Abstract
The standard explanation for increasing party polarization in Congress is based on factors that would affect all policy issues simultaneously. We show that this has not happened. We examine the dimensionality of legislative choice in the House of Representatives, scaling eighteen issues in each Congress from 1965 through 2004. We detect considerable variability in issue dimensionality, an evolution in the structure of choice over time, and changes in the relationship between party unity and issue dimensionality. Our findings suggest that polarization has occurred on an issue-by-issue basis, reinvigorating the debate over the role of policy substance in shaping congressional politics.
Polarization and partisanship in Congress have trended upward since the 1970s and conflict has increasingly become organized around ideological concerns (Poole and Rosenthal 1997; McCarty, Poole, and Rosenthal 2006; Lee 2009). These trends are typically viewed as institution-wide phenomena, intensifying across the years in both the House and Senate. Some scholars have pointed to aspects of the legislative process that are more or less resistant to these trends; for example, procedural votes are more likely to polarize the parties (Theriault 2008). For the most part, however, scholars have viewed the process of increasing polarization and the corresponding simplification of the choice space as consequences of societal developments (McCarty, Poole, and Rosenthal 2006), regional political realignments (Poole and Rosenthal 1997; Stonecash, Brewer, and Mariani 2003), or reforms to legislative rule and procedure (Rohde 1991; Theriault 2008).
These explanations are clearly credible, but they omit a very important component of congressional politics: the issues that generate voting coalitions. Some issues, like agriculture, tend to diffuse partisan polarization while others, such as macroeconomics, exacerbate it. Most if not all issues contain a left–right ideological component centering on government intervention in economic affairs. But as we show, many also contain crosscutting components that divide legislators by concerns that appear neither ideological nor partisan in nature.
The question is not whether Congress has become more polarized; this has obviously happened. Poole and Rosenthal (2001, Figure 1) document a strong and sustained decline during the postwar period in the improvement offered by the second dimension. The question is how this has occurred. In this article, we examine the evolution of issue dimensionality since the mid-1960s by scaling roll call votes within the Policy Agendas Project’s topic categories. Our analysis indicates that the period of study was characterized by a series of issue evolutions in which individual issues became integrated into the partisan and ideological fabric of the political system.
Extracted dimensions for 18 issue areas, 1965-2004.
We examine three aspects of issue dimensionality that together provide evidence that party polarization has proceeded on an issue-by-issue basis. First, we show that issues vary in their dimensionality. Some issues are readily captured by the unidimensional, ideological model of Congress while others are more complicated, requiring additional dimensions to explain members’ voting behavior. Second, we identify an evolution in dimensionality over time, from multidimensional spaces toward unidimensional spaces. A simplifying choice space suggests that concerns orthogonal to ideology and party are becoming less central to the organization of congressional politics. Variation in the process of simplification—that is, different rates and different periods—suggests such changes cannot be the result of institution-wide phenomena that affect all issues similarly. Third, we demonstrate that party cohesion has something to do with these patterns, supplementing our study of issue-based dimensionality with an examination of issue-based party unity measures. We find evidence of an increasing correspondence between ideology and party over time and this cannot be explained by changes to chamber operations that cause procedural or amendment votes to be distributed differently across issue categories.
In our analysis, we do not distinguish between enduring coalitions that cut across the dominant dimension in a consistent fashion, such as the regional dimension that Poole and Rosenthal isolate in their studies, from those that may be more transient. Transient coalitions play an important role in pluralist accounts of American politics and hence must not be excluded from consideration. The disappearance of such transient coalitions is an important part of the polarization story and we offer evidence that this disappearance has occurred on an issue-by-issue basis.
Issues, Dimensionality, and Partisanship
A dimension is simply a spatial arrangement of preferences with respect to a set of alternatives. In unidimensional choice spaces, individual preferences are arrayed along a single continuum. The coalitions that determine the outcome of the choice are flexible only insofar as legislators occupying the middle of the unidimensional space shift positions (as when a centrist moves from the liberal coalition to the conservative coalition). A multidimensional choice space generates a different arrangement of preferences. The issue of agriculture may pit supporters of farm subsidies against critics. But some of the supporters and critics may also favor greater regulation of the nation’s food supply. In this two-dimensional choice space, legislators’ preferences on one dimension (e.g., farm subsidies) cluster together while on the other dimension (e.g., food safety) are dispersed.
The notion that congressional voting can be organized around a single, ideological dimension of conflict is a relatively recent development. Early research on the dimensionality of voting choice emphasized the fluidity of coalitions across different issues such that Congress was organized around distinctive “policy dimensions” (MacRae 1965; Clausen and Cheney 1970; Clausen 1973). With Poole and Rosenthal’s (1997) consideration of roll call votes since the dawn of the Republic, issue-centric conceptions of congressional voting were replaced by ideological ones emphasizing the power of party and ideology to readily simplify the complexities of congressional life.
