What Has Made Congress More Polarized?

By Sean Trende - May 11, 2012

‹‹Previous Page |1 | 2 | 3 |

Thomas Mann and Norm Ornstein ignited a bit of a firestorm with their column describing Republicans as main drivers behind Washington's problems. The Washington Post's Chris Cillizza has made some excellent points with regard to Mann and Ornstein's qualitative arguments, and other thorough responses abound. Rather than revisiting these points, I would like to focus my attention on the quantitative arguments made at the end of the article. Mann and Ornstein note that:

“[P]olitical scientists Keith Poole and Howard Rosenthal, who have long tracked historical trends in political polarization, said their studies of congressional votes found that Republicans are now more conservative than they have been in more than a century. Their data show a dramatic uptick in polarization, mostly caused by the sharp rightward move of the GOP.”

Mann and Ornstein are referring to data generated by Poole and Rosenthal’s famous DW-NOMINATE program -- one of the great achievements of modern political science. Poole and Rosenthal’s results regarding polarization have received a fair amount of media attention lately, popping up on “The Rachel Maddow Show,” NPR, and in the work of the American Prospect’s outstanding analyst Jamelle Bouie. The claim that Republicans are mostly to blame for the increase in polarization is usually accompanied by the claim -- also based upon DW-NOMINATE data -- that Republicans are now more conservative than they have been at any point since the 1880s.

This is a misapplication of DW-NOMINATE. As we’ll see below, DW-NOMINATE scores don’t really tell you how conservative or liberal a member of Congress is, at least not in the sense that most pundits use the term. It tells you how conservative or liberal a member of Congress is relative to other members of Congress. The latter is useful information, to be sure, but it really is an important distinction to be aware of.

Before explaining why I think this is a misapplication of DW-NOMINATE -- at least as pundits are using the data -- I should give a basic account of where I’m coming from in the big picture. On the one hand, I’m happy to see more rigorous analytical techniques make their way into political journalism. Having spent 18 months of my life applying a first cousin of DW-NOMINATE (known as OC, or Optimal Classification) to measure Supreme Court voting patterns for my master’s thesis (portions of which are paraphrased for this piece), my initial reaction was that the increased use of it was a good thing.

On the other hand, I’m concerned about relatively sophisticated statistical techniques like regression analysis and DW-NOMINATE being increasingly used by a journalist class that doesn’t really understand the assumptions or limitations behind them. With some of these techniques, journalists and their audience simply aren’t equipped to tell whether a technique is being misused or not. The result can be embarrassing misapplications, which the audience is then ill-equipped to analyze critically.

Remember, while there are a few political scientists who behave like partisans once they enter the public arena, the vast majority don’t. What occurs instead is what we’ve seen with regression modeling. These techniques all have certain assumptions and limitations built into them -- with which political scientists are familiar. When they speak to each other, there’s no need to list all of the caveats. The problem is that this manner of speaking is habit forming, and when addressing the general public, these assumptions and limitations really do need to be explained. All too often, however, they’re glossed over.

So it’s important to set forth at least a rudimentary explanation of what DW-NOMINATE is and how it works before analyzing this particular application. DW-NOMINATE grew out of a disenchantment with interest group ratings (think Americans for Democratic Action or American Conservative Union scores) as a measure of ideology. Interest groups typically cherry-pick 20 or so “key” roll call votes out of hundreds cast in Congress, and rate legislator ideology on this basis. Political scientists had noted that these scores often failed to grasp complexities in a legislator’s voting patterns, and could artificially place a lawmaker on the extremes of Congress.

DW-NOMINATE represents an attempt to rectify this by looking at every vote cast in a given Congress. Rather than starting with a pre-conceived notion of what a conservative vote would look like and picking and choosing votes to evaluate based upon that notion, DW-NOMINATE attempts, in essence, to reverse-engineer the “ideology” from the votes, without looking at the content of the votes themselves (stick with me here).

So how do we do this? DW-NOMINATE proceeds under the assumption that legislators who occupy similar positions on the ideological spectrum will vote together more frequently than they will with legislators who hold dissimilar positions on the ideological spectrum. In other words, because Bernie Sanders votes with Barbara Boxer frequently, and votes with Tom Coburn infrequently, we may conclude that Sanders and Boxer occupy similar ideological spaces, and that Sanders and Coburn occupy dissimilar ideological spaces.

