Modelling longitudinal polytomous animal data using Bayesian hierarchical models

2025-07-22

AI Summary

Modeling longitudinal nominal multicategorical data (e.g., animal behavior) faces challenges including excessive parameters, poor convergence under small sample sizes, and overdispersion. This paper proposes a Bayesian hierarchical modeling framework that employs noninformative priors and MCMC sampling for robust estimation in R, markedly improving convergence and parameter stability. Its key contributions are: (i) the first systematic application of hierarchical Bayesian methods to longitudinal nominal categorical data, jointly accommodating inter-category heterogeneity and within-subject temporal dependence; and (ii) effective mitigation of overparameterization via prior shrinkage and structured random effects. In an empirical analysis of 12-week longitudinal behavioral data from pigs across seven categories, the model achieved stable convergence and yielded interpretable fixed- and random-effect estimates. Open-source code ensures reproducibility and establishes a novel analytical paradigm for agricultural science and related domains.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Machine Learning: Bayesian Learning

Application Category

• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The analysis of longitudinal categorical data can be complex and unfeasible due to the number of parameters involved, characterised by overparameterisation leading to model non-convergence, in addition to problems related to sample size and the presence or absence of overdispersion. In this context, we introduce Bayesian hierarchical models as an alternative methodology to classical statistical techniques for analysing nominal polytomous data in longitudinal studies. The theoretical foundation is based on the use of non-informative priors and advanced computational techniques, such as Markov Chain Monte Carlo (MCMC) methods, which enable a robust and flexible data analysis framework. As a motivating example, the procedure is illustrated through an applied study in agrarian science, focusing on animal welfare, which assessed seven types of behaviours exhibited by pigs over twelve weeks. The results demonstrated the efficacy of Bayesian hierarchical models for the analysis of longitudinal nominal polytomous data. Since the computational procedures were implemented in the R software and the codes are available, this work will serve as support for those who need such analyses, especially in agricultural designs, where longitudinal categorical data are frequently encountered.

Problem

• Addresses complexity in longitudinal categorical data analysis
• Proposes Bayesian hierarchical models for polytomous data
• Demonstrates efficacy in animal behavior studies

Innovation

• Bayesian hierarchical models for polytomous data
• Non-informative priors with MCMC methods
• R software implementation for agricultural studies