Some scholars have suspected that the notion of a unidimensional choice space was an overgeneralization (Jackson and Kingdon 1992; Kingdon 1977; Snyder 1992). Scholars have found multidimensional choice structures in cosponsoring behavior (Talbert and Potoski 2002) and distributive awards to congressional districts (Potoski and Talbert 2000). Two or more dimensions can emerge in assessing the choice structure of individual bills or particular policy areas (Hurwitz, Moiles, and Rohde 2001; Smith 2007). Crespin and Rohde (2010) find members of Congress vote differently across annual appropriations bills, a finding incompatible with a unidimensional choice space.
We contribute to the burgeoning literature challenging unidimensional conceptions of Congress and reinvigorate the debate over the role of policy substance in shaping congressional politics. There is little doubt that the central dimension in modern politics is the extent of government intervention in the economy. But, not all issues are so readily integrated in the ideological concerns of members. Party leaders have long struggled to retain control of policy “subsystems” led by powerful committee chairs and supportive interest groups. Issues dominated by funding initiatives that target particular sectors such as defense contracting or farmers generate less partisan and ideological conflict (Hurwitz, Moiles, and Rohde 2001; Stein and Bickers 1995; Thorpe 2010). Distributive policy is more likely to emphasize constituency service, and legislators are less likely to cede authority to party leaders, especially when it has electoral costs (Deering and Smith 1997). Some regional divisions among legislators split traditional ideological coalitions but only for particular issues such as trade or farm policy (Sinclair 1978; Turner 1952). These distributive and regional bases of conflict imply multidimensionality in legislative choice.
The well-documented rise in party polarization strongly suggests that pluralistic coalitions organized around concerns orthogonal to party are on the decline. But, why is this? One possibility is that parties are increasingly acting to organize issues into the ideological structure of politics, effectively reducing the influence of crosscutting dimensions of conflict. Parties face strong incentives to draw distinctions among themselves. As Francis Lee (2009) suggests, the pressure to compete leads parties to increasingly emphasize issues for which they hold strong disagreements and she documents a steady rise in the proportion of ideological votes in the Senate.
Some anecdotal evidence suggests that the parties’ power of integration is limited, however. Consider Republicans’ attempts to make farm subsidies a partisan issue with the Freedom to Farm Act of 1996 (Hurwitz, Moiles and Rohde 2001). Their efforts ultimately succumbed to the overwhelming power of distributive politics. And sometimes, parties face few incentives to engage in opposition—consider the high-profile debate on mandatory minimum sentences culminating with the bipartisan passage of the Anti-Drug Abuse Act of 1988 (see Jochim 2012).
There is a second explanation that is almost always omitted in discussions of party alignments—the issue itself. Issues may evolve in a manner that evokes fundamental ideological disputes.Transportation, education, environmentalism, health care, and even business regulation all involved less ideological rancor early in their respective developments. Consider, for example, the evolution of education from one involving no-strings-attached federal funding to one in which strong federal oversight is the norm. It is possible that the parties shaped this agenda change, but it is also possible that the issues evolved in response to other political pressures. Many government programs begin small and get bigger over time, as rules and statutory adjustments are made to solve problems within the existing policy arrangements and interest groups emerge to defend and expand their purview. As this happens, they may tap into existing divisions over redistribution and regulation that characteristically divide the parties.
This issue-based evolution in dimensionality is particularly relevant given the expansion of issue agendas in the 1960s and 1970s. Not only did government expand its reach, it also incorporated issues previously left to state or local governments or the private sector, including race, environmentalism, consumer relations, housing subsidies for the poor, education, and crime. It is possible that as these issues generated controversy, they became fodder for partisan conflict and less subject to distributive politics. If parties and ideology are evolving to incorporate new issues, the structure of legislative choice should simplify over time as the issues become integrated into the ideological preferences of the parties’ members. This evolution has clear implications for the extent of party polarization.
Data and Methods
Our strategy in examining these issue-centric dynamics is to scale roll call votes within well-defined issue areas. If issues act to magnify or attenuate ideological polarization and partisanship, then we should find evidence of variability across issues, an evolution over time (from higher to lower), and a relationship between issue dimensionality and party cohesion.
