Christopher Hare is a PhD student in Political Science at the University of Georgia.

Howard Rosenthal is Professor of Politics at New York University and the Roger Williams Straus Professor of Social Sciences, Emeritus, at Princeton University.

The recent outburst of scholarly and popular interest in political polarization has attracted attention to the methods we use to measure this phenomenon. One frequently voiced concern (see a recent column by Sean Trende) is that Congress may not have polarized as we have claimed in publications and blogs stretching as far back as 1984. The concern is that the meaning of ideological (NOMINATE) scores are tied to the legislative and historical context of the roll call votes that are used to estimate them. For example, the content of roll calls votes cast by members of 90th Senate that dealt with the Vietnam War, civil rights, and funding for LBJ’s “Great Society” programs are quite different than those votes cast in the current Senate. Thus, being the most conservative Senator (with a score of 1.0) in 1968 would mean something different than having an identical 1.0 score in 2012.

Indeed, temporal comparisons should not be made for ideal points generated from static scaling methods. Static methods (like W-NOMINATE) treat each legislative session separately and there is no valid way to compare the scores of legislators from different years. However, we developed a dynamic methodology, DW-NOMINATE (McCarty, Poole and Rosenthal 1997), to allow for over-time comparisons of legislator ideological positions. The key innovation is the use of “bridge” legislators — members of Congress (MCs) who have served in multiple sessions — to compare the positions of legislators who have never served together.

A sports analogy to the overlapping cohorts method is the “common opponents” statistic. If we want to compare two teams who have not played each other, we compare their performances against a common opponent(s). Likewise, MCs who have not served together can be compared with the use of a “bridge” legislator who has served with both. For example, if we know that Sen. George McGovern (D-SD) is more liberal than Sen. Patrick Leahy (D-VT), and that Sen. Leahy is more liberal than Sen. Max Baucus (D-MT), then we can say that Sen. McGovern is more liberal than Sen. Baucus. Though intransitivities may arise cases involving 3 or more sports teams, Poole shows in his 2007 Public Choice article “Changing Minds? Not in Congress!” that MCs remain remarkably static in their ideological positions over the course of their careers. Thus, we are on much firmer ground in making over-time comparisons between MCs with the caveat that we cannot compare members outside of one of the stable, two-party periods of American history. For that reason, when we discuss current polarization we focus on the period from the end of Reconstruction in 1877 to the current period.

With the use of overlapping cohorts, we can make the over-time comparisons needed to analyze polarization. A good example is Senator Richard Lugar (R-IN), who, after his primary defeat last week, will have served in the Senate between 1977 and 2013. As David Karol points out, Lugar himself did not change very much over time: he was a reliable conservative who moved only somewhat towards the center during a 30-plus year career (from a DW-NOMINATE first dimension score of 0.348 to 0.241). DW-NOMINATE scores range (with slight simplification) from -1 to +1 or a band of two units. So in 30 years, Senator Lugar moved just five percent on the liberal-conservative dimension.¹

For Lugar, what is more dramatic is the change in his ideological position relative to the Senate Republican Caucus. In his first term in Congress, Senator Lugar was the 23rd most moderate Republican in the Senate; in the most recent term (through 2011), he was the fifth most moderate. Even if he had maintained his freshman score of 0.341, he would still have been the 12th most moderate Republican in the 112th Congress. This repositioning occurred because almost every new cohort of Republican Senators has been more conservative than Senator Lugar. That fact is the basis for our claim that the Republican party has moved to the right.

Indeed, we find that contemporary polarization is not only real — the ideological distance between the parties has grown dramatically since the 1970s — but also that it is asymmetric — congressional Republicans have moved farther away from the center than Democrats during this period. In two figures below, we plot the mean first dimension DW-NOMINATE scores of the two parties in the House and Senate from 1879 to the present. Since the mid-1970s, Republicans have moved further to the right than Democrats have moved to the left. This rightward shift is especially dramatic among House Republicans, from a mean of 0.22 in 1975 to 0.67 in 2012.

