Abstract
Data from the 2010 Cooperative Congressional Election Study are examined to determine the extent to which partisan polarization is a popular as well as an elite phenomenon. The evidence indicates that ordinary Americans as well as political elites divide consistently along party lines over a broad range of issues and legislative preferences. Partisan ideological divisions are not confined to the political class; they extend downward into the citizenry to the degree that people are actively involved in politics. The coincidence of partisan differences across issues results in a bimodal distribution of aggregate preferences among voters, showing that partisan sorting can by itself polarize the electorate even if sorted partisans do not adopt extreme positions on individual issues. The distribution of partisan preferences is even more sharply bimodal and polarized when the electoral constituencies of the congressional parties are compared, so the electoral process sustains elite polarization. Primary electoral constituencies tend to be even more extreme, particularly on the Republican side, deterring departures from party orthodoxy and thus movement toward the median voter.
The ideological divide between the American national parties has been widening for several decades. The trend has been most conspicuous in Congress, and by the most commonly cited measure—McCarty, Poole, and Rosenthal’s DW-Nominate scores 1 —the congressional parties were further apart ideologically in the 111th Congress (2009-2010) than at any time since the Civil War. This and a wide variety of additional evidence has forged a strong consensus among political scientists (e.g., Baumer & Gold, 2008; Fiorina, 2009; McCarty, Poole, & Rosenthal, 2006; Sinclair, 2006; Stonecash, Brewer, & Mariani, 2003) and other professional observers of politics (Bishop, 2008; Edsall, 2006) that partisan polarization, as the trend is usually characterized, has fundamentally transformed national politics since the 1970s. Anyone observing the bitter legislative battles between Barack Obama and congressional Republicans both before and after the 2010 election would be hard-pressed to disagree.
Consensus breaks down, however, over the extent to which ordinary Americans have participated in the polarizing trend. Morris P. Fiorina and his collaborators (Fiorina, 2009; Fiorina, Abrams, & Pope, 2006, 2008; Fiorina & Levendusky, 2006) argue that by and large they have not. In their view, movement toward the ideological extremes has been confined to a narrow political class—politicians, activists, pundits, and other assorted political junkies. They cite evidence that the distribution of Americans’ opinions on a variety of social and economic issues has not become more polarized or extreme (Evans, 2003)—except among the minority of citizens most visible in politics as candidates, activists, and professional observers. A large majority of ordinary citizens, according to this argument, do not share the zealotry or extremism of the active stratum but cannot express their moderation at the polls because candidates offer only polar options. This critique is summed up in one of their titles: “Disconnected: The Political Class Versus the People” (Fiorina & Levendusky, 2006).
This critique, as its proponents recognize, challenges the classic theories of electoral and congressional politics developed by Anthony Downs (1957) and David Mayhew (1974). Downs’s spacial logic predicts that competing candidates or parties will strive to maximize their share of votes by taking positions appealing to the median voter on any unidimensional (e.g., left-right) scale; in a Downsian world, extremists would always lose to centrists. How do extremist politicians nonetheless get elected? Mayhew imagines what Congress would look like if its members were “single-minded seekers of reelection,” and his astute depiction of the actual institution drawn from this perspective has made the electoral incentive the guiding premise of most subsequent congressional research. How does “a disconnect between the American people and those who purport to represent them” (Fiorina & Levendusky, 2006, pp. 51-52) emerge if the primary objective of representatives is to please enough voters to win reelection?
One possibility, argued most thoroughly by Alan I. Abramowitz (Abramowitz, 2006, 2010; see also Abramowitz & Saunders, 2005, 2008), is that there is no disconnect: “Polarization in Washington reflects polarization within the public, especially the politically engaged segment of the public” (Abramowitz, 2010, p. x). Abramowitz concedes that a substantial portion of the public—the politically disengaged who pay little attention to politics and do not vote—has shown no movement away from the center. But he offers evidence that other Americans have grown increasingly divided along party lines on issues and ideology, and the more involved they are politically, the further apart they have moved. Polarization is not merely an elite phenomenon but extends downward through the citizenry to the degree that people are politically aware and active (Abramowitz, 2010, chapter 3).
