Factor analysis versus multi-dimensional scaling: binary choice roll-call voting and the US Supreme Court

Abstract

We evaluate factor analysis and multi-dimensional scaling (MDS) as tools for the analysis of voter decisions over a series of dichotomous choices. We simulate binary voting data with a known form and illustrate that standard factor analyses of these types of data yield additional artifactual dimensions. This effect may be exacerbated by the choice of inter-voter measures of similarity used as input. These results call in question the conclusions of others based on standard factor analyses of empirical voting data from the US Supreme Court. We demonstrate that MDS methods produce more parsimonious descriptions of these data.

Environmental changes in Lebanon during the Holocene: Man vs. climate impacts

“…6b shows that the variation of the scores of the second axis is dependent on the variation of the scores of the first one (Guttman/ arch effect, see for example: Kruskal and Wish, 1991;Brazill and Grofman, 2002). Consequently, the Aammiq pollen record seems importantly related to the variation of Cichorioideae percentages and consequently to sediment perturbations since c. 10 ky cal.…”

Civil Society and Cyber Society: The Role of the Internet in Community Associations and Democratic Politics

“…6. Additionally, since the data on offline associations are binary (yes/no), we prefer MDS to a factor analysis because factor analyses have been shown to produce false factors, particularly when working with binary data (Brazill & Grofman, 2002). 7.…”

Analyzing Voting Behavior in Italian Parliament: Group Cohesion and Evolution

“…We now analyze the voting behavior of Italian Parliament by applying the well known multidimensional scaling technique known as Singular Value Decomposition (SV D) [8], whose advantages with respect to other techniques have been discussed in [9]. Let A be an n × m voting matrix where rows correspond to Representatives and columns to the votes cast to approve a law.…”