To be sure, political polarization is not entirely asymmetric. Congressional Democrats have moved slightly to the left during this period, but most of this is a product of the disappearance of conservative Southern “Blue Dog” Democrats. But the northern Democrats of the 1970s are ideologically indistinguishable from their present-day counterparts, with average scores around -0.4.

Though Democrats have not moved nearly as much to the left as the Republicans have to the right, they have also contributed to polarization, in our opinion, by embracing identity politics as a strategic tool. In Roosevelt’s New Deal, the Democrats advocated redistribution and regulation of business. These issues remain active to some extent, but with time emphasis has shifted to issues centered on race, gender, ethnicity, or sexual preference (Gerring 1998). As this issue evolved, it mapped onto the existing liberal-conservative dimension. The mapping is marked by members of the Black Caucus anchoring the liberal end of the dimension. What our roll call analysis shows is that Democrats did not vote much further to the left on the new issues than on New Deal issues. The comparison works because some New Deal issues, such as minimum wages and regulation of the financial sector, continue to lead to roll call votes.

Nonetheless, we should be careful not to equate the two parties’ roles in contemporary political polarization: the data are clear that this is a Republican-led phenomenon where very conservative Republicans have replaced moderate Republicans and Southern Democrats. Thomas Mann and Norm Ornstein do an excellent job of navigating these trends in their new book: It’s Even Worse Than It Looks: How the American Constitutional System Collided With the New Politics of Extremism.

Moreover, the rise of the “Tea Party” will likely only move Congressional Republicans further away from the political center. For example, the five Tea Party-backed Senators elected in 2010 (Senators Rubio, Paul, Toomey, Lee, and Johnson) have an average first dimension DW-NOMINATE score of 0.795. Moderate MCs (especially Republicans) are increasingly likely to be “primaried” out (e.g., Sens. Bob Bennett (R-UT), Joe Lieberman (D-CT), and Lisa Murkowski (R-AK), as detailed in a recent post on voteview.com).

The public policy consequences of polarization are immense. Bipartisan agreements to address looming issues like the budget deficits, spending on entitlement programs, and immigration are now almost impossible to reach. In contrast, during Ronald Reagan’s administration, about 40% of the members of Congress could be described as moderates. Reagan was thus able to forge major bipartisan agreements to cut taxes in 1981, raise taxes in 1982, fix Social Security (the Greenspan Commission) in 1983, and pass immigration reform (which included amnesty) and major tax simplification in 1986.

As shown in the second pair of figures below, only about 6% of Representatives and 13% of Senators in the 112th Congress can be described as moderates (defined as having a first dimension DW-NOMINATE score between -0.25 and +0.25). This absence forces major legislation, such as President Obama’s health care package, to be passed by one party. But unlike major bipartisan efforts (e.g., the Social Security Act of 1935, the Civil Rights Act of 1964, or welfare reform in 1995), legislation passed by one party is less likely to earn popular acceptance (as evidenced by the partisan breakdown in opinion on “Obamacare”).

Polarization is real. Arlen Specter was reelected to the Senate as a moderate Republican in 2004. In the 2010 election, he was replaced by Pat Toomey. Do academics and pundits really want to argue that Republicans have not moved to the right and that Pat Toomey might be more moderate than Arlen Specter because the congressional agenda has changed? Let’s not get picky about polarization. It’s for real, and it is making the United States dysfunctional.

   ¹ Legislators’ DW-NOMINATE scores are allowed to move as a linear function of time, while a single coordinate is estimated for each legislator with the Common Space procedure; methodological issues aside, the linear and constant methods produce yield the same pattern of contemporary political polarization.

References:

Gerring, John. 1998. Party Ideologies in America, 1828-1996. Cambridge: Cambridge University Press.

McCarty, Nolan M., Keith T. Poole, and Howard Rosenthal. 1997. Income Redistribution and the Realignment of American Politics. AEI Studies on Understanding Economic Inequality. Washington, DC: AEI Press.

The vote was largely party-line with five Democrats voting “Yea” and twenty-seven Republicans voting “Nay.” The Democratic defections do not appear to be ideologically-driven, but the twenty-seven Republicans who broke party ranks are generally less conservative than their counterparts who supported the amendment.