In the first step of this research, we coded all roll call votes from 1947 to 2004 using the Policy Agendas Project’s content coding protocols. These data, including a full description of each of the eighteen issue areas we consider are available for public access through the Policy Agendas Project website (www.policyagendas.org). In the second step, we scaled the votes within each issue area separately from the 89th to the 108th Congresses. 1 While a more extensive historical analysis of the roll call voting record would be advantageous, too few of the issue areas have enough activity in them in the earlier period to reliably scale them. 2
While there has been some debate over desirable approaches used to examine the dimensionality of roll call votes (Heckman and Snyder 1997; Poole and Rosenthal 1997), our choice of methodology is driven by the desire to facilitate direct comparisons with other studies. We utilize Poole and Rosenthal’s W-NOMINATE, which assesses dimensionality in single Congresses. The NOMINATE procedure reduces floor votes to one or more underlying dimensions and then identifies legislators’ scores on those dimensions utilizing a logit specification. 3
While ideal point estimates from W-NOMINATE across different Congresses are not directly comparable, there is no evidence that the extracted dimensions are affected by this limitation. A different algorithm developed by Poole and Rosenthal (1997), DW-NOMINATE, produces ideal point estimates that are comparable over time using a dynamic estimation procedure that moves as a function of time, effectively bridging members’ behavior across Congresses. In contrast, W-NOMINATE constrains legislator ideal points to lie within the unit circle. In neither case is it suggested that the dimensional structure of the voting space changes by introducing the dynamic component (see also Poole 2005). 4 Indeed, there is little reason to believe that the extracted number of dimensions is affected in a similar fashion for it relies on variance explained and is not sensitive to the scale of the preference space. Our approach does not differ markedly from other research in this area (see Talbert and Potoski 2002; Potoski and Talbert 2000). Appendices A and B present summary statistics for the eighteen issue areas we analyze (see appendices at http://prq.sagepub.com/supplemental/).
Judging the appropriate number of dimensions for characterizing a pattern of votes is at least part art. All approaches relate the criterion variable to the number of dimensions extracted. We use the straightforward “elbow test” developed by Cattell (1966). The elbow test assesses dimensionality through the visual examination of scree plots, which plot the size of the eigenvalues against each dimension. An elbow or sharp change in curvature in the plot suggests more dimensions do little to explain variation in the data. See Appendix C for a set of sample scree plots. To assess the degree to which a unidimensional model fails to account for the underlying variance in congressional voting we examine the strength of the first dimension by calculating the proportion of variance explained by the first eigenvalue. Higher values indicate a stronger, more robust first dimension while lower values indicate a weaker first dimension.
The above procedures allow us to examine the issue-bases of ideological structure and how they change over time. To assess the connection between ideology and partisanship, we calculate the proportion of party unity votes with data drawn from Rohde (2004). A party unity vote is a vote that pits the majority of one party against the majority of the other party. These data are used to reveal the extent to which parties structure choice in each of the areas we consider and how this relationship has evolved over time.
The level of aggregation can affect dimensionality in congressional voting. Roberts, Smith, and Haptonstahl (2008) find that dimensional structures simplify as the level of aggregation increases (i.e., bills to issues to congresses). While Poole and Rosenthal aggregate at the highest level, Crespin and Rohde (2010) consider the annual appropriations bills for the Department of Housing and Urban Development and Independent Agencies. One could also aggregate at the committee level. It is unclear that any of these is the one correct approach; it is likely that different levels of aggregation reveal different elements of the choice structure. Studying individual bills and committees could miss issue-based political dynamics, as legislative proposals may reflect a coherent set of programs and policies but they may also represent an assembly of unrelated issues pulled together via legislative maneuvering (Krutz 2001). Our choice of the issue as the level of aggregation is due to our interest in issue-based political dynamics and their evolution.
Issue Dimensionality, 1965-2004
Figure 1 displays the extracted number of dimensions for each issue area across the period of the study. 5 All of the issues we consider face the same institutional and electoral contexts. Yet, despite this, they reveal remarkably different patterns. Some issues exhibit multidimensionality often (e.g., agriculture, science) while others are mostly stable at a single dimension (e.g., economics, labor and employment). Appendix D reports dimensionality across each issue-congress combination in a tabular format.
What might be the content of the dimensions beyond the general first dimension? For the purposes of our analysis, the content is irrelevant. The patterning evident in Figure 1 indicates there may be considerable instability in the dimensions, with extra dimensions appearing within one issue and then fading, which indicates that the voting coalitions underlying these dimensions are transient. The ability of legislators to form and disband transient coalitions is one hallmark of pluralistic (as opposed to ideological) politics. More generally, the extent to which an issue is viewed as having distributive characteristics could make it more resistant to ideological incorporation. However, the bases of distributional coalitions need not be constant; they may form and re-form depending on what is being distributed. Hence one would expect to see dimensionality “bounce around” as issues evolve.
