popular war-time hero, inspired significant levels of cross-party voting during the 50s, and levels of differentiation between Democrats and Republicans began to erode. On key issues, such as national security, Democrats and Republicans were quite similar: Democrats like Kennedy matched Republicans in their Cold War stances. On issues related to Civil Rights, the Southern wing of the Democratic party kept the party from adopting civil rights stances until 64. Thus, until the early 60s, it was not clear if Republicans or Democrats would be supportive of civil rights. Similarly, as the class-based New Deal system of parties began to disappear, differences between Democrats and Republicans on issues related to the role of government weakened. To examine these historical trends, we looked at a set of issues over the period to determine the timing of change in the electorate. We use the National Election Studies (NES) crosssectional studies from 4 to Because questions on issue opinions were not asked consistently 0

11 over time, we went through each cross-sectional study and created an opinion index for each respondent on three key dimensions. In creating the scales, we followed methodology established in Stimson (). These issues are:. National Defense: This index encompasses respondents views on U.S. policy towards Communism and Communist countries, and levels of military power or defense spending. The scale runs from to, with indicating a hawkish position, and indicating a dovish position. 0 implies neutrality. 2. Race/Civil Rights: This index captures respondents views on issues regarding race and civil rights. This includes views on the Civil Rights movement, school integration, busing, and the federal government s role in supporting minorities. The scale ranges from - to, with implying less government intervention, and implying greater government intervention. 0 implies status quo.. Role of Government: This index includes questions about the role of government in providing social welfare. It encompasses questions about government s role in supporting full employment, health care, housing and utilities, social security, food stamps, and other government services. This scale also runs from to, with indicating less government, and indicating government. 0 is status quo. We recoded any question in the NES cross-sections that pertained to one of the three issues above (see ppendix for the questions included in each scale). We then created a composite score for each individual that represented the mean of her answers on the to scale. We could then aggregate these individual scores to create mean scores for Republicans versus Democrats in each presidential election year. Figure 4 shows the mean differences between parties over time on issues related to race, defense, and role of government. INSERT FIGURE 4 BOUT HERE

12 These graphs reveal a basic historical pattern: the voters converged on key issues in the 50s, but patterns of polarization began to reemerge in the 60s. 6 On defense issues, the differences between the parties were zero in the 50s, and only began to rise thereafter. On race issues, the difference between voters from each party hovers around zero and only begins to rise in 64. On issues related to the role of government, the differences between parties fluctuate, but we nevertheless witness a relative convergence between voters from each party in the mid-50s, and a subsequent divergence. In other words, it becomes clear that on these three major issues of the immediate post-wwii period, the electorate was not divided along clear partisan lines. Issues which had previously determined the partisan basis of the electorate were fading in importance as cues to mass voting behavior. s levels of partisanship in the electorate and in Congress declined in the 50s, members of Congress learned to become independent of party in their attempts to seek re-election. The rise of the personal vote is well-documented in political science scholarship through studies of the incumbency advantage (lford and Brady ; Erikson 2; nsolabehere, Snyder, and Stewart III 2000; Gelman and King 0; Jacobson ; Cain, Ferejohn, and Fiorina ). Beginning in the 50s, the incumbency advantage began to rise and increased or less consistently until its peak in, when incumbents had a 2% electoral advantage over non-incumbents. One commonly used method of measuring this is the slurge, or the mean value of the sopho surge and the retirement slump. Measures of slurge over time show that the sharpest rise in the incumbency advantage came in the late 50s and 60s. Simultaneously, the percentage of districts with split partisan results at the presidential and congressional levels moves from zero at the start of the 20 th century, to a dramatic rise between 4 and 2 (Burden and Kimball 2002). With respect to the election of Eisenhower in 52 and 56, the level of split ticket voting was understandable as voters from both parties chose a popular war hero for 6 lthough a precise comparison is not possible, we find that these results are consistent with Stimson s () results. See Stimson, figures 4.4 and 4.5. The sopho surge is calculated as the difference between the winning vote share of candidates running as incumbents for the first time, versus their winning vote share in the previous election. The retirement slump is the mean decrease in the party s vote from an election in which an incumbent was running, to an election in which the incumbent retires and the seat is open. 2

