Keith T. Poole is the Philip H. Alston Jr. Distinguished Professor in the Department of Political Science at the University of Georgia and Christopher Hare is a Ph.D. candidate in Political Science at the University of Georgia. They, along with Dave Armstrong, Ryan Bakker, Royce Carroll, and Howard Rosenthal, are co-authors of the forthcoming book Analyzing Spatial Models of Choice and Judgment with R.
To the extent that any topic in the social sciences can be considered settled, the polarization of American political elites almost certainly qualifies. In Congress, the distance between the parties’ ideological positions–whether measured using roll call voting records or campaign contributions–has continued to rise at a record-setting pace since the 1970s. The ideological center in Congress, the driving force behind major policy reforms as recently as the 1980s, has hollowed out.
A much more contested question concerns the degree to which the mass electorate can also be described as polarized. Political scientist Morris Fiorina and his colleagues (especially with their book Culture War? The Myth of a Polarized America) have been the leading advocates for the position that ordinary Americans remain predominantly centrist and are alienated by elite polarization. While it is true that most Americans consider themselves to be near the center of the ideological spectrum (either moderates or “slightly” liberal or conservative using the standard seven-point liberal-conservative scale), there are two reasons to expect that many of these moderates are illusory. In truth, the American electorate is more polarized ideologically than it might seem at first glance. And that is especially true for those voters who are more knowledgeable about politics.
First, moderates possess lower levels of political information and are less likely to be politically engaged than those who are closer to one of the ideological poles. As a result, centrists are underrepresented in the electorate. Emory University political scientist Alan Abramowitz has extensively detailed this phenomenon in his book The Disappearing Center.
Second, the term “moderate” is a favorable term in politics, one that connotes pragmatism and reasonableness, as opposed to epithets like ideologue, radical, or extremist. Accordingly, many individuals describe themselves and their preferred candidate and party as “moderate.” Indeed, in the 2012 American National Election Study (the data we use here), 31 percent of Obama voters who placed both themselves and Obama on the liberal-conservative scale rated both themselves and Obama as “moderate” or “slightly liberal,” while 21 percent of Romney voters rated both themselves and Romney as “moderate” or “slightly conservative.” Of course, if these respondents view themselves as sharing the ideological views of the Democratic or Republican parties, they are probably not really centrists. This is a common problem in survey research known as “differential item functioning,” and it occurs any time that respondents interpret scales differently.
We use the Aldrich-McKelvey scaling procedure to uncover both types of moderates: the politically uninformed/unengaged and the “moderates-in-name-only.” Put differently, we use respondents’ placements of common objects (e.g., the Republican and Democratic parties) to estimate how they distort the space and correct for bias in their self-placements. For example, someone who rated herself and Obama as “moderate” and placed Romney at the right-most, “extremely conservative” category would be moved to the left.
The scaling procedure also essentially “sorts” respondents by their level of political information. For example, if they place Romney and the Republican Party to the left of Obama and the Democratic Party, they have a very low level of political information. Conversely, respondents who can correctly place the objects are more likely to be politically informed and participate in the political arena.
The plot below shows respondents’ self-placements on the liberal-conservative scale in the left panel, and we see a familiar bell curve pattern where most respondents place themselves in the middle and only a small proportion consider themselves to be “extremely liberal” or “extremely conservative.” This paints a clear picture of a centrist electorate.
The right panel shows the distribution of respondent ideal points estimated by Bayesian Aldrich-McKelvey scaling, which corrects for bias in respondents’ self-placements. The distribution of all respondents is shown in black, while the gray line shows only respondents who correctly place the liberal stimuli to the left of the conservative stimuli (in technical terms, have positive weights in the scaling). Both distributions exhibit greater polarization than the raw self-placements. There is still a peak at the center, but there are also modes around the estimated locations of Obama and Romney (the blue “O” and the red “R” tokens). Many moderates also disappear when we look at only respondents who place the parties appropriately, and the heights of the three modes are nearly equal in this distribution.
Another way to measure polarization from the two sets of ideology scores (the raw self-placements and the adjusted ideal points) is to calculate the overlap between partisan groups and Obama/Romney voters indicated by each set of scores. The more Obama and Romney voters overlap, the lower the level of polarization. Political scientists Matt Levendusky (University of Pennsylvania) and Jeremy Pope (BYU) have pioneered this approach as an alternative measure of polarization between two opinion distributions. The overlap coefficient ranges between 0 and 1, with higher values indicating greater overlap and lower polarization. Below are some examples of what the overlap coefficients look like in three different cases.
We calculate the overlap between Obama and Romney voters and between Democrats and Republicans (including leaners) on the liberal-conservative dimension using each of the two measures: the raw self-placements and the DIF-corrected ideal point estimates. As we might expect, the overlap between both groups is smaller—and thus polarization is greater–using the corrected estimates of ideology as compared to respondents’ own self-placements on the liberal-conservative scale. The differences are statistically significant (see graph below) and substantial in magnitude: a 38 percent drop (from .45 to .28) in the overlap between Democrats and Republicans and a 46 percent drop (from .48 to .26) in the overlap between Obama and Romney voters when moving from raw self-placements to the BAM ideal point estimates. Once we account for differences in how Americans define “liberals,” “moderates,” and “conservatives,”, we get a picture of a more polarized electorate than if we rely on raw self-placement data.
These results are a classic instance of “a glass half full vs. a glass half empty.” Compared to members of the House and Senate, the public is clearly more moderate. The overlap for the House and Senate is zero. There are still moderates in American politics, but they are in the electorate and not in Congress. Still, we cannot characterize the American public as overwhelmingly centrist—doing so is more an artifact of survey responses than an apt description of Americans’ ideology.
This is the latest post in our ongoing series on political polarization. The previous posts are listed below. -Dan Hopkins