A1240
Title: New robust estimation methods for diagnostic classification models
Authors: Elena Castilla - Universidad Rey Juan Carlos (Spain) [presenting]
Abstract: A new estimation framework is presented for Cognitive Diagnosis Models (CDMs), focusing on the Loglinear CDM (LCDM). CDMs aim to provide detailed information about individuals mastery of multiple skills, making them especially useful in educational assessment. While joint maximum likelihood estimation (JMLE) has recently been shown to be statistically consistent for these models, it may be sensitive to model misspecification and data contamination. To address this limitation, a class of joint minimum divergence estimators based on the Cressie Read family of divergences is introduced. This approach generalizes JMLE and allows for a flexible trade-off between efficiency and robustness. The theoretical properties of the proposed estimator are established and its performance is evaluated through simulation studies and real data applications. Results suggest that the new method preserves the desirable asymptotic properties of JMLE while offering improved robustness in finite samples.
