Hierarchical Ideal Point Models using Diffusion Trees

• In Event: Scaling and Ideal Point Estimation

Abstract

Understanding hierarchies that exist within political institutions allows for more in-depth analysis of the processes which lead to policy decision making. Hierarchies within voting bodies can exist in multiple ways - organization of voters into voting coalitions, salient policy areas that are nested into multiple categories, the choice to vote on certain topics as a function of their sources, etc. These hierarchies are key to many of the political processes that are central to the study of U.S. political institutions, but are rarely considered in quantitative models that attempt to model these phenomena.

Ideal point models are frequently used to understand the voting behavior of various voting bodies in the U.S. These models rely on strong assumptions of independence for both estimation purposes and identifiability. More recent approaches, however, have begun to relax these assumptions and use the ideal point model to perform rich inferences about the complexities of voting in political institutions. These methods create flexible frameworks for estimating ideal points and are ripe for extension.

In this paper, I extend the ideal point model to include notions of hierarchical decision making. I utilize Dirichlet diffusion tree priors (DDT) to model hierarchies in both the topic space and ideal point space. This specification allows for a version of hierarchical clustering among the ideal points to uncover a rich representation of voting within an institution. Unlike previous models which disregard hierarchical organization within voting coalitions (or completely ignore coalitions all together), DDT allows for a richer understanding of how voting members organize when making decisions. Along these lines, DDT is also used to allow for a hierarchical specification of the topic space in which the ideal points are represented. This approach uncovers a hierarchical representation of this topic space, relating the various issue dimensions from ideal point analysis to one another in a tree-based fashion. Estimation of this new structure promises to provide a thorough analysis of the relationships between important issue areas and coalitions in ideal point models.

To demonstrate the power of this model, I examine the entirety of the U.S. Congress to understand the evolution of coalitions over time. Along with uncovering membership in coalitions, the set of issues which create divisions within and across coalitions are uncovered. This analysis is used to create a more fine-grained measure of polarization within the U.S. Congress and shows that various groups have played roles in the perceived differences between parties over time.

A particular case of interest is the Southern Democratic coalition - a group that formed during the Civil War and proliferated until the 1990s. Previous research suggests that this group differed from Democrats and Republicans on key sets of issues until most Southern Democrats and their respective districts switched support to the Republican party in the late 1900s. Similarly, scholars attribute much of the increase in polarization within the U.S. Congress to this group. While attempts have been made to examine these notions quantitatively, the majority of analyses have relied heavily on qualitative approaches. Using the DDT ideal point model, I provide a quantitative analysis using only roll call votes and show that the voting behavior of Southern Democrats coalesced with the Republican party on many key issue areas long before the 1980s. However, there was still a strong allegiance to the Democratic party on procedural issues until the switch in the late 1980s. This finding corroborates previous results which suggest that Southern Democrats were Democrats in name only, though it differs by showing that Southern Democrats still voted with Democrats on matters related to how the legislative body made rules and distributed funds.

Author

• Kevin McAlister, University of Michigan