Fiorina and Levendusky concede that ordinary partisans have indeed become more distinct in their political opinions and ideologies over the past few decades but prefer to describe the change as partisan sorting (Fiorina & Levendusky, 2006; Levendusky, 2009). Americans may not have grown more extreme in their views, but they have brought their partisan identities, ideological self-locations, and issue preferences into closer alignment. Thus, conservative Democrats and liberal Republicans are vanishing if not quite extinct breeds, and average partisan differences on many if not all issues have widened. Sorting may occur either when people choose to identify with the party that better fits their ideological and policy preferences or when they adopt the canonical positions of their chosen party. Levendusky (2009, chapter 6) argues that the latter is the more common path, but in either case the process is elite driven: Polarized elites give clearer cues about what the parties stand for, making it easier for people to chose the appropriate party or position. And certainly the timing demonstrates that, whatever the process, politicians polarized—or sorted—first, the public, afterward (Adams, 1997; Baumer & Gold, 2008; Carmines & Stimson, 1989; Jacobson, 2000; Levendusky, 2009).
Fiorina and Levendusky (2006) point out that sorting can occur without any change in the distribution of people’s opinions on political matters and hence argue against labeling the results of sorting “polarization” unless it coincides with an increase in the adoption of extreme positions. But sorting alone may foster polarized politics by generating distributions of opinion summed across sets of issues that are in aggregate more polarized and more extreme. This is essentially what, at the congressional level, has produced the increasing partisan differences in DW Nominate scores: As the House and Senate have split along party lines with increasing uniformity across a wide variety of issues, the aggregate pattern of roll-call votes, measured by the DW Nominate scale, has become more polarized. Even if ordinary people do not take sharply polarized positions on any particular issue, they may still end up at opposite ends of a scale constructed from multiple issues. Moreover, as reinforcing cleavages replace cross-cutting cleavages, 2 sorted partisans will find themselves in disagreement on issue after issue, perhaps to the point where they appear to inhabit separate cognitive and moral worlds (Hetherington & Weiler, 2009). To the extent that the other side is viewed as wrong about everything, elite intransigence (on one’s own side) will be admired, compromise disdained.
In this article, I demonstrate how sorting and aggregation can polarize the electorate and how their effects, compounded by electoral processes, sustain elite polarization. The evidence I examine supports the most common explanation for the failure to observe Downsian convergence—the greater extremism of primary electorates—and suggests that Mayhew’s “electoral connection” remains robust. My principal data source is the 2010 Cooperative Congressional Election Study (CCES; Ansolabehere, 2011), which provides both a very large national sample (more than 54,000 cases) and a rich array of questions regarding political attitudes, beliefs, and behaviors, including participation in both primary and general elections. 3
I begin by reviewing the distribution of opinion on nine distinct issue questions posed to CCES respondents. Then, I compare responses to these questions across categories of partisanship and political participation. I also examine how the distribution of responses to questions about Barack Obama, his legislative agenda, and several other political attitudes and opinions varies across these categories. I then create a pair of scales based on respondents’ issue positions and legislative preferences to show how sorting and aggregation can create a polarized electorate. Finally, I show that Republicans and Democrats in Congress hold their seats at the behest of distinct and ideologically distant electoral constituencies and of even more distant primary constituencies. Downsian convergence is thus thwarted, while the electoral connection remains robust.
Opinion on Issues
The 2010 CCES asked respondents for their opinions on a diverse set of social and political issues: the Iraq War, gun laws, dealing with climate change, immigration, abortion, the tradeoff between environmental protection and jobs, gay marriage, affirmative action, and how to deal with the federal budget deficit. Table 1 displays the specific questions and distributions of responses they elicited. 4 Where response categories allow a judgment, the responses do not suggest sharp public disagreement on these issues. When available, middling positions tend to be the most popular except when a plurality prefers one of the polar positions: support for abortion rights, opposition to affirmative action, and opposition to taxes to cut deficits. The more conservative position is the more popular in one of these latter cases, the more liberal position in two.
Issue Questions (CCES Question Identifiers in Parentheses)
Source: 2010 Cooperative Congressional Election Study.
Note: The number of cases for each question ranges from 47,216 to 47,882.
