Legislator Ideology Scores for Divided Government Project
Abstract
This project develops ideal point estimations of all members of the U.S. Congress between 1947 and 2010. It uses Bayesian item response theory in the R statistical program using the pscl package developed by Clinton and Jackman. These estimates will provide important leverage into legislative voting behavior since the votes used in this study are separated by specific type, policy area, and time series than is commonly used in most legislative scholarly work. Access to FutureGrid will provide me with the processing power that exceeds a standard laptop. Moreover, because the matricies used in this estimation process are quite large, and because the pscl package is serial and not parallel, the Stampede system is unable to process most of my requests as the I tend to exceed the 24 hour wall clock limit. FutureGrid is my last option.
Intellectual Merit
The intellectual merit of this project is that it is a new way of creating legislator ideal points across large sections of time. Where most legislative scholars use Poole and Rosenthal's DW Nominate scores, these estimates are limited since they use all votes across each two-year congressional limit. While comparable across time, the theoretical limits of these scores are apparant if one hope to look a specific votes and time periods. In other words, these popular scores are estimated from all votes, whether they are final passage, procedural, or commerative, and comprise two-year intervals. In this project, I analyze final passage votes only across specific policy areas as well as across each Congress, year, and quarterly time series. The estimates from DW Nominate can lead to biased estimates in my statsitical models if I am looking at one policy area while usiing estimates that come from all votes.
Broader Impact
The broader impact of this study is that it develops a way of estimating specific votes across large and unique sections of time. Currently, if scholars want to use the Bayesian techniques of Clinton and Jackman limited processing strength requires that they create estimates that span small time periods which cannot be compared to votes at other times due to the differences in the policy space as well as the potential of different members of Congress voting at one period and not others. While I did not develop the item respone program or the format for creating the matricies that build the bridge votes and members to compare across time, this study is the first that uses these techniques on such a large time series. The FutureGrid system will make these estimates a model for future political scientist to move away from DW Nominate into a world where HPC can create any number of possible estimates. This will allow political scientists to gain better leverage on congressional voting behavior.
Use of FutureGrid
I simply need FutureGrid to provide the processing strength and length of time to show my profession an alternative technique for processing unique legisaltor ideal points across large sections of time.
Scale Of Use
I need to run a number of matricies. I plan to run four different estimates based on foreign policy votes, and another four based on domestic votes. These will be separated by congress and year. I wll also have a massive quarterly times series matrix that I will run. So far, I estimate that the longest dataset will take eight days based on one node speeds in Stampede.
Futuregrid is a resource provider for