Click image to enlarge

Click images to enlarge

To address part of the question of the role of the party primary system in the overall decline of moderates, we focus in this post on whether incumbents who are defeated in the primary are more likely to be ideologically centrists, whether those who defeat them are more ideologically extreme, and if there are differences between the parties (for an innovative study related to this question, see Aaron King, Frank Orlando, and David Sparks’s analysis using the Twitter accounts of primary candidates to estimate their ideological location).

Below we plot the DW-NOMINATE Common Space scores (which allow for cross-comparability between the House and Senate) of the 19 congressional incumbents (11 Republicans and 8 Democrats; 5 Senators and 14 Representatives) who were defeated in their party primary between 2006 and 2012. We also include a rug plot at the bottom of the graph of the first dimension scores of all members of Congress during this period (Democrats in blue ticks, Republicans in red ticks) to show the overall ideological distribution of the parties. For both parties, defeated incumbents are slightly more moderate as a group than their partisan counterparts, though the difference of the mean (ideological) first dimension scores between the two groups is larger among Republicans (0.07, t = 1.22) than Democrats (-0.04, t = -0.87).

What is more dramatic is the difference between successful primary challengers and defeated incumbents. In the second plot below, we use an arrow to show the difference between the DW-NOMINATE Common Space scores of incumbents and the primary challengers who have defeated them (the arrow pointing to the location of the challengers). We lack DW-NOMINATE scores for 6 of these 19 challengers because they were not elected (most of these six ran from the ideological flanks of the party – e.g., Republican Richard Mourdock of Indiana and Ned Lamont of Connecticut – and so data on them would likely strengthen out findings). Of the 5 challengers who defeated Democratic incumbents, 2 were more liberal than the incumbent, 2 more conservative, and for one there is no substantive difference (mean difference = 0.01, t = 0.06). Of the 8 Republican challengers, however, 6 were more conservative than the incumbent and only 2 more liberal, an average shift of 0.20 units to the right (t = 2.23). Though preliminary and based on a small number of cases, these results lend some support the idea that moderate Republicans in Congress are more imperiled in the primary process than are moderate Democratic incumbents (whose ranks have been cut more heavily by retirements and in general elections).

Click images to enlarge

This provides a partial explanation for polarizing trends in Congress. Below we show the party means on the first (ideological) dimension over time. The Republican mean has moved steadily to the right since the mid-to-late 1970s, and the effects of the primary process are consistent with this trend.

Click images to enlarge

Though both votes — in particular the highway funding/Keystone pipeline vote — featured a large number of partisan defectors (69 Democrats joined all but 14 Republicans to support the transportation bill; while 18 Democrats and 10 Republicans defected on the Small Business Tax Cut Act), the two-dimensional spatial (geometric) model of choice performs well in classifying members’ votes based on their ideological preferences.

On the vote to mandate construction of the Keystone XL pipeline, all moderate Democrats voted “Yea,” and the cutting line separates these approximately two dozen members from the more sizable liberal bloc of House Democrats. It is worth noting that a considerable number of House Democrats to the left of the cutting line supported the bill (for example, Congressional Progressive Caucus members Reps. Ed Pastor (D-AZ) and Chaka Fattah (D-PA)), illustrating the challenge facing President Obama’s to maintain a political coalition to oppose the proposed construction of the Keystone XL pipeline.

The Republican Small Business Tax Cut Act attracted some, though not nearly as much, Democratic support as the highway funding bill: 18 generally moderate Democrats voted “Yea”, but 10 conservative Republicans voted “Nay” because they instead prefer a long-term tax reform measure that overhauls the current tax code. The Senate, which rejected a cloture motion on the “Buffett Rule” on Monday, is not expected to bring up the Small Business Tax Cut Act for a vote.

Click images to enlarge

Foremost among the ignored issues are those that concern America’s unsustainable fiscal trajectory. The US Congress Joint Committee on Taxation estimates that adoption of the “Buffett Rule” would generate an additional $47 billion in revenue over the next ten years. In comparison, the federal government spent $240 billion on interest on the national debt in 2012 alone. The total 2012 national budget deficit is $1.10 trillion, and the national debt now totals $15.66 trillion.