When responses are broken down by party identification and electoral participation, however, stark divisions appear, with voters more divided by party than nonvoters, and primary election participants even more divided than voters who participated only in the general election. 5 The evidence is displayed in the first section of Table 2. 6 Like other surveys relying on self-reported turnout (Burden, 2000; McDonald, 2003), the 2010 CCES overstates voting participation; 59% of respondents reported voting, whereas only 42% of eligible citizens actually cast ballots in 2010 (McDonald, 2011), so the first row of the table cannot be descriptive of the population. Nonetheless, comparisons across participation categories remain informative. The data show that a majority of partisans took opposing sides on nearly every issue, and the more involved they were in elections, the wider the gap. On average, in the main component of the first eight questions (the questions that are flush left, not indented), nonvoting partisans were 26 percentage points apart, partisans who voted only in the general election voters 44 points apart, and primary election voters 51 points apart. 7
Opinions of Partisans, by Voting Participation
Table 2 also reports the distribution of responses to some additional political questions across the parties and levels of participation. The distributions again reveal large differences in the average positions of partisans, with the parties’ respective primary electorates furthest apart. Barack Obama and his agenda provoked even wider partisan divisions than the political issues. In aggregate, Democrats and Republicans have strikingly divergent beliefs about Obama’s ideology, and their assessments of his performance are as far apart in this survey as were typical assessments of his predecessor, George W. Bush, during his second term; Bush has so far been the most polarizing president on record (Jacobson, 2011a), but Obama may end up matching him in this regard. Partisan difference in support for eight items on Obama’s legislative agenda varied by issue; health care reform produced the widest split, with partisan gaps of 82 points between the parties’ primary voters, 76 points between general-election-only voters, and 51 points between nonvoters. Obama’s economic stimulus package and Elena Kagan’s nomination to the Supreme Court were also very divisive. In every case, primary voters show the widest partisan differences on the agenda items, and nonvoters the smallest; the average partisan gap for the eight agenda items is 63 points for primary voters, 55 points for general-election-only voters, and 33 points for nonvoters. 8 Notable partisan differences, again increasing with participation, also emerge on several other indices derived from the CCES data: a version of the racial resentment scale, 9 mean self-location on the 7-point liberal-conservative scale, 10 opinions of the Tea Party movement, and religiosity. 11
Opinions Aggregated Across Issues
By this evidence, partisans were quite well sorted according to their political views in 2010, and the more involved they were in electoral politics, the more consistently they took the modal positions of their preferred party. 12 The opinions of primary voters are also more closely related to one another (average correlation between pairs of items, .42) than those of general-election-only voters (average correlation, .31), but opinions of both types of voters are considerably more closely related than those of nonvoters (.14). A factor analysis of the responses to the issue questions by each of three participation categories indicates that a single latent variable underlies responses to these issues questions. For each category, the analysis uncovers a single factor, 13 but the eigenvalue and the average factor loadings are highest for primary voters (3.83, .64), somewhat lower for general-election-only voters (2.88, .55), and considerably lower for nonvoting partisans (1.38, .37). Thus, the greater a person’s involvement in electoral politics, the more ideological “constraint” he or she displays (Converse, 1964). For all respondents analyzed together, the eigenvalue for the first factor is 2.71, 14 and the average absolute factor loading is .57. The results are reported in Table 3.
Factor Analysis of Issue Question Responses
Note: Eigenvalue = 2.71, N = 53,522.
Individual factor scores computed from these results serve to locate respondents on a scale representing the latent variable underlying the extracted factor, unambiguously interpretable in this case as a liberal-conservative dimension. Average factor scores across participation categories (adjusted to range from -1 to 1 for clarity) are shown in the bottom row of the first section of Table 2. The kernel density distribution of respondents (including now independents as well as partisans) along the scale is displayed in Figure 1. The distribution for nonvoters is strongly unidimensional, reiterating the finding accepted by scholars on both sides of the debate that the politically inert, at least, have not polarized along party lines or any other dimension (Abramowitz, 2010; Fiorina, 2009). The distribution among general election voters is, in contrast, clearly bimodal, and the subset of voters who also participated in a primary is even more so. Primary voters clearly differ in this regard from voters who did not participate in primaries in states and districts where this was an option; for this latter subset of respondents, the distribution is barely bimodal. The figure also displays the distribution for people in an additional subcategory of participation, those who engaged in one or more political activities in addition to voting 15 ; the distribution of these activists is the most sharply bimodal and polarized of all.
The Distribution of Factor Scores of Partisans, by Electoral Participation
Individuals’ locations on the scale are of course strongly related to party identification. Among primary voters, 89% of Democrats had negative scores, and 93% of Republicans had positive scores; the respective figures for general election voters are 88% and 87%, for nonvoters 80% and 71%, and for activists 93% and 95%. That is, partisan and ideological differences coincide.
These results demonstrate that, independent of partisan differences on specific issues, partisan sorting can produce a polarized distribution of responses when opinions are summed across a diverse range of issues. They also show that the degree of polarization varies by level of political activity. The largest change in the shape of the distribution occurs in the move from the nonvoter to the voter category; the former show no signs of aggregate polarization, the latter clearly do.
