I chose roll calls used to create scores on the various issues for members of Congress based on their policy content and validated by use of principal components factor analysis to insure that scales tapped into one dimension. I took votes from ICPSR Study no. 4, which includes all recorded votes in the House and Senate through the 104th Congress. I took more recent votes from the web sites www.senate.gov and www.house.gov.
I initially selected roll calls in the same substantive policy area based on ICPSR descriptions as well as the summaries in the Congressional Quarterly Almanacs and – in ambiguous cases – by consulting the Congressional Record itself. After this initial substantive selection I retained only those votes that remained in a factor analysis in which only one component had an eigenvalue with a value of one or greater, and in which each roll call loaded at 0.6 or more on the dominant factor.
This is a fairly stringent standard that resulted in discarding a fair number of votes, but I found that more liberal criteria for determining which roll calls to keep, such as using all those that loaded primarily on the factor that explained the greatest share of the variance, did not meaningfully alter results.