Bayesian Semiparametric Longitudinal Inverse-Probit Mixed Models for Category Learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Statistics

Background:

• Understanding adult human brain learning of novel categories is crucial.
• Drift-diffusion models are common for mimicking neural mechanisms in learning.
• Existing models often require response times, which are not always available.

Purpose of the Study:

• To derive a novel class of biologically interpretable 'inverse-probit' categorical probability models.
• To address identifiability and inferential challenges in models without latent response times.
• To adapt the model for group and individual-level inference in longitudinal studies.

Main Methods:

• Building upon Paulon et al. (2021), latent response times were integrated out.
• A novel projection-based approach with a symmetry-preserving identifiability constraint was developed.
• An efficient Markov chain Monte Carlo algorithm was designed for posterior computation.

Main Results:

• A new marginal model for observed categories was derived, overcoming previous limitations.
• The method successfully handles identifiability and inferential challenges.
• The model was adapted for longitudinal group and individual inference.

Conclusions:

• The developed 'inverse-probit' model provides a new tool for analyzing category learning data.
• The projection-based approach and MCMC algorithm enable robust inference.
• The method's practical efficacy is demonstrated in longitudinal tone learning studies.