Comparing Collapsed Gibbs Sampling and Posterior Mode Estimation on Estimating DCM (Diagnostic Classification Model) Attribute Mastery Profiles (Poster 17)

• In Event: Thursday Poster Session 10:50 am
  In Poster Session: Division D Poster Session 1

Abstract

Diagnostic Classification Models (DCMs) have emerged as valuable tools for assessing individuals' skills and attributes in various fields. In estimating attribute mastery profiles in DCMs, Collapsed Gibbs Sampling (CGS) and Posterior Mode Estimation (PME) are two reliable methods. This research compares the performance of CGS and PME by conducting simulation studies under various conditions, including sample size, number of items and attributes, item discrimination, and Q-matrix mis-specification. The R, Python, and Mplus software are employed for data generation and analysis. Results indicate that (1) both CGS and PME achieve high Classification Correct Rates (CCR) for individual attributes and the overall pattern, and (2) CGS outperforms PME on classification accuracy, presenting it a promising alternative for attribute profile estimation in DCMs.

Authors

• Jingyang Li, University of Georgia

• Zhenqiu Laura Lu, University of Georgia

• Matthew J. Madison, University of Georgia

• Zuchao Shen, University of Georgia