The existing literature suggests some bases for our findings. For example, regional coalitions may emerge on some issues but not on others, and even then hold only for some votes within an issue area. Occasionally an “ends against the middle” pattern emerges, as was the case for the Emergency Economic Stabilization Act of 2008 (the Wall Street bailout bill)—on the initial vote conservatives and liberals voted against moderates in both parties. There can be different reasons for the persistence of multidimensionality, and some (such as regional coalitions) may be fairly stable but others may be fleeting.
Whereas Poole and Rosenthal (1997) highlight the strength of the unidimensional model in the absence of political transformation, our findings suggest that multidimensionality at the issue level occurs regularly, though some issues are more susceptible to it. While civil rights have been central to explanations of multidimensionality in aggregate analyses, our results suggest that most multidimensional issues have few connections to racial politics. There is little doubt ideology constrains the structure of legislative choice and that civil rights transformed that structure, but these factors do not seem to be the primary shapers of member behavior on issues like agriculture, the environment, or transportation.
In light of this variability, can we characterize trends in the structure of choice across issue areas? Figure 2 displays the average number of extracted dimensions for each issue. These averages are calculated on slightly different bases because for some issues and congresses there were too few votes to scale reliably. The most complex issue area, on average, was science (D = 2.7) followed by trade (2.5), agriculture (D = 2.4), public lands (D = 2.1), transportation (D = 2.0), and energy (D = 2.0). Each of these areas exhibited two or more dimensions. The simplest policies were, again on average, labor and employment (D = 1.2), housing (D = 1.2), economics (D = 1.3), health (D = 1.4), and social welfare (D = 1.4).
Mean extracted dimensions for 18 issue areas, 1965-2004.
The issues with the simplest dimensional structures are those most directly associated with intervention in the economy, either through regulation, monetary and fiscal policy (labor, economics), or through major spending initiatives (health, housing). These are the arenas that have been most aggressively organized by the political party system and subject to the most intense political scrutiny in contemporary American politics. They largely reflect the divide between Republicans and Democrats established in the wake of the New Deal and the centrality of redistribution to ideological conflict.
Those with more complex structures are traditionally characterized as distributive, subject to logrolling, or simply outside of the realm of partisan politics. Science, trade, agriculture, public lands, and transportation each have strong distributive elements with large grant programs targeting businesses and/or states. In many of these, partisan politics plays out alongside distributive politics. Consider, for example, the case of energy (D = 2) in which subsidies for ethanol producers generate one set of conflicts among legislators while drilling in the Arctic National Wildlife Refuge produces a very different set.
How different are the patterns of dimensionality across these issue areas? Of the 153 possible issue pairings, more than 20 percent reveal statistically significant differences according to the results of a post-ANOVA difference of means test (results available in Appendix E). Of particular note is the contrasting pattern between those issues that emphasize traditional ideological debates around redistributive and regulatory concerns (e.g., economics, health care, labor and employment, and social welfare) and those that are traditionally portrayed as primarily distributive in nature (e.g., agriculture, transportation, science, and public lands). Issues that have tended toward redistributive and regulatory policies are not significantly different from one another. Likewise, issues with a strong distributive component do not differ from similarly situated issues.
Do higher order dimensions account for enough variation to warrant their inclusion, or are they reflective of random error? We calculate the proportion of variance accounted for by the first dimension for each of the eighteen issues. Table 1 presents the results. The strength of the first dimension varies widely among issues. Some issues are characterized by a stronger, more robust first dimension (e.g., civil rights, social welfare, environment, and labor) while others are characterized by a considerably weaker first dimension (e.g., agriculture, trade, transportation, and public lands). For example, the strength of the first dimension within agriculture policy is 60.7 percent, which suggests that in an average year, the first dimension fails to account for nearly 40 percent of the variance in congressional voting. This stands in contrast to areas such as economics, housing, and labor where the first dimension accounts for nearly 90 percent of the variance.
The Strength of the First Dimension.
Note: Cell values display the average proportion of variance explained by the first dimension across eighteen issue areas from 1965 to 2004.
The Evolution of Issue Dimensionality
Anecdotal evidence suggests that the content of issues can evoke lower dimensional conflict and increasing levels of partisanship. For example in the multidimensional arena of agriculture, dimensionality collapsed to a single dimension in the 108th Congress. During this Congress, the House took up and passed H.R. 554, the Personal Responsibility and Food Consumption Act, which sought to protect producers and retailer of foods from class action lawsuits from consumers alleging negligence. Party unity votes on agriculture increased from their historical average of 28 percent to 65 percent of votes cast.