13 President. However, in the 60s, as issues related to race, the Vietnam War, and later the environment emerged as dividing issues, voters still split their tickets. Both of these patterns the rise in the incumbency advantage and the rise in split districts demonstrate the increasing tendency of Congressional elections to be subject to forces other than partisanship. Research shows that there is, in essence, a trade-off between the effect of the personal vote (as represented by the incumbency advantage) and the effect of partisanship in determining electoral outcomes (nsolabehere and Snyder 2002). Instead of relying on parties to secure re-election, members of Congress in the post-wwii began to rely on other factors. Throughout the immediate post-wwii era, then, partisan lines blurred in both Congress and the electorate. Levels of bi-partisanship were higher than ever before in the House and the Senate, and the New Deal system of parties began to fade in the electorate. Democrats and Republican were not very differentiated on key national issues like national security, civil rights, and the role of government. Members of Congress, in turn, learned to win office on their own, relying less on party and on other independent factors. What changed? How did this system give way to the partisanship of the latter decades of the twentieth century? Part II: The Historical Lag in Congressional Party Polarization Following the unique bi-partisanship of the immediate post-wwii era, partisan lines began to reappear in the electorate and in presidential elections throughout the 60s. The social movements of the 60s, striking demographic changes, and polarized presidential elections prompted a redefinition of partisan lines. Both the House and the Senate, however, were slow to respond to these changes. s shown in Figures and, polarization in Congress did not reappear until the late 0s or early 0s. This section discusses this lag in the re-emergence of congressional party polarization. rich body of research has explored the sources of the personal vote. See, for example, Cain, Bruce E., John Ferejohn, and Morris P. Fiorina.. The Personal Vote: Constituency Service and Electoral Independence. Cambridge: Harvard University Press, Fiorina, Morris P.. Congress: Keystone of the Washington Establishment. New Haven: Yale University Press.

14 In addition to plotting the mean differences between parties on three key issues, Figure 4 also shows the trajectory of polarization in Congressional voting, as measured by the difference in DW- Nominate scores by party. It is clear that on defense issues, the partisan differences in the electorate were growing in the early 60s and again in the 0s, but remained relatively constant in Congress throughout the 60s and 0s. On race issues, partisanship in the electorate grows steadily starting in 64, but it is not until 2 that congressional voting begins to polarize. On the issues related to the role of government, partisan differences in the electorate rise rapidly in the early 60s and stay consistently high (with the exception of 2), even though polarization in Congress does not appear until the 0s. In each case, it appears that voters were using party to sort out policy choices long before aggregate levels of congressional voting reflected such polarization. To understand the re-emergence of polarization after the 50s, we start by examining existing explanations for House party polarization in the current literature. s previously noted, the parallel trends in polarization in the House and the Senate intimate that explanations focusing solely on endogenous institutional change may not tell the whole story. The key exogenous political changes that scholars cite as sources of polarization in the House are: () the impact of redistricting on creating polarized congressional districts (Carson et al. 200; Cox and Katz 2002), (2) the impact of changing populations within the districts (Stonecash, Brewer, and Mariani 2002), () the impact of changing levels of income inequality (McCarty, Poole, and Rosenthal 2002), and (4) the impact of political change in the South (Rohde ; Black and Black 2002). Clearly, we can rule out redistricting as an explanation for polarization in the Senate, since reapportionment of congressional districts has no impact on state boundaries. The remaining three explanations for increased polarization make the same basic point: some exogenous political shock altered the electoral bases of the parties, creating greater intra-party homogeneity and greater inter-party heterogeneity. The change in these electoral coalitions prompted greater congressional polarization. The key exogenous shocks, however changing trends in income inequality, changing demographics, and change in the one-party South occurred primarily in the mid-60s. In developing 4

15 their arguments about the changing constituency bases of the parties, McCarty, Poole, and Rosenthal (2002), Stonecash, Brewer and Mariani (2002), Rohde (), and Black and Black (2002) find some common ground in a set of exogenous shocks occurring in the 60s that led to a reconfiguration of constituency bases of the parties. Perhaps most prominently, the adoption of the 64 Civil Rights ct, and the selection of Barry Goldwater as the Republican nominee in the 64 presidential election marked clear divergence of the parties on civil rights. The subsequent Voting Rights ct in 65 sparked the enfranchising of black voters throughout the South. lso in 65, Congress passed the Immigration and Nationality ct mendments of 65, reinforcing increasing levels of legal immigration to the United States that had begun to rise in the mid 40s. Finally, beginning in the 60s and particularly in the 0s, economic inequality began to rise (Piketty and Saez 200; Stonecash, Brewer, and Mariani 2002, 64-4). s noted by multiple authors, the 60s were a tumultuous time of much social and political change that redefined the political parties. Despite the tumult of the 60s, House and Senate party polarization did not begin to appear until the late 0s; this is true even though presidential elections and partisanship in the electorate reflected the changes. Many authors, including Stonecash, Brewer and Mariani (2002) and Black and Black (2002), note that despite the changes in the 60s, the demographic changes take some time to take effect. We agree that this is the case, but argue that there were also institutional characteristics that insulated congressional parties from exogenous changes and slowed the pattern of change. In fact, looking at levels of black voter registration, it is clear that the key changes were in the 60s: in 60, black voters comprised 0.6% of the pool of registered voters in the South, and by 6 that number had jumped to 6.5%, a number that held relatively constant for the next twenty years until (Stonecash, Brewer, and Mariani 2002, 55). The demographic changes, thus, took hold and congressional parties still lagged behind the change. This point also comes from a talk by Keith Poole, Economic Inequality and Political Polarization, pril 22, 2004, Stanford, C. 5

