Technical and Practical Considerations for Fitting Loglinear Cognitive Diagnosis-Latent Space Item Response Model

• In Event: AERA Research in Progress Roundtable Session Two
  In Invited Roundtable: Research-in-Progress RT002 (Thursday)

Abstract

Pairing the Latent Space Item Response Model with the Loglinear Cognitive Diagnosis Model (LSIRM-LCDM) can reveal local dependence and other structures a cognitive diagnosis model alone may miss. The obstacle is fitting it. This work creates a practical roadmap for doing so. Using benchmark datasets and realistic simulations, it highlights typical causes estimation stalls in MCMC, sets run-time expectations for common test and sample sizes, and identifies defaults that reliably work. It also offers guidance on when to use JAGS or Stan, as well as code templates and model-fit checks. The focus is on estimation mechanics as opposed to interpretation.

Author

• Brian Harrold, University of South Carolina