By plugging in all votes from a given Congress, the DW-NOMINATE program in essence measures the similarities and dissimilarities in members’ voting patterns, and then assigns these legislators “scores” that measure where they would fall in a given Congress.

Rather than giving a primer in linear algebra, a simple example from my thesis may be more productive. The following is the raw data I used for the 1938 Supreme Court term. A vote with the majority is classified as a “1,” a vote against the majority is a “6,” and a failure to vote is classified as a “9.” By placing the votes in rows by the justice casting them, and in columns for each opinion (for Congress, the columns represent roll call votes), a matrix is created. Note that there are no unanimous votes -- DW-NOMINATE actually excludes unanimous votes, for reasons I won’t go into (a necessary shortcoming that can actually skew its perceptions of polarization).

This is quite literally all that DW-NOMINATE ever “sees.” Just to re-emphasize what these data are, however, here’s a more user-friendly depiction of the first 11 votes.

Just a simple glance at the “raw” matrix indicates that Justices Black and Douglas rarely voted with Justices McReynolds and Butler (who vote together in every instance except two). Indeed, McReynolds and Black never voted together on a split decision during the entire 1938 term. Thus, they represent the two poles. In the middle are Justices Roberts, Hughes, Reed, Frankfurter, and Stone, in descending order of likelihood of voting with McReynolds and Butler. Thus, the expected order would go something like McReynolds, Butler, Roberts, Hughes, Reed, Frankfurter, Stone, Douglas, and Black.

DW-NOMINATE-based programs perform the analysis much more precisely than our eyeballing of the data (although in the 1938 term, the result was exactly what our eyeballing suggested it should be), though with 435 legislators casting over 1,000 votes, the analysis is much more involved. Based upon these voting pairs -- how frequently members vote with each other -- DW-NOMINATE will also estimate “ideal points” for each legislator, which is essentially the “ideological score” that everyone is referring to when they discuss DW-NOMINATE. These typically fall between -1 and 1.

If you’ve been skimming this, now’s the time to stop doing so, as I’m about to stop speaking “geek.” There are three important limitations to the program. First, you are probably wondering: “Is -1 liberal or is -1 conservative?” The answer is “DW-NOMINATE doesn’t tell you.” It simply tells you how the legislators should be rank-ordered, relative to each other, and how far apart they are from each other in terms of ideology. We have to fill in the “conservative” and “liberal” labels based upon our real-world understandings of the legislators.

This isn’t a problem today, as we know that Bernie Sanders is a liberal and Tom Coburn is a conservative. Identifying the poles is therefore easy. But what about in the 1880s? What do “conservative” and “liberal” mean in that context? As discussed later, it is difficult to say.

Second, you are probably thinking: “But do we really need to look at all roll call votes? I don’t care how someone votes on the motion to proceed to the highway bill.” This is a more serious objection. NOMINATE absolutely weights votes on renaming post offices the same as the vote on the 2010 health care bill. You can see why this type of completeness might be important for political scientists hoping to divine granular distinctions between legislators, but might not be useful for purposes of general punditry.

In fact, this “completeness” might make things look more polarized than they really are. Ever since the mid-1970s, votes in the House for the “rule” and against motions to table or to recommit on bills have been more or less taken for granted as a matter of party loyalty if a member wants to progress in the House. It really is a big deal when a member casts a vote against the party on these. By including all of these votes, which are frequently party line votes even if the final vote is not, DW-NOMINATE can tend to make it look like liberals and moderates (or conservatives and moderates) within a party agree on more than they really do. This also goes to another problematic assumption behind DW-NOMINATE, which is that legislators cast “sincere” votes, e.g. they vote purely with regard to ideology and not with respect to other things, like logrolling, party loyalty, or cornhusker kickbacks.

The third, and most important, limitation is this: a score of -.5 means is that a member is quite liberal but only (a) relative to other people voting in that particular Congress and (b) relative to the issue agenda of that Congress. The scores have no absolute, fixed meaning.