But even these figures obscure some of the more structural fiscal challenges facing the United States. As a nation, we are now technically bankrupt: the Federal Reserve estimates the total net worth of all American households at $58.46 trillion; while we face $62.30 trillion in unfunded liabilities and promises, the major source of which are entitlement programs like Social Security and Medicare.

Cutting spending on these programs is extremely unpopular. According to a March 2012 Economist/YouGov poll, only 10% of voters approve cuts to Social Security or Medicare. Even among conservatives, only 19% favor cutting Social Security spending and 18% support Medicare cuts.

Reducing the budget deficit by raising taxes on the rich polls better, but the disparity between existing shortfalls and the amount that can be raised through this approach is sobering. In 2008, the taxable wealth of millionaires totaled $0.93 trillion (of which they paid $0.25 trillion in taxes). This leaves a pool of about $680 billion which can be taxed from those making over $1 million a year.

The United States faces serious fiscal challenges, some of which are more quickly approaching than others. Polarization contributes to the dysfunction of the political elites by not only making hard compromises more difficult to reach, but also by obscuring important debates and choices in favor of a year-round “silly season” in political dialogue.

The vote was party-line with only two exceptions: Senator Mark Pryor (D-AR) voted “Yea” and Senator Susan Collins (R-ME) voted “Nay.” These two Senators are among the most moderate members in the chamber, and from a spatial standpoint are among the most likely to most likely to defect from their co-partisans. Indeed, Optimal Classification correctly classifies Sen. Collins as a “Nay” vote, placing the cutting line (dividing predicted “Yea” from predicted “Nay” votes) just to her right.

Later this week, the House is expected to vote on (and likely approve) a Republican proposal to issue a one-year, 20-percent tax deduction for small businesses (businesses with less than 500 employees).

Click image to enlarge

In the ANES 2010-2012 EGSS 4, respondents were asked to place themselves and six political stimuli–the Republican and Democratic parties, the Tea Party, President Barack Obama, Mitt Romney, and Newt Gingrich–on a seven point ideological scale, from very liberal to very conservative. We first analyze this data with the Aldrich-McKelvey scaling procedure using the “aldmck” function. The Aldrich-McKelvey procedure extracts a latent dimension with the stimuli positions and then estimates simple linear maps of the respondents onto this ideological (liberal-conservative) dimension.

The Aldrich-McKelvey procedure divides respondents based on whether they have positive or negative weights. Respondents have negative weights if they see the space backwards– that is, if they perceive President Obama as being more conservative than the Republican Party. This is a strength of the Aldrich-McKelvey procedure, since it acts as a filter for individuals with very low levels of political information (as shown by Palfrey and Poole (1987)). The first plot show the distribution of respondents with positive weights, the second those with negative weights. Note that those respondents with positive weights (who are more politically sophisticated are more likely to be politically engaged) approximates a bimodal distribution where voters are ideologically polarized. Conversely, in the second plot, respondents with negative weights are located primarily in the political center– but it is these citizens who have lower levels of political sophistication and less likely to vote (furthering political polarization).

In addition, the locations of the stimuli on the ideological continuum are interesting. Note that President Obama and the Democratic Party are placed at virtually identical locations, while the four Republican stimuli are tightly clustered– with Mitt Romney closest to the center among these and the Tea Party furthest to the right.

Click images to enlarge

We next use the “blackbox_transpose” function to scale the same six political stimuli in two dimensions from respondents’ ideological rankings of each candidate or party. Poole’s Basic Space methods are a generalization of the Aldrich-Mckelvey procedure, and can handle missing data–even at high levels–and estimate the locations of stimuli and respondents in multiple dimensions. The “blackbox” function handles preferential data (where respondents express choices), while the “blackbox_transpose” is appropriate when dealing with perceptual data (where respondents locate stimuli using the scales). The “blackbox_transpose” function produces the following plot which locates the six stimuli in latent, ideological space. The first dimension (the liberal-conservative continuum) explains most (about 86%) of the variance in respondents’ evaluations of these political figures, with the Tea Party perceived as the most conservative and President Obama as the most liberal. The second dimension appears to be picking up differences between traditional Republicans (most prominently, Mitt Romney) and the Tea Party. Interestingly, this parallels a pattern also found in recent congressional roll call voting.