Given the increase in party polarization in Congress, it is likely that issue dimensionality simplifies as the parties increasingly intrude into areas once dominated by distributive politics. Alternatively, dimensional structures may exhibit a high degree of stability across time as durable coalitions consistently structure the choice space. To examine the changing structure of congressional voting, we calculated the average number of dimensions extracted per issue for two periods: an early period (1965-1980), allegedly characterized by weaker legislative parties, more bipartisanship, and a stronger committee structure, and a later, more partisan era characterized by weaker committees, stronger parties, and more polarization (1981-2004). Any attempt to divide American politics into “eras” can be misleading, but limitations in our ability to scale the roll calls in all issue areas for all congresses mean that some simplifications are necessary to reveal possible trends. 6 Table 2 presents the results.
Dimensionality Over Time.
Notes: Table displays the mean number of extracted dimensions for each policy area across two eras. The Diff. column is the difference between era two and era one.
p > .01, **p > .05, *p > .10 (two tailed).
Table 2 shows that the evolution of legislative choice has not followed a single pattern. Some issues are charac-terized by fewer dimensions in the contemporary period. We find statistically significant (p > .10) declines in dimensionality in education (−1.0), science (−0.89), public lands (−0.88), transportation (−0.83), health care (−0.55), and business (−0.54). Others, including labor, agriculture, defense, and trade, have remained relatively stable. Still others, such as the environment and foreign affairs exhibit increases in their average dimensionalities (0.29 and 0.20, respectively), but these differences are not statistically significant.
We may draw three conclusions from this analysis. On the one hand, the majority of issues exhibited declines in dimensionality across the two eras, a trend that is perhaps unsurprising given increases in party polarization. On the other hand, a variety of issues have been resistant to the trends toward unidimensionality (agriculture, foreign affairs, science, transportation, trade, and public lands), which suggests that issue-based politics continues to survive despite party polarization. Finally, a variety of issues (health care, education, civil rights, crime and law, and defense) have evolved from multidimensional space to unidimensional spaces. This suggests that party and ideology have come to intrude into issues once less susceptible to their power.
Parties and Issue Dimensionality
Legislative parties have a distinct role in shaping legislative choice (Aldrich 1995; Cox and McCubbins 2005; Poole and Rosenthal 1997; Rohde 1991). If a set of votes is characterized by only two coalitions (e.g., Republicans and Democrats), the choice space is necessarily unidimensional, but the converse need not be true—low-dimensional spaces are not necessarily partisan ones. And, this relationship can change over time. So here we examine the possibility that the evolution of legislative choice described above reflects a strengthening of the connection between partisanship and voting over time. Because coherence in the respective party coalitions is critical to establishing and maintaining low-dimensional structures, we study the proportion of party unity votes for each issue area.
Table 3 presents the findings. The correlation between the number of extracted dimensions and the proportion of party unity votes for all issues regardless of era is −0.59 (p = .01), indicating that increased levels of partisan conflict in issues are associated with a reduction in the dimensional space of issues. Partisan conflict is highest in the areas we would expect it to be—economics, labor, and housing—all issues that exhibit stable unidimensional choice spaces. Partisan conflict is lowest in the areas of environment, foreign affairs, agriculture, and transportation. Among the areas with fewer party unity votes, the relationship is more ambiguous. For example, environmental issues are characterized by fewer party unity votes than we might expect given its relatively low-dimensional structure (D = 1.5). 7
Relationship between Dimensionality and Partisanship.
Note: Cell values display the average extracted dimensions and party unity votes from 1965 to 2004 and the average proportion of party unity votes across two eras. The Diff. column displays the difference between era 1 and era 2. Statistics include the correlation between number of dimensions and the proportion of party unity votes and the statistical significance of these correlations. ***p > .01, **p > .05, *p > .10 (two tailed).
Second, the relationship between dimensionality and partisanship has changed across the two eras and the results are striking. In the earlier era, partisanship is not reliably associated with dimensional structures—the correlation between the number of extracted dimensions and party unity is a statistically insignificant −0.14. In the later era, the correlation jumps to −0.74. In the earlier era, low-dimensional structures within issues are not related to party organization of those issues; party unity is not significantly different across the issue categories. In the later era, they are tightly related—unity and low dimensionality are connected. It is not much of an overgeneralization to claim that legislative conflict has evolved from a system in which party and ideology across issues were not related to one in which ideological and partisan conflict correspond.
The pattern plays out across issues in different ways. Some traditionally partisan issues have become more partisan over time. Issues such as economics and labor had high levels of partisanship in the earlier era (59 and 52 percent party unity votes, respectively), but the associations are even higher in the era beginning in 1980. Because relatively high levels of partisanship already existed, the effect on the choice space was minimal—both became slightly more unidimensional (in the case of labor and employment, choice space was reduced from 1.29 to 1.08 while in the case of economics, it was reduced from 1.50 to 1.17).
