A Dynamic Dirichlet Process Mixture Model for the Partisan Realignment of Civil Rights Issues in the U.S. House of Representatives

TL;DR

This paper introduces a nonparametric Bayesian model combining hidden Markov and Dirichlet process mixture models to analyze the gradual and sudden shifts in U.S. political parties' civil rights positions over time.

Contribution

It develops a novel dynamic Bayesian model that captures multiple latent clusters and shifts in political party positions using longitudinal legislative data.

Findings

Evidence of gradual racial realignment in the 20th century

Identification of two periods of rapid change: 1948 election and Civil Rights Movement

Model uncovers coalitions and shifts in party positions over time

Abstract

Evolutionary societal changes often prompt a debate. The positions of the two major political parties in the United States on civil rights issues underwent a reversal in the 20th century. The conventional view holds that this shift was a structural break in the 1960s, driven by party elites, while recent studies argue that the change was a more gradual process that began as early as the 1930s, driven by local rank-and-file party members. Motivated by this controversy, this paper develops a nonparametric Bayesian model that incorporates a hidden Markov model into the Dirichlet process mixture model. A distinctive feature of the proposed approach is that it models a process in which multiple latent clusters emerge and diminish as a continuing process so that it uncovers any of steady, sudden, and repeated shifts in analysing longitudinal data. Our model estimates each party's positions on civil rights in each state based on the legislative activities of their Congressional members, identifying cross- and within-party coalitions over time. We find evidence of gradual racial realignment in the 20th century, with two periods of fast changes during the 1948 election and the Civil Rights Movement.