The first test I explored, Keith Poole and Howard Rosenthal's NOMINATE, limited itself to an observation of US Congressmen, employed an “action-based” observational method, and sought to describe individual Congressmen within a dynamic number of dimensions. Curiously, most roll-call votes measured using NOMINATE revealed either a single relevant dimension or, at most, two relevant dimensions, leading Poole and Rosenthal to conclude that American politics was itself defined along two axes at most. While on the surface this method appears sound, difficulties do arise. First of all, the model can only measure the actions of members of Congress, and does not readily port to non-legislative actors. Second, the data used are the records of roll-call votes within Congress, and because of such Congressional practices as logrolling, as well as the limited response problem (one votes either in favor or against, or else he says 'present') the data captured may not necessarily constitute a reliable indicator of ideological orientation, but possibly a tendency towards partisanship within two predominant parties in Congress. Third, the ideological spectrum originally used by Poole and Rosenthal to identify Congressmen presupposes that “voting in Congress is entirely driven by one basic dimension--liberalism and conservatism” (Poole 3), and therefore either obscures or ignores other ideological positions such as libertarianism, communitarianism, or “centrism”. Nevertheless, NOMINATE may constitute a reliable indicator of the degree to which the imposed linear ideological spectrum is evident in Congressional activity. Fourth, NOMINATE appears to rely for its error correction upon the self-identification of the Congressmen under scrutiny, which may be adversely distorted by a halo effect, or else by a limited set of responses (in this case, liberal/conservative).
The
second test I explored, Benjamin Bishin’s FILTER, sought to address
some of the difficulties encountered with NOMINATE. Instead of
relying upon action-based data such as roll-call votes, FILTER
included an individual demographic survey of the members of Congress
to offer a predictive model for ideological self-identification in
Congress. It includes such variables as geography, race, education,
and farming background as a part of its data set, quantifying each
term by assigning both positive and negative numerical values to each
item in the data set, and through a sophisticated statistical formula
arrives at a numerical value to describe how liberal or conservative
the test subject may be. This index was then compared with the
self-reported ideological stance of the test subjects. In many
respects the FILTER method produced reliable predictions of
self-perceived ideological dispositions, but several outliers did of
course appear, whereby test subjects’ responses did not closely
match the predicted ideological position. This may be because the
FILTER method, while relying on historical and demographic
information to produce an ideology index, was nevertheless limited to
the same basic one-dimensional ideological measure. Furthermore,
this model continued to rely on direct self-reporting on the part of
test subjects to offer a check against the formula developed by the
authors.
The limitations inherent in the one-dimensional political/ideological spectrum notwithstanding, it appears that a critical need exists for a measure that neither inherently limits responses nor relies on self-reporting for error correction. For the purposes of measuring ideological preference, both of these errors invalidate any measure that uses them.
Comments