A similar pattern appears when we examine levels of support for Obama’s legislative agenda. Figure 2 shows distribution of the frequency with which respondents supported the administration’s position on the eight items. Among nonvoters the distribution is again unimodal, although with the mode at the supportive end; 42% of nonvoters supported Obama’s position on at least seven of the eight items, while only 8% supported it on no more than one item. A total of 63% of general election voters were at these extremes (supporting no to 1 item, or seven to eight items); among primary voters, it was 66%, with 43% supporting either all or none of the agenda items. Nearly three quarters of the activists supported or opposed Obama on at least seven of the eight items, and for a majority of activists, it was all or none.
Distribution of Support for Obama’s Agenda, 2010
The Polarization of Electoral Constituencies
The distribution of issues positions and roll call vote preferences reveals an electorate in which issue cleavages tend strongly to reinforce one another, leaving majorities of Republican and Democratic voters consistently on opposite sides. The electoral process compounds these partisan differences and provides a foundation for the partisan polarization observed in Congress. Although the full constituencies represented by the winning Republicans and Democrats do not differ sharply in the distribution of aggregate opinions, the congressional parties’ respective electoral constituencies—defined as those respondents who reported voting for the winning Republican or Democrat—are far apart, and their primary electoral constituencies even more so. This is evident from Figure 3, which compares the distributions of adjusted factor scores in House districts won by Democrats and Republicans in 2010. The distributions for the full constituencies of the two congressional parties are distinctive but overlap a great deal; the difference in average scores is 0.14 points on the 2-point scale (−1 to +1). 16 There is much less overlap between the congressional parties’ electoral constituencies, and the gap between their average locations is 0.61 points. Primary electoral constituencies—consisting of respondents who voted for the representative and who participated in the representative’s party’s primary—are even more distinct, with an average difference of 0.89 points on the scale. And the activists in the congressional parties’ electoral constituencies (not displayed to avoid clutter) were an even wider 0.98 points apart.
Factor Scores for Electoral Constituencies
A similar pattern emerges in average levels of support for Obama’s agenda (Table 4). Full constituencies differ by less than one item on average, but voters who supported the victorious Republicans and Democrats differ by 4.2 items, those who also voted in the primary by 5.5 items, and activists among electoral constituents by 6.1 items. Thus, although red and blue districts as a whole are not particularly polarized on issues or roll-call votes, voters in the constituencies who were responsible for electing each congressional party’s contingent are quite polarized, and the parties’ respective electoral constituents who participate in primaries are even more so.
Support for Obama’s Agenda, by House Constituencies (Average Number of the Eight Items Supported)
According to the distributions in Figure 3, Republican electoral constituencies lean clearly to the right, and Democratic electoral constituencies lean clearly to the left, corresponding to the respective locations of the Republican and Democratic representatives they have elected. Although the issue scales are not on the same metric as DW Nominate scores, the patterns observed in Figure 3 do raise serious doubt that there is a disconnect between moderate voters and extreme representatives. 17 The data in Table 5, which displays average locations on a 7-point liberal-conservative scale chosen by respondents for themselves, the House candidates, and the parties, provide further reason to question the disconnect. In every instance but one (on average, nonvoting Democrats locate themselves slightly to the right of their party), respondents typically locate themselves further from the ideological center than their candidates or parties; Republicans typically place themselves to the right of their candidates and party, Democrats place themselves to the left of theirs. Clearly, partisans did not typically view their candidates or party as more extreme than they are. They do, however, view the other party and its candidates as considerably more extreme than the party’s own adherents view it, and the more active they are in elections, the more ideologically distant they perceive the other party and candidate to be.
Average Placements of Self, Candidates, and Parties on the 7-Point Liberal Conservative Scale
Note: Scale ranges from 1 (very liberal) to 7 (very conservative), with 4 as “middle of the road”; data are from cc334a, cc334d, cc334e, cc334j, and cc334k.
Data from a 2010 survey can offer only a snapshot, but the idea that partisan divisions in Congress reflect an ideological disconnect between representatives and the people who elect them is contradicted by time-series data as well. If elite polarization had separated representatives from the more moderate citizens who elect them, we would expect the distance between respondent’s self-described ideological locations and the locations they estimate for their party and its candidates to have grown over time. National Election Studies (NES) data show that it has not; what has grown is the distance between ideological self-locations and placement of the other party and its candidates. Increasingly polarized representatives are not seen as more distant by their own partisans, but they are viewed as substantially more distant ideologically by those on the other side (Jacobson, 2011a). Just as in the 2010 CCES, it is the other party and its politicians, not their own, that people have come to see as extreme. 18
Failure to Converge
The 2010 CCES data support a common explanation for candidates’ and parties’ failure to adopt positions close to those of the median voter: the need to appeal to primary electorates in which ideological extremists are overrepresented. The reality of this predicament was underlined during the 2010 Senate campaigns, during which Tea Party favorites challenging from the right denied two incumbent Republican Senators renomination and in several other states defeated more mainstream conservatives supported by the Republican establishment. The difficulties this can create for a party was also demonstrated in 2010, when the nomination of far-right extremists cost the Republicans certainly two and possibly three Senate seats (Jacobson, 2011b, p. 39).