For other issues, the effect of increasing partisanship is more significant. High levels of bipartisanship and a multidimensional choice space once characterized issues such as science, business regulation, and defense. In each of these cases, growing levels of partisanship is associated with a reduction in dimensionality. Education votes in the earlier era were characterized on average by two dimensions and just 22 percent of votes revealed party unity; after 1980, the choice space collapsed to one dimension and almost 50 percent were party unity votes. Of all the issues we consider, civil rights exhibits the largest increase in party unity voting, and the choice space has simplified correspondingly. This largely fits with conventional depictions of the issue—as southern Democrats aligned with Republicans, the issue became couched in ideological terms (Poole and Rosenthal 1997; Carmines and Stimson 1989; Sinclair 1982).
Procedural and Amendment Votes
It is possible that our findings regarding the relationship between partisanship and dimensionality is driven in part or in whole by changes in the type of votes in each area. The overall composition of the roll call record changed with the introduction of electronic voting in 1974, with substantial increases in the number of procedural and amendment votes (Smith 1989). Theriault (2008) suggests that increases in party polarization can largely be traced to the growing number of procedural votes on the agenda. Roberts and Smith (2003) argue that amendment votes are more polarizing than final passage votes because they pit two alternatives against each other, rather than an alternative against (an unacceptable) status quo. Yet, Shafran (2011) shows that votes on amendments generate more deviations from first-dimension NOMINATE scores for the period 1985-2004.
In either of these cases, it is possible that changes in dimensionality reflect differences in the types of votes—procedural or amendment—on the congressional agenda, rather than differences in the content of issues. To assess this, we examined the relationship between procedural and amendment voting and dimensionality for each of the eighteen issues. We calculated the proportion of procedural and amendment votes using data from Rohde (2004). We are interested in the overall level of procedural and amendment votes by issue area, whether these incidences have changed as a proportion of votes over time, and the relationship between the type of vote and extracted dimensionality.
Tables 4 and 5, which present the relevant analyses, reveal little relationship between the proportion of procedural or amendment votes in each issue area and the extracted number of dimensions. There is a modest but insignificant negative correlation between dimensionality and the proportion of procedural votes across issues. If we consider the periods separately, we find no relationship for the first period between dimensionality and procedural votes, but a larger negative (but still insignificant) correlation during the second period. Across the two periods, only health and macroeconomics have increased their proportion of procedural votes significantly, and several issues, including housing, science, foreign affairs, and civil rights exhibit a nontrivial decline in procedural votes over the two eras we consider. Procedural votes cannot account for issue differences in dimensionality overall, nor can they explain differences across eras.
Relationship between Dimensionality and Procedural Votes.
Note: Cell values display the average extracted dimensions and procedural votes from 1965 to 2004 and the average proportion of procedural votes across two eras. The Diff. column displays the difference between era 1 and era 2. Statistics include the correlation between number of dimensions and the proportion of procedural votes and the statistical significance of these correlations (tdf = 18). ***p > .01, **p > .05, *p > .10 (two tailed).
Relationship between Dimensionality and Amendment Votes.
Note: Cell values display the average extracted dimensions and amendment votes from 1965 to 2004 and the average proportion of amendment votes across two eras. The Diff. column displays the difference between era 1 and era 2. Statistics include the correlation between number of dimensions and the proportion of amendment votes and the statistical significance of these correlations (tdf = 18). ***p > .01, **p > .05, *p > .10 (two tailed).
The story for amendment voting is more complex. A higher proportion of amendment voting is associated with higher dimensionality, albeit not significantly. But the role of amendment voting shifts across the two periods—from a moderate −0.44 correlation between dimensionality and the proportion of amendment votes in the first period to basically no relationship in the second. What strikes one, however, is the vast increase in the proportion of voting in Congress on amendments across time; these changes affect some issues (civil rights, health care, labor, and employment) much more strongly than others. While the proportion of amendment votes cannot account for the issue-based differences we report, the radical changes over time in amendments and the consequent decline in final passage votes bears further examination.
Discussion and Conclusion
Poole and Rosenthal (2001) report a steady increase in the power of the first, ideological, dimension in organizing congressional roll call voting. How has this occurred and what consequences has it had? Our findings may be summarized as follows. Our analyses have revealed considerable variability in the structure of legislative choice among the issues we consider. Some issues are characterized by a low-dimensional structure throughout the period of study, others revealed robust multidimensionality, and some moved from multidimensional spaces to unidimensional spaces. Our analysis of party unity votes within the issues indicated that the parties are implicated in the issue-by-issue polarization process. These findings suggest that issues have been brought into the increasingly unidimensional structure of congressional politics sequentially and that a variety of issues remain resistant toward full incorporation.
