A finite mixture mixed proportion regression model for classification problems in longitudinal voting data

Area of Science:

• Social Sciences
• Political Science
• Statistics

Background:

• Continuous clustered proportion data are common in social and political sciences.
• Analyzing voter proportions over time across different regions presents challenges due to regional variations.
• Identifying groups with similar electoral behavior profiles is crucial for understanding political dynamics.

Purpose of the Study:

• To propose a novel statistical model for analyzing continuous clustered proportion data.
• To develop a finite mixture of a random effects regression model using the L-Logistic distribution.
• To effectively identify and analyze clusters of similar electoral behavior over time.

Main Methods:

• A finite mixture of a random effects regression model based on the L-Logistic distribution is proposed.
• A Markov chain Monte Carlo (MCMC) algorithm is employed for Bayesian inference.
• The method is applied to analyze presidential election voting proportions in municipalities over time.

Main Results:

• The proposed model successfully analyzes clustered proportion data, revealing distinct groups with similar voting patterns.
• The Bayesian approach with MCMC provides robust posterior distributions for model parameters.
• The analysis identified clusters of municipalities based on electoral behavior at different levels of favorable votes.

Conclusions:

• The developed finite mixture model offers a powerful tool for analyzing complex clustered proportion data in political science.
• The method facilitates the identification of regional electoral behavior clusters, enhancing understanding of political trends.
• This approach provides a robust Bayesian framework for analyzing time-series proportion data with clustering effects.