The data displayed in Tables 2 and 4 and in the figures suggest that the primary electorate’s pull toward the extreme was stronger among Republicans than among Democrats in 2010. The differences between general election and primary electorates are wider on the Republican side on virtually all of these measures, as is evident from visual inspection of all three figures as well as the data in Table 2. Republican primary voters are an average of 5.6 points to the right of other Republican voters on the core questions in the first section of the table (those not indented) and an average of 7.1 percentage points less supportive of the Obama’s agenda. The comparable differences among Democrats average 1.9 and 1.0 percentage points, respectively.
The greater extremism of Republican primary electorates reflects in part the peculiarities of 2010, in which Tea Party sympathizers, a great majority of them very conservative Republicans or independent-leaning Republican, were the most enthusiastic participants. 19 But it is consistent as well with two longer term components of the polarization process. First, most of the congressional polarization documented by DW Nominate scores was produced by Republicans becoming more conservative rather than by Democrats becoming more liberal; 82% of the increase in polarization in the House between the 92nd Congress (1971-1972) and the 111th Congress (2009-2010), and 69% of the increase in Senate polarization, reflects increasing Republican conservatism. 20 Second, data from the NES time series indicate that the growing ideological gap between Republican and Democratic identifiers has been more a consequence of Republicans moving right than of Democrats moving left (Jacobson, 2012, pp. 97-98). Thus, the Tea Party phenomenon may be simply the most recent manifestation of the increasing conservatism of Republicans at both the elite and mass levels. Republican intransigence in the budgetary wars of 2011 certainly reflected fear of retaliation in the primary from the Party’s Tea Party wing, but with a large majority of voting Republicans favorably disposed toward the Tea Party movement (Table 2), this did not necessarily disconnect Republican representatives from their electoral base.
Conclusion
To summarize: the 2010 CCES offers considerable evidence that ordinary Americans as well as political elites divide consistently along party lines over a broad range of issues and legislative preferences. The coincidence of partisan differences across issues results in a bimodal distribution of aggregate preferences among voters, showing that partisan sorting can by itself polarize the electorate even if sorted partisans do not adopt extreme positions on individual issues. Partisan polarization is by no means universal, but clear ideological divisions are not confined to the political class; they extend downward into the citizenry to the degree that people are actively involved in politics. The distribution of partisan preferences is even more sharply bimodal and polarized when the comparison is between the electoral constituencies of the congressional parties, so the electoral process itself contributes to elite polarization. And because electoral constituencies determine the fates of representatives, ideological differences between the congressional parties have a firm electoral foundation. Finally, primary electoral constituencies tend to be even more extreme, particularly on the Republican side, deterring departures from party orthodoxy and thus movement toward the median voter.
Despite their disagreements, Fiorina and Abramowitz would probably accept the empirical evidence offered here as generally valid. Their dispute is actually much more over the interpretation of the evidence than the evidence itself—of necessity, as they both rely heavily on the same NES time series data to make their points. Both could draw support from the 2010 CCES data for elements of their arguments. For example, Abramowitz could point to the bimodal distribution of voters on the two scales, indicating that partisan ideological divisions extend well beyond political elites, although not to the politically uninvolved. He would also find support for his argument that partisan sorting can by itself generate polarization. Fiorina could find confirmation of his view that electoral institutions—particularly primary elections, but also election processes more generally—contribute to keeping elites from converging on a middle ground: members of Congress who faithfully represent their electoral coalitions do not necessarily represent their full constituencies nor, collectively, the average American. The electoral process’s capacity for perversity is illustrated by the fact that, insofar as voters were expressing a desire for more moderate national policies in 2010, as rational balancing theories of midterm voting behavior suppose (e.g., Alesina & Rosenthal, 1989), they did so by voting out the moderates. 21 The 112th Congress was consequently even more polarized than the 111th, its divisions exemplified by the potentially calamitous showdown over deficit reduction and the debt ceiling in the summer of 2011.
Footnotes
Declaration of Conflicting Interests
The author declared no conflicts of interest with respect to the authorship and/or publication of this article.
Funding
The author received no financial support for the research and/or authorship of this article.