Liberal Senators

It’s pretty impressive that the algorithm achieves perfect separation between Reps and Dems. As the authors state, this is the most polarized Senate since Reconstruction. No doubt bloggers are to blame.

It’s interesting to note how close to the bottom McCain shows up. I can’t say I’m surprised, but it’s interesting given how many people seem to think of him as a moderate.

Ah, simultaneity.

I was struck by how conservative Ron Paul actually is. The metric used fails the more for Paul then anyone else, and he was the second worst classified House member in the 109th. I am guessing that this is his opposition to the war.

Paul is not a libertarian. The man is isolationist Republican. I now know who my older relatives voted for.

“It’s pretty impressive that the algorithm achieves perfect separation between Reps and Dems. As the authors state, this is the most polarized Senate since Reconstruction. No doubt bloggers are to blame.”

Posted by John Quiggin

And this doesn’t help the libertarian ‘two orthogonal axes of politics’ theory.

When Zell Miller was still there (a few congresses ago- I can’t remember which it was and am too lazy to re-click this morning) he was placed well into the Republican pack. No wonder he was invited to give a speech at the Republican National Convention in ’04.

And this doesn’t help the libertarian ‘two orthogonal axes of politics’ theory.

So a method that shows up all the senators against a single axes doesn’t show anyone appearing on a second axes. Astonishing.

And this doesn’t help the libertarian ‘two orthogonal axes of politics’ theory.

So a method that shows up all the senators against a single axes doesn’t show anyone appearing on a second axes. Astonishing.

Poole and Rosenthal’s research really does tell against the libertarian two-axis theory. (I did some coding for them around 1990, helping write some software for displaying their results, so this is a non-professional’s take on something I read a while ago — but here goes. WARNING: Long comment follows.)

Basically, they begin from the observation that, even though there’s no obvious political theory connecting gun control and abortion and defense spending (for instance), knowing someone’s vote on one does help you predict their vote on another. So they ask: can we rank Members of Congress in a way that helps us predict their votes? And how many dimensions do we need to rank them in in order to predict their votes?

That is: For a given congress, give every member a position along one or two or more dimensions. Then, for each roll-call vote,* draw a line on your plot, trying to put all the no votes on one side of the line and all the yes votes on the other. Can you make a map of the members that minimizes the number of times that you’re forced to put someone on the wrong side of the line? (That is, they voted yes, but they’re on the same side as everyone who voted no.)**

The thing about this is: The more dimensions you use in your plot, the easier it should be to get the votes right, because you have more flexibility in how you draw the line. So it’s significant if moving from two dimensions to three dimensions doesn’t give you any appreciable increase in accuracy; it shows that there are basically two factors that determine how someone’s voting.

And in fact, for most of the congresses they studied, they only needed two dimensions. (Fortunately for me, since I was working on the display software.) Except for one congress around the breakup of the Whig party (IIRC), plotting people on two dimensions let you classify their votes about as accurately as going to more dimensions. So there was grounds for subscribing to a two-axis theory, but it wasn’t the libertarian theory; one axis was more or less a left-right axis, one was a regional axis corresponding roughly to views on race.

But, as I understand, since about the mid-80s the two axes have collapsed into one. The decline of the Dixiecrat party has meant that the second dimension doesn’t yield much more predictive power anymore. In the data I was working with, you could really see this: in most of the recent congresses, the parties were separated along the left-right axis, and then spread out along the vertical axis, with the southern Democrats apart from the others. But in the very last congress the southern Democrats started to mush in with the rest of the party. This accords with what the methodology link says about “the 3-Party period from the late 1930s to the mid-1980s when the Democratic Party was split into Northern and Southern blocs.”

And the great thing about this method is that it doesn’t rely on any preconception of what’s liberal or conservative or what. It only looks at the correlation between votes on different issues. In this respect it is very unlike the libertarian two-axes analysis.

*I think they throw out votes that pass with only a few no votes.

**This is what is meant when they say (in the methodology link Kieran provided) “Given a matrix of binary choices by individuals (for example, Yes or No) over a series of Parliamentary votes, OC produces a configuration of legislators and cutting lines/planes that maximize the correct classification of the choices.” That was about the last thing I understood in that link.

matt, isn’t this just restating the obvious political reality – a party in congress (or any similarly set up political insitution) can’t be split over too many issues to the extent that members vote agaisnt the party majority view regularly, or that party will become completely ineffective and the opposing faction will get everything their own way.

