LiMMCov: An interactive research tool for efficiently selecting covariance structures in linear mixed models using insights from time series analysis

• Published in: PloS one Vol. 20; no. 6; p. e0325834
• Main Authors: Savieri, Perseverence, Stas, Lara, Barbé, Kurt
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 11.06.2025; Public Library of Science (PLoS)
• Subjects: Algorithms; Autoregressive models; Complexity; Computer and Information Sciences; Covariance; Data analysis; Data visualization; Engineering and Technology; Estimates; Humans; Linear Models; Longitudinal method; Longitudinal Studies; Methods; Mixed-effects models; Parameter estimation; Physical Sciences; Research and Analysis Methods; Researchers; Software; Specifications; Statistics; Structural equation modeling; Time series; Time-series analysis; Belgium
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0325834

The correct specification of covariance structures in linear mixed models (LMMs) is critical for accurate longitudinal data analysis. These data, characterised by repeated measurements on subjects over time, demand careful handling of inherent correlations to avoid biased estimates and invalid inferences. Incorrect covariance structure specification can lead to inflated type I error rates, reduced statistical power, and inefficient estimation, ultimately compromising the reliability of statistical inferences. Traditional methods for selecting appropriate covariance structures, such as AIC and BIC, often fall short, particularly as model complexity increases or sample sizes decrease. Studies have shown that these criteria can misidentify the correct structure, resulting in suboptimal parameter estimates and poor assessment of standard errors for fixed effects. Additionally, relying on trial-and-error comparisons in LMMs can lead to overfitting and arbitrary decisions, further undermining the robustness of model selection and inference. To address this challenge, we introduce LiMMCov, an interactive app that uniquely integrates time-series concepts into the process of covariance structure selection. Unlike existing tools, LiMMCov allows researchers to explore and model complex structures using autoregressive models, a novel feature that enhances the accuracy of model specification. The app provides interactive visualisations of residuals, offering insights into underlying patterns that traditional methods may overlook. LiMMCov facilitates a systematic approach to covariance structure selection with a user-friendly interface and integrated theoretical guidance. This paper details the development and features of LiMMCov, demonstrates its application with an example dataset, and discusses its potential impact on research. The app is freely accessible at https://zq9mvv-vub0square.shinyapps.io/LiMMCov-research-tool/ .