This can have important implications. Consider the following example: Thirty-one Southern Democrats served in both the 71st Congress (1929-30) and the 76th (1939-40). Their median score in the 71st Congress was -.165. In the 76th the median was -.085. All of them except three saw their scores migrate “rightward” during this time period. One legislator, Fritz Lanham of Texas, who represented Fort Worth and some surrounding counties from 1919 to 1947, swung from negative .139 to positive .114. In other words he went from voting mostly with Democrats to voting mostly with Republicans (at least on economic issues; for reasons beyond the scope of this article, DW-NOMINATE effectively filters out racial issues in this time period)!

The explanation most would immediately reach for here is that the Southern Democrats became more conservative in response to Franklin Roosevelt. This is possible. But it is also possible -- and consistent with the historical record -- that they didn’t shift at all. Instead, we should consider that the House was populated with many more relatively liberal members from 1929 to 1939, and that the issue agenda changed from a conservative Republican one, which the Southern Democrats opposed, to a progressive Democratic one, which the Southern Democrats also opposed. Going from mostly opposing conservative Republican legislation to mostly opposing the legislation put forward by their progressive brethren would create an illusion of movement on the part of the Southern Democrats in terms of their absolute conservatism, when in reality they stayed the same. DW-NOMINATE simply gives us no particularly useful tools for discriminating between these competing explanations.

This brings us Mann and Ornstein’s claims, which I think we can now dispense with rather quickly given this understanding of what DW-NOMINATE actually does. One simply cannot say that Republicans are the most conservative they’ve been since the 1880s, because the DW-NOMINATE scores are created without any fixed meaning. A legislator with a score of .2 in 1880 could actually be quite a bit more conservative, based upon our current understanding of “conservative,” than one who received a score of .7 in 2010. Put differently, a legislator with a score of .2 in 1880 would almost certainly never have voted for Medicaid; a similarly situated legislator today would almost certainly vote to renew it.

In fact, the historical record suggests that while there are some similarities between the basic Republican platform in 1880 and today, on other issues such as free trade, the differences are quite radical (and the Republican position on free trade today would have been considered “liberal” in the 1880s). In fact, it’s debatable whether Republicans should even be assigned the conservative pole in the 1880s, or that the Democrats should be assigned the liberal pole (which would also make the current Democratic caucus the most liberal one in history as well), given the strength of laissez-faire “Bourbon Democrats” in the Democratic Party during this time and the Republican Party’s position on internal improvements and spending on things like education.

For the same reason, while the DW-NOMINATE data can show that the current Congress is the most polarized in history (since that is a relative measurement), they cannot really show us who is to blame for that polarization. Without any lodestone to hold the “center” steady, we can’t tell who has moved the farthest away from that “center.” And because we have a pretty good sense that the center shifts from Congress to Congress, especially when control of that body changes hands, it becomes almost impossible to make meaningful comparisons between even two Congresses, let alone over the course of multiple decades.

Again, to illustrate this point, DW-NOMINATE suggests that Congress became much less polarized in the 1920s and 30s, which is probably true.  But it suggests that both parties moved toward the center (and that both Northern and Southern Democrats, on average moved toward the center).  There simply isn't any support for the idea that the Democrats in the 1936 Congress were, on average, more conservative than Democrats in the 1926 Congress, at least in the sense that contemporary pundits use the term.  But as the agenda moved leftward, the rise of the Conservative Coalition has the effect of pulling both parties toward the center, even though, overall, both were probably becoming more liberal.

DW-NOMINATE remains a powerful tool, especially if you keep its limitations in mind and are looking at discrete Congresses (or really are interested purely in polarization). It even sheds interesting light on realignment theory. But it really doesn’t do any of the things the popular press is claiming it does, at least not particularly well. 

‹‹Previous Page |1 | 2 | 3 |

Sean Trende is senior elections analyst for RealClearPolitics. He is a co-author of the 2014 Almanac of American Politics and author of The Lost Majority. He can be reached at Follow him on Twitter @SeanTrende.

A President Who Is Hearing Things
Richard Benedetto · November 12, 2014
The GOP Will Set the Agenda
Jack Kelly · November 9, 2014

Latest On Twitter