Multivariate versus univariate latent trajectory modelling: a comparison of class recovery performance and implications for multivariate model building strategies

The Group-based multi-trajectory model (GBMTM) extends the univariate Group-based trajectory model (GBTM) to analyse multivariate longitudinal data by identifying subgroups with similar developmental patterns across multiple outcomes. This method has gained popularity for exploring complex phenomena from developmental and relational perspectives across various empirical fields. Despite its utility, comparing GBMTM with preliminary GBTM analyses poses challenges due to potential discrepancies in trajectory characteristics such as numbers, sizes, levels, and shapes across outcomes. These differences suggest a complex data-generative process not fully understood. Our study aims to bridge this knowledge gap by examining how longitudinal data features impact class enumeration and parameter recovery in GBMTM and GBTM through extensive simulations. We highlight the influence of several factors on multivariate clustering, notably outcomes' class separation and the strength of univariate class correspondence. By addressing analytical and interpretational challenges, our findings offer practical guidelines for GBMTM, illustrated with real-world data examples.