山口 一大
Some Model Comparisons in Cognitive Diagnostic Models
Relatively Small Sample Situation with Simulation Study
Kazuhiro Yamaguchi
April, 2018
Abstract
A lot of cognitive diagnostic models (CDMs) have been developed in several decades. The objective of this study is to check how we can detect misspesifications among data generation models and analysis models in relatively small sample size situations. We employed simulation study for the purpose. We got three results. First, that Bayesian information criterion (BIC) indicated LLM (linear logistic model) as optimal model when G-DINA (generalized deterministic noisy inputs “and” gate) model was true model. Second, when the LLM and A-CDM (additive CDM) were true models, it was difficult to distinguish these model with Akaike information criterion (AIC) and BIC. Third, AIC and BIC can select R-RUM (reparametarized reduced unified model), DINA (deterministic noisy inputs “and” gate) model and DINO (deterministic noisy inputs “or” gate) model models as correct model. We discuss these results.