Originally posted in March, 2011. We had a disk crash last winter and we are attempting to reconstruct lost posts from 2011.
We have been asked over the years whether or not DW-NOMINATE is affected by the composition of the congressional agenda; that is, the mix of issue areas that are being voted on over time. In particular, how does the agenda affect polarization (the policy distance between the two major parties)? Could it be that party leaders in Congress are exerting more influence over procedural votes, which is artificially increasing the frequency of party-line, polarized votes?
In Congress: A Political-Economic History of Roll Call Voting (1997), in Ideology and Congress (2007), and in Spatial Models of Parliamentary Voting (2005) we show some experiments with agendas and find that the pattern of polarization is unchanged if the Southern States are removed and if the distribution of the roll call margins is held fixed across all margins.
Below we show another experiment where we exclude almost all close votes from the analysis so that 70 percent of the roll calls have a majority margin of two-thirds or higher and 94 percent of the roll calls have margins of 61 percent or higher. Specifically, we use the following distribution of margins:
From each House, we sample roll calls with replacement until we have 400 roll calls with the distribution of margins shown above. We then analyzed the first 111 Houses with the standard DW-NOMINATE two-dimensional linear model. Below we show the party polarization measure for the Democrat-Republican Party system for the 46th to the 111th Houses (1879-2010).
As can be seen from the figure, the pattern of polarization is reproduced almost exactly. Removing the close votes (which almost certainly removes all the party unity votes) has the effect of dampening the polarization, but the pattern remains the same.
Below, we also plot the means of the two political parties (the distance between the respective means is the polarization measure shown above). Removing the party unity votes lessens the absolute level of polarization, but for both parties the patterns are the same.