Stuart, I don’t know — I’m not a political scientist, so I’m not the best interpreter of these results. Maybe one of the social scientists on the board will chime in.

It is my understanding that the analysis does show a big change: in the three-party era from the 30s to the 80s you needed to rank people on two dimensions, but with the breakup of the Dixiecrats one dimension became adequate. Which also reinforces the idea that increased party polarization is due partly to the breakup of the Dixiecrats (when they migrated to the GOP the Democratic party stopped being polarized on race). But my understanding here is pretty sketchy — and I only worked on this for one summer, before the post-Dixiecrat era was really underway, so I may completely have the wrong end of the stick.

A party in congress (or any similarly set up political institution) can’t be split over too many issues to the extent that members vote against the party majority view regularly

Given that we had such a situation in the past, (minimally from 1948 to about 1968, but maybe for ten or fifteen years before that), the most we can say is that such arrangements are unstable — and maybe not even that. (Matt’s “in the three-party era from the 30s to the 80s” is quite misleading; except in 1948 and 1968, the Dixiecrats were always Democrats).

I’d guess that the present Republican party split, if there is one, could be graphed as big-government vs. little-government conservatives; the actual libertarians don’t seem to be a factor any more.

My guess is that big government conservatives are doubtful about free trade and enthusiastic about pork, whereas little-government conservatives are the opposite. Beliefs about war and social issues would be scattered — Ron Paul is anti abortion and anti war, Giuliani and Romney are pro-war and historically pro-choice, but most Republicans of any description (like Huckabee and McCain) are pro-war and anti-abortion.

Bush’s deferred tax cuts and reliance on pork to enforce unity on Congressional Republicans allowed the Republicans to punt the big/little split into the future, but the future might be now.

Or not. I just whipped that one up right in front of your eyes.

I think that Barry is right in saying that the fact that a one-dimensional ordination like this one explains a lot of what’s going on does hint pretty strongly that a second dimension wouldn’t be that important.

And Matt’s information that in some periods, the second dimension was needed is pretty telling too, especially it suggests that the second dimension becomes important in times when the system of political alliances gets more complicated, rather than at times when an additional aspect of political philosophy comes into focus.

I’m a political practitioner myself, and I’m bringing my profession’s stong bias toward the theory that alliances are more important than political philosophy.

In fact, if I could add a second dimension, I’d want it to cover “Senator Pothole” type instate problem solving and service. That would do a lot to separate the two New York Senators, who show up as tied in this one.

The two-party system in all its forms.

Howevermuch fault such a system may have in reality, I wasn’t referring to that but too a technical point in Matt Weiner’s explanation. If there is a system that forces politcal choice to one axis locally, and your data is about the outcome of that proces, namely Congress, then your calculations will tend to ignore secondary axes, even if they are important to voters’ decisions.

In fact, the only secondary axes such an analysis is going to show is when the single political axis in some districts is different from that in other districts.

26: It seems to me that the two party hierachies in Congress and the influence of the national party on local races are more important factors, though I suppose the winner-take-all might contribute.

Local winner-take-all races seem to me to be more susceptible to the patching together of weird local coalitions, for example a mix or ethnic, pork-barrel, and ideological appeals with personal charm and personal connections.

Yes, I mentioned that in my first post, but in a bit obscure language ( I said local politics would map on the main political axis of national politics, instead of its own first principal component). The trouble is that I used very similar math as behind these kinds of analyses, but in a different context and with different terminology.

Megan McArdle wants to know something

And God forbid she should do some actual work to find out.

Just what our discourse was crying out for: a libertarian Jonah Goldberg.

Matt, thanks for the explanation.

Suart: “So a method that shows up all the senators against a single axes doesn’t show anyone appearing on a second axes. Astonishing.”

I sort of figured that people had heard of this; I was obviously incorrect. The general statistical principle is that things like this can be perfectly explained by a sufficiently complex model (which may have poor prediction ability, please note). The whole point of their analyses, from my casual readings was just how few axes were needed, when the second or third was significant, and what the axes seemed to represent.

