Cluster Analysis of Congressional Votes With the Bc Try System

• DOI: 10.1177/106591296601900402
• Published date: 01 December 1966
• Author: Duncan Macrae
• Date: 01 December 1966
• Subject Matter: Articles

CLUSTER ANALYSIS OF CONGRESSIONAL VOTES
WITH THE BC TRY SYSTEM
DUNCAN MACRAE, JR.
University of Chicago
OLL-CALL
VOTES have attracted increasing interest as sources of infor-
mation about issues in the legislative process.~ The availability of computer
-t-
programs for data processing has made it possible to search for salient issues
by examining all the roll-call votes in a given Congress, rather than limiting analysis
to pre-selected subsets of votes reflecting the judgment of the investigator. One par-
ticularly promising set of computer programs for this purpose is the BC TRY sys-
tem, designed by R. C. Tryon.2 The purpose of the present paper is to illustrate the
applicability of this system to roll-call analysis.
Various procedures of multivariate analysis are applicable to the identification
of legislative issues from roll calls. These include Guttman scaling, factor analysis,
and cluster analysis. Computer programs are available for these procedures and sev-
eral versions of them can be performed by particular programs in the BC TRY sys-
tem.3 We
select for presentation the procedure of non-communality cluster analysis,
which is not widely available outside the BC TRY system, and which also has the
advantage of making fewer assumptions about communalities than other methods.
In addition we present the spherical analysis (SPAN) diagram, which can be com-
puted for any three-dimensional factor space identified by the system. To illustrate
these features, we shall use the votes of the Republican members of the U. S. House
of Representatives in the 84th Congress ( 1955-56) .
CLUSTER ANALYSIS
Cluster analysis, like factor analysis, is a method of analyzing the associations
between pairs of items in order to summarize them more simply. In the search for
legislative issues, these items are roll calls; in the related search for legislative blocs,
NOTE: This research was aided by a grant to the author from the Social Science Research
Council. Assistance from R. C. Tryon and from the Computer Center, University of
California, Berkeley, is gratefully acknowledged.
1
Two recent examples are L. N. Rieselbach, "The Demography of the Congressional Vote on
Foreign Aid, 1939-1958," APSR, 58 (September 1964), 577-88; and H. R. Alker, Jr.,
"Dimensions of Conflict in the General Assembly," ibid., 642-57.
2
The operation of the component programs of the system, as well as their underlying theory,
will be described in R. C. Tryon and D. E. Bailey, Cluster and Factor Analysis (in
preparation). Draft versions of sections of the book are available from Professor Robert
C. Tryon, Department of Psychology, University of California, Berkeley. See also
Tryon and Bailey, "The BC TRY Computer System of Cluster and Factor Analysis,"
Multivariate Behavioral Research, 1 ( January 1966 ) , 95-111.
3
A program for Guttman scaling, which performs functions similar to those of cluster analysis,
is described in J. C. Lingoes, "Information Processing in Psychological Research,"
Behavioral Science, 7 ( July 1962 ) , 412-17. Functions available in the BC TRY system
include the computation of correlations, estimation of communalities in several ways,
principal-components and centroid factor analyses, cluster analysis using communalities,
orthogonal rotation of axes, computation of factor scores, and comparison of factor
structures. For a classification of procedures in this general area, see Tryon, "Domain
Sampling Formulation of Cluster and Factor Analysis," Psychometrika, 24 (June 1959),
113-35
631

---

632
they are legis!ators.’ A matrix of correlations (or other coefficients of association)
can be analyzed by cluster-analysis procedures, to allocate individual roll calls to
mutually exclusive clusters, and these clusters can then be examined to reveal cor-
responding issues or other bases of division.
Cluster analysis differs from factor analysis in stressing groupings of the items
themselves, rather than presumed underlying variables that may be related to the
items in varying degrees.5 Cluster analysis typically begins by searching for sets of
roll calls that resemble one another within sets but are distinct between sets; if co-
ordinate axes are defined, they are passed through such sets.6 It thus proceeds im-
mediately to what is in general an oblique coordinate system, rather than reaching
it indirectly through oblique rotation of initially orthogonal axes. In the search for
legislative issues, this possibility of oblique structures is of great importance, and clus-
ter analysis provides a relatively easy and direct means of revealing it.
The possibility of oblique axes to represent legislative issues is important because
one cannot expect issues, as legislators see them, to be completely independent of one
another. Legislators distinguish among various general issues that confront them,
recognizing the presence or absence of these issues in individual roll calls.’ They,
and other interested parties, place themselves and their colleagues relative to one
another on these issues. While one issue is distinguishable from another, the posi-
tions of legislators on two issues (e.g., foreign aid and reciprocal trade) are normally
related or associated. The relation between issues is actually an important feature
distinguishing one period of time from another, or one party from another.
Any method that finds only independent dimensions...