Generalized Linear Mixed Proficiency Models for Cognitive Diagnosis

Models for cognitive diagnosis combine discrete latent trait models with constrained latent class analysis, allowing the user to diagnosis the set of abilities an examinee may possess. Following a review of relevant cognitive diagnosis models and commonly used methods of estimation, implementation of a generalized linear mixed model for the proficiency space of examinee abilities (GLMPM) is developed using the Reparameterized Unified Model (or RUM, Hartz, 2002). The GLMPM provides a one-factor correlational model along with a covariate-weighted mean structure model for mixed type (discrete or continuous) latent variables. The effectiveness of the GLMPM is demonstrated in simulation studies across a wide variety of conditions designed to mimic realistic analyses. Also shown is a lack of robustness when latent variable correlations are not modeled. As an extension of the models presented, the RUM and GLMPM are generalized to allow for discrete-polytomous latent variables. Again, the effectiveness of the polytomous RUM and polytomous GLMPM is demonstrated in simulation studies. Finally, to demonstrate the usefulness of the GLMPM, all types of the proficiency space model are fit to two real data sets.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Psychology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Templin, Jonathan L.

Contributors dc:contributor

• Douglas, Jeffrey

• Psychology, Psychometrics

Language dc:language

eng

Identifier

(MiAaPQ)AAI3160960

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/82077