This is all exceedingly interesting (especailly Matt W.’s 15.)

But I think one problem is that all votes are not created equal. To get a mapping that corresponds to our intuition, you’d have to give greater weight to certain key votes, and there doesn’t seem to be any obvious “automatic” way of doing that.

In this case, most of the anomalies (Lieberman, Paul, etc.) come from positions on the war. Foreign policy by its nature sees fewer roll-call votes, so it’s going to get underweighted in this analysis. If you could somehow correct for that, you might see two axes again. Or not — Lieberman and Paul may be sui generis.

More generally, given the winner-take-all system, a lot of what you’re going to see is regional patterns. Whcih, as has been noted, are a lot less strong now that the South has been somewhat civilized.

The mind reels:

http://blogs.wsj.com/economics/2008/01/31/volcker-i-endorse-obama/

http://www.vnews.com/02012008/4601389.htm

The Performance-artist community splits:

http://www.youtube.com/watch?v=I2JoSo17Azk&e
http://www.youtube.com/watch?v=BEmiavw_-sE

Forget axes, we have a crisis of representation.

“More generally, given the winner-take-all system, a lot of what you’re going to see is regional patterns. Whcih, as has been noted, are a lot less strong now that the South has been somewhat civilized.”
Posted by lemuel pitkin ·

From memory, what this sort of analysis showed was that there *had been* a two-axis system, which was chiefly due to the Dixiecrats (i.e., a hard-right regional faction allied with a generally more liberal party). Once the Dixiecrats died/retired/converted to the Republican party, the two axes collapsed to a single, left-right axis.

In addition, to answer your question about technique, IIRC they only counted votes with some disagreement; e.g., a vote for ‘National Flowers’ week, passing with close to 100%, would not have been counted.

Thanks to everyone for actually reading that comment! Let me say again that I’m not an expert.

greatzamfir: isn’t there a problem in this analysis as result of the winner-takes-all system in elections?

Well, the analysis is just meant to capture congressional voting records, not the records of individual members. But you may be right that, even if individual voters are spread along two dimensions, the winner-take-all system would compress their representatives onto one dimension. I’d wonder why it compresses one dimension but not both, OTOH. Is this the point of talking about the party system — that it forces local disagreement onto one axis?

John Emerson: Matt’s “in the three-party era from the 30s to the 80s” is quite misleading; except in 1948 and 1968, the Dixiecrats were always Democrats

It’s Poole’s phrase, more or less, but it’s not meant to suggest that there were three de jure parties; rather than there were three de facto parties. Check out the picture of the 95th House on their site; the southern and non-southern Democrats are pretty well separated, although there was some overlap. (They have the program display any Democrat from the Confederacy +Kentucky/Oklahoma as an S rather than a D.)

Barry, what you describe is exactly how I remember it too. But I think Lemuel is right that a real liberal-conservative ranking (which this isn’t exactly supposed to be) might weight some votes more than others. Then you’re into subjective territory, though.

Others have also pointed out that there’s a lot more than the final vote that goes into liberalism and conservatism. There’s amendments and negotiations in Congress, and there’s the rhetorical status you take outside of Congress. Feinstein often seems to be illiberal on the procedural stuff, and Lieberman loves to undercut the liberal position on talk shows. Still, this is interesting at measuring what it does measure.

“Megan McArdle wants to know something

And God forbid she should do some actual work to find out.

Just what our discourse was crying out for: a libertarian Jonah Goldberg.”

Don’t be a putz. All she did was ask a simple question on her blog, something that folks around here do all the time. It’s not like she asked anyone to do her research for her. Don’t be a putz.

To give you a hint, Megan McArdle’s reaction to the Lancet II article estimating deaths in Iraq was that her ‘gut’ told her it was wrong. This would be the trained gut of a Chicago MBA, please not, not the trained gut of somebody with knowledge of war zones.

One of these days I’m going to finish my four-axis political map. This will allow me to show Libertarians exactly where they lie within the interior of a tesseract.

Then, all I have to do is fold up the tesseract, place it in my pocket, and the Libertarian will vanish into the aether. Problem solved.