Exploratory and confirmatory neurocognitive modeling with latent variables
| Agency | National Science Foundation |
| Panel | Methods, Measurement, and Statistics |
| Location | University of California, Irvine |
| Start date | June 2021 |
| End date | May 2024 |
| Budget | $ 349,551.00 |
| Agency code | 2051186 |
| Lead | Joachim Vandekerckhove |
| Other | Ramesh Srinivasan |
| Project text | pn-2051186.pdf |
Abstract
This research project will improve methods for analyzing brain and behavior measurements. Existing methods for combining multiple sources of brain data are limited. They cannot take advantage of the joint measurement of different kinds of brain activity to predict behavior. This project will combine multiple recent developments in mathematical psychology and cognitive neuroscience to arrive at a powerful new data analysis method that can process many sources of data at once. The methods to be developed will identify the shared and unique contributions of brain regions to behavior and can be used to discover new functions of brain regions. Broader use of these methods also will facilitate theory-building in cognitive (neuro)science. The project will involve early-career scientists who will receive training in neuroscience or cognitive science. The investigators will document their modeling techniques as part of a series of publicly available video lectures. The project also will develop a large code base and user-friendly software, thus contributing to the global research and education infrastructure.
This project will combine recent developments in mathematical psychology (Bayesian latent variable modeling with cognitive models) and cognitive neuroscience (modeling of neurocognitive relationships) to develop new methods for analyzing brain data. Current approaches to brain-data analysis emphasize looking for separate relationships between individual physiological measures and behavior. Direct analysis of the relationship between different types of physiological measures usually does not involve behavioral data. Identification of the processes that give rise to patterns in different measures is carried out without direct links to behavior. However, multiple neuroimaging techniques can provide common underlying information that informs cognitive models of human behavior as well as the physical location of the ongoing processes. For instance, both fMRI and EEG are driven, at least in part, by the same underlying neural processes. At the same time, each technique also contains unique information, so that some cognitive processes underlying human behavior can potentially only be informed by specific techniques or combinations thereof. This project will make it possible to translate such theories about these complex relationships into principled statistical models whose predictions and assumptions can be put to the empirical test. The new methods will be applied to archival data sets, thus contributing to generalizable knowledge in a way that is cost-effective and ethical.