Title: Longitudinal Modeling of Age-Dependent Latent Traits with Generalized Additive Latent and Mixed Models
Abstract: In cognitive neuroscience, the goal is to understand how properties of the brain relate to cognitive function, and particularly how changes in the brain lead to changes in cognitive function. Typical datasets are large and may contain brain scans and cognitive test scores from tens of thousands of individuals, sampled over decades across countries. While latent variable models have well-known advantages, their use in cognitive neuroscience have been limited by restrictive assumptions not met by the typical datasets, such as regularly spaced timepoints, nested hierarchical structures, and parametric effects. In this talk I will start by presenting a framework for semiparametric latent variable modeling which relaxes most of these assumptions (Sørensen, Fjell, Walhovd, Psychometrika, 2023, https://doi.org/10.1007/s11336-023-09910-z). In the second half of the talk, I will discuss important future challenges for latent variable modeling with big and complex data, including computational algorithms, nonsensical zero variance estimates, and accuracy of confidence intervals.