Reference
Sørensen, Ø., Fjell, A. M., & Walhovd, K. B. (2023). Longitudinal Modeling of Age-Dependent Latent Traits with Generalized Additive Latent and Mixed Models. Psychometrika, 88(2), 456-486. https://doi.org/10.1007/s11336-023-09910-z
Review
In their 2023 study, Sørensen, Fjell, and Walhovd introduced generalized additive latent and mixed models (GALAMMs) to analyze clustered data. They developed these models primarily to address applications in cognitive neuroscience. Their method leverages a scalable maximum likelihood estimation algorithm, utilizing advanced computational techniques like the Laplace approximation, sparse matrix computation, and automatic differentiation. Crucially, this approach allows for a variety of mixed response types, heteroscedasticity, and crossed random effects.
The authors further illustrated the applicability of GALAMMs by presenting two case studies. The first highlighted how these models could comprehensively capture lifespan trajectories of various cognitive abilities, including episodic memory, working memory, and executive function. Such findings were drawn from widely used cognitive tests like the California Verbal Learning Test, digit span tests, and Stroop tests. In their second case, the researchers explored the impact of socioeconomic status on brain structure, specifically delving into the relationship between educational and income levels with hippocampal volumes, gauged via magnetic resonance imaging (MRI). Their results posited that by integrating semiparametric estimation with latent variable modeling, GALAMMs can offer a more nuanced depiction of how both the brain and cognition evolve throughout an individual's life.
Overall, this study presents a promising tool for the analysis of complex data structures, especially in the realm of cognitive neuroscience. While the authors provided solid evidence from their case studies, it would be beneficial to see how GALAMMs fare in a broader range of applications. Moreover, the efficacy of these models in different sample sizes, beyond moderate ones, remains a question worth exploring in future research.