For each issue, we invariably find a major role for the left–right dimension; the questions for us is how strong that dimension is across issues and why it has strengthened over time. We find systematic differences that are substantively meaningful and this strongly suggests the additional dimensions are not simply random error. We cannot say from our analyses whether the dimensions beyond the first are similar for issues or across time, but we do argue that the particular nature of the dimensions is less important than the fact that they indicate that the party-based coalitions are split across some of the votes in the cluster comprising the issue. Something is interfering with the ability or will of each party to construct ideologically based coalitions for these. The standard suspects are constituency and interest group pressures, topics that merit further study.
Three separate mechanisms may be implicated in the transformation of issues from pluralistic, multidimensional arenas to ideological, unidimensional ones. The first is based in external party dynamics. Parties invade the issue space and increasingly insist that the ideological dimension is of prime importance. This may be due to the Southern realignment, the operation of primaries, the increasing role of specialized media or money in politics, or due to the strategic action of legislative parties. Regardless of how it happens, issues increasingly get defined along a left–right ideological dimension because parties are becoming increasingly structured that way and the causes are largely exogenous to Congress. Polsby’s (2004) study of congressional evolution is almost entirely based on changes external to the body itself.
The second mechanism is internal to the legislature and reflects the classic struggle between central legislative party control and the centrifugal forces of issue-based politics. Power over legislative rules, procedure, and organization has become increasingly concentrated in party leaders and these forces are strongest when party goals align with the ideological preferences of members. Speaker John Boehner’s struggle to maintain control over the Republican caucus during the 112th Congress reflects not the typical historical struggle between the Speaker and his committee chairs; it is a Speaker struggling with a caucus that is more ideological than he is.
The third mechanism is based in the evolution of issues. Issues may develop in a manner that provokes an ideological response from the parties. What begins as a problem-solving exercise to address a specific policy problem becomes an intrusive government program through what Aaron Wildavsky termed “policy as its own cause.” Policy solutions “create their own effects, which gradually displace the original difficulty, . . . public agencies are ever more involved in making adjustments to past programs, creating new ones to overcome difficulties, and responding to forces originating in other sectors or in society” (Wildavsky 1979, 81). Each governmental attempt to address issues generates other problems that cause more governmental action. As James Q. Wilson (1979, 41) wrote, “Once the ‘legitimacy barrier’ has fallen, politics takes a very different form. . . . New programs need not await the advent of a crisis or an extraordinary majority, because no program is any longer ‘new’—it is seen, rather, as an extension, a modification, or an enlargement of something the government is already doing.” If it is true that there exist no social limits on the issues that government may involve itself in, then perhaps it is not surprising that those issues get organized along a left–right ideological dimension.
In the 1960s and 1970s, the rise of new issues previously beyond the “legitimacy barrier,” including civil rights, education, health care, crime and justice, and the environment, resulted in a large expansion of the congressional agenda (Baumgartner and Jones 2009). During the 1950s, Congress conducted around 1300 hearings per year, but that rose in the late 1960s and throughout the 1970s, peaking in 1979 at 2246, according to the data from the Policy Agendas Project. By 1995, it was down to around 1500, where it has stabilized. Our analyses suggest few of these issues polarized the political parties immediately. Rather, it was later, as efforts to implement specific policy initiatives revealed the full range of regulatory and/or redistributive actions implied by the initial legislation, that polarization seems to have increased.
Whatever the mechanism, the implications are important. It would be hard to contest the notion that labor, economics, and housing are, at least in the voting stage, organized in unidimensional decision arenas and subject to partisan and ideological constraint. On the other hand, for issues such as trade, science, and agriculture, it is more likely the traditional organization in subsystems and committees continues unabated by ongoing polarization with constituency and interest groups jostling in a pluralistic decentralized arrangement. As partisan ideological concerns intrude into issue areas once handled through the pluralist adjustments of the interest-group system, there is less room for issue-centered bargaining and the emergence of subsystem partial equilibria—or at least there is less room for adjusting these interests on the floor of Congress.
We contend that the content of issues generate the results we report here. While we have ruled out several explanations for the patterns we observe, including the differential occurrence of procedural and amendment votes across issues, we have not been able to examine all potential explanations. One is the possibility that the majority party assembles bigger majorities among its members in order to engage in bargaining with the Senate and the President. This would have the consequence of increasing party unity votes on the more controversial issues, leaving lower levels of party unity on less controversial votes. Given the patterns we observe, this would require an interaction between time and what we might term “sophisticated majoritarianism,” which might occur more in divided governments. However, both periods we studied were characterized by extensive periods of divided government.
