Tracking Attribute Mastery Change among Individuals: Longitudinal Diagnostic Classification Models with Random Intercepts

Tracking an individual’s learning status is important to ensure improvement. previously developed longitudinal diagnostic classification models (LDCMs) are limited in interpreting attribute mastery transitions as individual processes because they ignore group- and individual-level components. Therefore, the RI-LDCM was developed by considering such group- and individual-level variations as random intercepts (RI) in the measurement model. A Bayesian estimation method for the RI-LDCM was also developed. A simulation study revealed that ignoring multilevel structures causes biased parameter estimates and serious under-coverage of posterior credible intervals. A real data example provided attribute mastery transitions from the four-year longitudinal data of mathematics tests in the third to sixth grades of Japanese elementary schools. The extensibility and limitation of the RI-LDCM were also discussed.