16 Presidential elections did not follow the same patterns as Congress and, instead, paralleled changes in the electorate. The highly bi-partisan elections of Eisenhower in the 50s gave way to the turbulent electoral processes of the 60s. The Johnson-Goldwater elections of 64 and the Nixon- Humphrey elections of 6 were both divisive, prompting party identification to become a better predictor of vote choice. fter the 6 convention, in fact, liberal activists in the Democratic party garnered sufficient power to pressure the party into reforming the nomination process (Polsby and Wildavsky 0). These reforms, which eventually occurred in the Republican Party as well, forced presidential candidates to cater to extreme primary voters. To win a party s nomination, in other words, presidential candidates had to win the support of a narrow band of primary voters. series of studies, however, have shown that these party activists are often not representative of the larger party s interests (Jackson, Brown, and Bositis 2; McClosky, Hoffman, and O'Hara 60; Nexon ; Verba and Nie 2; Soule and McGrath 5; Kirkpatrick 6). 0 Presidential politics, thus, began to evince clear patterns of divergence between parties, particularly among party elites. Despite these changes in the electorate and in presidential elections, congressional voting remained surprisingly immune from this increase in partisanship. To examine this assertion, we performed a logit regression of the probability of voting Democratic in presidential and congressional elections on the issue indexes of electorate opinions on race, role of government, and defense used earlier in the paper. The model is as follows: Where, Pr(D it =) = Φ(α + β Role it + β 2 Race it + β Defense it + error it ) D it = Probability of voter i voting Democrat in the election t Role it = Voter i s score on the role of government index in election t Race it = Voter i s score on the race index in election t Defense it = Voter i s score on the defense/national security index in election t 0 Other studies have found that presidential nomination activists take electability of the candidate into account as much as ideological congruence (see esp. Stone and bramowitz ; Stone, Rappoport, bramowitz 2). lthough the primary voters may weight general election electability in making their vote choices, however, this does not preclude the fact that the new nominating system forces candidates to appeal to a group that is not the same group they must appeal to in the general election. 6

17 If our assertion is that congressional elections remained immune from the resurgence of partisanship in the 60s, then predicted probabilities from this regression should show that voters congressional vote choice should remain less partisan throughout the 60s, thus lagging behind presidential vote choice. Figure 5 graphs the predicted probabilities from this regression. INSERT FIGURE 5 BOUT HERE In calculating the predicted probabilities, the issue indexes are set to ; in other words, this hypothetical voter consistently holds conservative views on issues related to race, defense, and the role of government. The predicted probabilities represent the percentages of conservative voters that are likely to vote Democratic. The figure shows that while presidential and congressional elections paralleled each other until the early 60s, they began to diverge thereafter. In the first two post-wwii elections, conservatives were quite likely to vote Democrat for President and Congress, from 2.2% in the 52 House election, and 25.% in the Senate election. In fact, in 52, conservatives are predicted to vote for Stevenson than for Democratic House and Senate candidates. fter 60, however, the percentage of conservative voters predicted to vote for a Democratic presidential candidate never reaches 20% and only in 64, 6, and 6 does it reach double digits. In 64, 2 and from 0 onward, the number of conservatives voting Democratic for President is predicted to be below 0% and averages less than 5%. In contrast, conservative votes for House candidates is predicted to be.5% in 56 and is over 20% from 60-4 with the average being over 2%. Not until the 4 election does the percent conservative voters predicted to vote Democratic in House elections fall to 2.%; and through 6, it averages about %. Similarly, in the Senate, from 60-0, the percent of conservative voters voting for a Democratic Senator is at least 20% and usually is higher. It is not until 0 that the percent of conservatives voting for a Democratic Senate candidate begins to drop. See ppendix 2 for the summary statistics from the regression.