But what is it about issue content that causes differential polarization? We speculate that the key to understanding this lies in the process of issue definition, in particular, the increasing tendency of political actors to view formerly distributive issues as redistributive ideological ones. The pork barrel and the logroll have become anthemia to more ideological parties. Recent events, such as the attack by many in the Republican Party on both the idea and the funding of specific projects in specific districts (termed “earmarking”) and the unwillingness to be swayed by loyalty to the Speaker or to internal organizational incentives controlled by the party leadership, suggests that the typical mechanisms for managing constituency interests may be on the decline. Yet in some areas deference to specialists still exists and some policies are made primarily by specialists interacting with interests and achieving deference on the floor. It is hard to imagine that all policy arenas will fall within the sway of the partisan and ideological fabric, but that remains to be seen.
Footnotes
Appendix
Post-ANOVA Difference of Means Test for 18 Issue Areas.
| Issue Area | Econ | C Rights | Health | Agric | Labor | Educ | Env | Ener | Tran | Crime | Welfare | Hous | Bus | Def | Sci | Trade | Foreign | P Lands |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Economics | – | Economics | – | |||||||||||||||
| Civil Rights | 0.99 | – | Civil Rights | 0.99 | – | |||||||||||||
| Health Care | 1.00 | 1.00 | – | Health Care | 1.00 | 1.00 | – | |||||||||||
| Agriculture |
|
|
|
– | Agriculture |
|
|
|
||||||||||
| Labor & Empl | 1.00 | 1.00 | 1.00 |
|
– | Labor & Empl | 1.00 | 1.00 | 1.00 | |||||||||
| Education | 0.99 | 1.00 | 1.00 |
|
1.00 | – | Education | 0.99 | 1.00 | 1.00 | ||||||||
| Environment | 0.99 | 1.00 | 1.00 |
|
0.99 | 1.00 | – | Environment | 0.99 | 1.00 | 1.00 | |||||||
| Energy | 0.43 | 0.86 | 0.73 | 0.99 | 0.16 | 0.91 | 0.98 | – | Energy | 0.43 | 0.86 | 0.73 | ||||||
| Transportation | 0.13 | 0.57 | 0.42 | 0.94 |
|
0.68 | 0.88 | 1.00 | – | Transportation | 0.13 | 0.57 | 0.42 | |||||
| Law & Crime | 1.00 | 1.00 | 1.00 |
|
1.00 | 1.00 | 1.00 | 0.80 | 0.49 | – | Law & Crime | 1.00 | 1.00 | 1.00 | ||||
| Social Welfare | 1.00 | 1.00 | 1.00 |
|
1.00 | 1.00 | 1.00 | 0.80 | 0.49 | 1.00 | – | Social Welfare | 1.00 | 1.00 | 1.00 | |||
| Housing | 1.00 | 1.00 | 1.00 |
|
1.00 | 1.00 | 1.00 | 0.37 | 0.11 | 1.00 | 1.00 | – | Housing | 1.00 | 1.00 | 1.00 | ||
| Business | 0.68 | 0.99 | 0.94 | 0.31 | 0.27 | 0.99 | 1.00 | 1.00 | 1.00 | 0.97 | 0.97 | 0.60 | – | Business | 0.68 | 0.99 | 0.94 | |
| Defense | 0.99 | 1.00 | 1.00 |
|
0.95 | 1.00 | 1.00 | 0.97 | 0.85 | 1.00 | 1.00 | 1.00 | 1.00 | – | Defense | 0.99 | 1.00 | 1.00 |
| Science |
|
|
|
1.00 |
|
|
|
0.73 | 0.44 |
|
|
|
|
|
Science |
|
|
|
| Trade |
|
|
|
1.00 |
|
|
|
0.92 | 0.72 |
|
|
|
0.17 |
|
Trade |
|
|
|
| For Affairs | 0.68 | 0.98 | 0.94 | 0.41 | 0.27 | 0.99 | 1.00 | 1.00 | 1.00 | 0.97 | 0.97 | 0.60 | 1.00 | 1.00 | For Affairs | 0.68 | 0.98 | 0.94 |
| Public Lands |
|
0.26 | 0.18 | 1.00 |
|
0.36 | 0.62 | 1.00 | 1.00 | 0.21 | 0.21 |
|
1.00 | 0.53 | Public Lands |
|
0.26 | 0.18 |
Notes: Cell values show the p-value for a post-ANOVA Tukey’s HSD difference of means test between the extracted number of dimensions for each issue area from 1965-2004. F17 = 6.8, p < .001. Values significant at the p <.10 highlighted in bold.
Acknowledgements
This paper has benefited from the helpful comments we received from Larry Dodd, Sean Theriault, Adam Rammey, John Wilkerson, and Peter May. Earlier versions were presented at the 2009 Southern Political Science Association and the 2010 Midwest Political Science Association. We appreciate the support of the National Science Foundation, the University of Washington, and the University of Texas, Austin for making the data collection reported in this project possible.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.