Click image to enlarge

There are two aspects of this vote that are interesting. First, in contrast to the recent Senate vote to approve $109B in highway funding that split the Republican caucus, this vote exposes an ideological divide among the Senate Democrats. The 22 “Nay” vote on the cloture motion drew mostly from among the most liberal bloc of Senators (e.g., Sens. Barbara Boxer (D-CA), Tom Harkin (D-IA), and Bernie Sanders (I-VT). The Optimal Classification procedure performs well in using the first dimension (representing the liberal-conservative scale) to split the group of more liberal “Nay” Democratic Senators from their more moderate “Yea” counterparts, producing only six classification errors (all close to the cutting line) with a PRE (proportional reduction in error) of 0.73.

Second, the location of the cutting line (which separates predicted “Yea” votes from predicted “Nay” votes) for the Senate vote to invoke cloture on the JOBS Act is strikingly similar to the House’s 390-23 vote on passage of the bill. In both cases, the cutting line runs along the left edge of the Democratic caucus. We include both plots below to allow comparison:

Both votes are good illustrations of the validity of spatial (geometric) voting models and how they provide a richer layer of information about legislative behavior beyond the party affiliation.

Click images to enlarge

Support for the bill was largely bipartisan, with all Senate Democrats voting “Yea” and Senate Republicans splitting 22-22 on the vote (with 3 abstentions). Here, the spatial (geometric) model of choice performs well in picking up the split within the Republican caucus based on the extent of anti-spending, anti-tax sentiment. On one side, GOP Senator James Inhofe (R-OK) was quoted as saying: “I’ve always said conservatives should be big spenders in two areas: national defense and infrastructure. We have to look at the future so we don’t have to go through this again.”

Conversely, the leading Tea Party Senators (Sens. Jim DeMint (R-SC), Ron Johnson (R-WI), Mike Lee (R-UT), Rand Paul (R-KY), and Pat Toomey (R-PA)) all voted “Nay” on the spending bill. Sen. Toomey was quoted as saying: “With a $15 trillion national debt, I cannot support a bill that plunges us further into debt and economic uncertainty. I hope this bill will become more fiscally responsible in conference and something I can support.”

The cutting line (which separates predicted “Yea” from predicted “Nay” votes) separates many of the more conservative Republicans from the more moderate ones, but also runs along the second dimension. The substantive meaning of the second dimension in contemporary American politics remains unclear, though it has been important in modeling votes like the payroll tax cut extension and raising the debt ceiling. Note that most of the anti-establishment legislators (the aforementioned Tea Party Senators on the right, or figures like Sen. Bernie Sanders (I-VT) or Rep. Dennis Kucinich (R-OH) on the left) have low second-dimension scores.

Click image to enlarge

Both votes divide Democratic members: 11 moderate Senate Democrats and 23 liberal House Democrats broke from the majority of the party in these votes. Consequently, these votes are good examples of the utility of the spatial (geometric) model of choice in modeling roll call voting behavior. OC performs well on both votes, drawing a vertical cutting line to divide members of Congress by their first dimension (x-axis) location (which represents the standard liberal-conservative scale). Indeed, this makes sense: both proposals involved questions of the desired extent of government intervention in the economy.

In the Senate’s Keystone vote, the cutting line runs through the moderate wing of the Democratic caucus; while in the House’s JOBS Act vote, the cutting line divides the most liberal bloc of House Democrats from the remainder of the party. The major outlier in the Senate vote was Senator Ben Nelson (D-NE), who voted “Nay” but was classified as a “Yea” vote (though in this case ideological factors were complicated by local concerns about how the routing of the Keystone XL pipeline through Nebraska).

Click images to enlarge