18 This lag in congressional party polarization adds another dimension to understanding the rise of polarization in the late 0s and 0s. The data shows that congressional party polarization took substantially longer to re-emerge than polarization in the electorate and at the presidential level. While presidential candidates and voters re-sorted themselves into parties around the polarizing issues of the 60s, members of both the House and the Senate lagged behind. Congressional elections thus differentiated themselves from presidential elections, and after the 50s, partisanship in the Congress was distinct from partisanship in the electorate. This casts the rise in congressional polarization in the 0s and 0s in a new light: it was a step behind the return to polarization in presidential elections and in the electorate. Part III: Why Did Congressional Party Polarization Take So Long to Emerge? The previous section examined the timing of historical change in polarization and revealed a lag in congressional party polarization. Despite tumultuous social and political changes in the 60s, congressional party polarization did not emerge until the late 0s and early 0s. This presents us with an altered historical perspective. Instead of asking why congressional parties polarized in recent decades, we ask why parties in Congress took so long to re-polarize despite changing trends in the electorate and at the presidential level. Our approach to understanding this question focuses on the rise of members of Congress cross-pressured between their district and their national party. In this section, we start by discussing the role of cross-pressuring in sustaining levels of bi-partisanship in Congress beyond the 50s, then demonstrate how polarization in Congress re-emerged as members who had been crosspressured between their geographic constituencies and their party retired or were electorally defeated. The Rise of Cross-Pressured Legislators s the national parties grew increasingly distinct throughout the 60s, we argue that the divergence of congressional elections cross-pressured a group of legislators. These legislators were pulled in one direction by their constituents, and in another direction by their national party. For example,

19 a Border State Democrat like Jim Jones of Oklahoma was pulled in a conservative direction by his district, but in a liberal direction by the Democratic party. Likewise, Northeastern Republicans during the 60s were pulled in a conservative direction by their national party, but in liberal direction by their districts. Pulled in two opposite directions by their party and their constituency, these members had a strategic dilemma: how could they balance the countervailing pressures they faced while still winning re-election? Caught between their parties and their constituencies, we argue that these cross-pressured members had several options that range from being very partisan to non-partisan. First, the most partisan response that members can have to cross-pressuring is to simply switch parties. We count this as a partisan response because members who are cross-pressured and switch parties subsequently reduce the dissonance between their parties and their constituencies, thus enabling them to comfortably vote with the national party. For instance, Democrats from conservative districts can simply switch to the Republican party as the national Democratic party becomes too liberal. The second (middle) option is for the members to try to balance the cross-pressures between their party and their district. This could lead to cross-pressured members being slightly out of step with both their parties and their constituencies. For example, cross-pressured Republicans who are trying to balance the cross-pressures will be too conservative for their constituencies, but too liberal for their parties (and vice versa for Democrats). This balancing can encompass a range of behaviors, ranging from members Gillis Long (D-L) who voted sometimes with his own party, and other times with the Republicans, to members like Phil Gramm (D- TX) who voted consistently with the Reagan Republicans. Members like Gramm represent the third option, which is the least partisan response that members can have: they can simply vote with the opposite party. We count this as the least partisan response because these members are bucking the pressures of their national party to vote consistently with the preferences of their constituents. Thus, by our measure of partisan overlap outlined in the second section of this paper, we would expect that crosspressured members are likely to fall in the overlap region. Finally, there is a fourth option that does

20 not lie on the continuum of most partisan to least partisan responses. This fourth option is for members to simply leave office through retirement or electoral loss. The middle option balancing is particularly interesting because these members had to find ways to compensate for being slightly out-of-step with both their parties and their constituencies. We hypothesize that the personal vote played a critical role in helping these members to stay in office. 2 Because of the growing levels of the personal vote during the 60s, a disproportionately high subset of these legislators could stay in office despite being out of step with their party and their constituencies. In other words, by insulating members of Congress from rising partisan tides in the 60s, the personal vote helped cross-pressured members who were out-of-step with the changing partisan tides of the 60s stay in office. This is also true for cross-pressured members who voted with the opposite party the source of their personal vote could have been their issue-based alignment with their constituency. In both cases, however, the personal vote played an important role in shielding cross-pressured members from the polarizing trends of the 60s. Cross-pressuring and the personal vote, thus could have helped delay the re-emergence of polarization in Congress. Based on our argument, then, we expect that members who are cross-pressured have higher personal vote scores than members who are not cross-pressured, since they have to rely on the personal vote to help them win re-election. Similarly, we expect that rates of switching parties and voting with the other party are higher for cross-pressured members. To examine this, we identify the members who were cross-pressured. We identified Democrats who had conservative constituencies and Republicans who had liberal constituencies. Consistent with previous work, we measured constituency liberalism by looking at three-year moving averages of presidential vote in the district or state (Canes- Wrone, Brady, and Cogan 2002; nsolabehere, Snyder, and Stewart III 200; Erikson and Wright ). 2 By personal vote, we are referring to the portion of the candidate s vote margin that is attributed to factors personal to the candidate, not the candidate s party. Thus, a candidate can generate a personal vote from a variety of sources including constituency services, pork barrel politics, or issue-based alignment with the district but the sources of the candidate s personal vote are always located in the candidate s district. This independent relationship that congressional candidates had with their geographic constituencies played a crucial role in insulating members of Congress from rising partisan tides in the 60s. 20

