Thursday, October 22, 2009

Polarization among state politicians

In a post about NY congressional candidate Dede Scozzafava, political scientist Boris Shor (who generally does what I do, only better and faster, which is a problem), produces the graph below. Just to explain what this is, Shor and Nolan McCarty have collected the NPAT scores for state legislators in all fifty states. The NPAT (National Political Aptitude Test) is administered by Project Vote Smart. He managed to produce a common scale across all the legislatures. The graph shows the ranges of issue stances within the legislatures.
This graph is very cool for me, in particular, because, well, look how darned polarized California looks! It's got the most divergent parties in the nation. This interests me because some roll call studies (particularly Jerry Wright's data) suggest that California has one of the most polarized legislatures, but not the most. It's occassionally surpassed by Wisconsin, Iowa, and a few others.

What Shor's data show, if I understand them correctly**, is that the ideologies of California's elected officials are the most polarized -- not as manifested in roll call votes, but as simple expressions of issues stances. This gets around the issue of agenda control, by which the majority party can pick votes that make the parties look more divergent than they really are. California's parties have been the most effective in the nation at screening out the moderates. There really is No Middle Ground. (Sorry for the shameless plug.)

*NPATs have a fairly high non-response rate, as I understand it. One presumes, however, that the non-response rate is uncorrelated with ideology or region.

**Late correction: I apparently did not understand them correctly. As Boris explains in comments, the ideal points are, in fact, derived from roll call data. Apparently 15 years of roll call votes from each state legislature. (I really need to know how they pulled that off.) They used the NPAT scores to create a common space across legislatures. So agenda control is still potentially an issue, but probably not a huge one.


bshor said...

Seth, thanks for the kind words. So actually the nonresponse rate isn't really a problem, because I use both roll call data and Votesmart survey data. The former is to get 100% coverage of all state legislators for the past decade and a half or so, the latter is to put every state into common space with each other and Congress. As long as respondents are consistent in their issue preferences between roll call voting and survey response, I'm ok...

About California -- yeah, that pretty much jumps out at you, doesn't it?

Anonymous said...

Apparently 15 years of roll call votes from each state legislature. (I really need to know how they pulled that off.)

They've set up tools so that as soon as they get an electronic copy of the journal, they can have a student input some appropriate parameters into a perl or python program that, once set, scans through the entire set of journals for roll-calls. It helps that the way journals display roll-call votes fall into a few relatively easily coded categories.

Then they combine that with OCR to turn paper journals they've ILL'ed into electronic journals.

Even apart from having the roll-call data too, nonresponse bias doesn't seem to be much of a problem. If you predict NPAT participation with vote-based nominate scores by party in Wright's data, they're only significant for a few chambers, so NPAT participants vote pretty much like their nonparticipant copartisans. Now that I think about it, they may even be jointly insignificant across chambers.

(Jim Battista, who's far too lazy to sign up for a proper userid)