Integrating Neuroimaging and Behavioral Data Using The Multidimensional Generalized Graded Unfolding Model.
Barrett, Matthew E.
MetadataShow full item record
A study investigating the relationship between two distinct data structures resulting from the same stimulus was examined. Participants made attractiveness judgments to computer generated models in two phases. Phase 1 of the study was conducted in the laboratory (behavioral) while phase 2 was conducted in the fMRI scanner (neuroimaging). Data from the behavioral component was composed of attractiveness ratings for computer generated models, whereas the neuroimaging component was composed of signal change in five pre-specified ROIs when responding to the identical stimulus. It was hypothesized that both of these outcomes were a function of the distance between a subject’s ideal point and the stimulus location in a latent multidimensional preference space. The attractiveness ratings were modeled with the multidimensional generalized graded unfolding model (MGGUM), which is an item response theory model for proximity-based data presumed to underlie the general preference ratings. The signal change data was simultaneously modeled as a function of the estimated distance between a subject and stimulus derived from the MGGUM. Estimation of models for both types of data was conducted simultaneously using a system of two simultaneous equations with parameters that are updated using a Markov chain Monte Carlo procedure. Information about signal change and its relationship to person-stimulus distances (i.e., idealness) in the multidimensional latent space was utilized to update estimates of the individual’s location in that space and this, in turn, lead to updated predictions of signal change in each ROI. This project was predicated on the notion that both behavioral and neural signal data are a function of the proximity between a given individual and stimulus, and was the first study to integrate models for neural signal into an item response theory framework.