21 Liberal constituencies were identified as areas that had voted than 55% Democratic. Conservative constituencies were identified as areas voting than 55% Republican. Cross-pressured Democrats, thus, were Democratic members who came from districts or states voting than 55% Republican at the presidential level. Conversely, cross-pressured Republicans were Republican members who came from districts or states voting than 55% Democratic at the presidential level. The data is largely supportive of our argument. First, looking at party-switchers, we find that there are sixteen cases of members who switch parties in the House over this time period, and three cases of members who switch parties in the Senate (see Nokken 2000 for discussion of these cases). mong the House and Senate members who switched parties, most are cross-pressured. However, it is clear that changing parties is not a frequent occurrence. Instead, cross-pressured members tend to try to balance. To look closely at members who try to balance the cross-pressures between their constituencies and their parties, and members who opt to vote with the opposite party, we display the results in Tables and 2. Table presents the results for the House and Table 2 presents the results for the Senate. The results are shown for Republicans, Southern Democrats, and non-southern Democrats. INSERT TBLES and 2 BOUT HERE The top line in Tables and 2 shows the mean DW-Nominate scores for members who are crosspressured and members who are not cross-pressured. If our argument that these members contributed to persistently low levels of partisan polarization in Congress is correct, then we would expect that crosspressured members are ideologically moderate than their partisan counterparts not in the overlap region. The results show that for both the House and the Senate, across both parties and both Southern and non-southern Democrats, our expectations are met. mong Democrats, the cross-pressured members are less liberal, and among Republicans, the cross-pressured members are less conservative. It is possible that there are members who are cross-pressured because their constituencies consistently vote with the opposite party at the presidential level, but do not meet our criteria. We use this standard of measuring cross-pressuring, however, because it is a conservative test. 2

22 The second part of Tables and 2 looks specifically at members who seek to balance the crosspressures between their constituency and their party. We look here at the cross-pressured members who stayed in office and compare them to non-cross-pressured members who stayed in office. We expect that these cross-pressured members have a harder time securing re-election since they are less in line with their national parties and their constituencies. Thus, we expect that they will have lower winning vote margins. In addition, to help themselves maintain office, we expect that cross-pressured members will build up a larger personal vote than members who are not cross-pressured. We examine slurge scores and hypothesize that they should be larger for cross-pressured members. Looking at the results in Tables and 2, it is clear that our hypotheses are supported. In all cases, cross-pressured members are winning with lower vote margins, even though they have higher slurge scores than their non-cross-pressured counterparts. The only exception to this is non-southern Democratic Senators, whose average winning vote margin is equal between cross-pressured and non-cross-pressured members. Even for these Senators, however, the slurge score for cross-pressured members is higher than the slurge score for noncross-pressured members. The third part of Tables and 2 looks at the probability that members will vote with the opposite party. We measure voting with the opposite party using our measure of partisan overlap outlined in the second section of this paper. Here, we expect that cross-pressured members should fall into the overlap region far than non-cross-pressured members. gain, looking at the results in Tables and 2, it is clear that cross-pressured members are in the overlap region at higher rates than non-cross-pressured members. This look at the implications of being cross-pressured in the House and the Senate makes it clear that cross-pressured members were ideologically moderate than their non-cross-pressured counterparts. In addition, they consistently had a larger personal vote than non-cross-pressured members. This is consistent with our argument that cross-pressured members contributed to sustained levels of bipartisanship in Congress throughout the 60s and 0s. Because the personal vote insulated crosspressured members from the changing national partisan tides in the 60s and 0s, these members 22