Local independence is one of the major assumptions when applying item response theory (IRT) models, which includes local item independence and local person independence. Recent studies have proposed models to account for one of the violation of local independence assumptions, for example, testlet models (two-parameter Bayesian random-effects testlet model, Rasch testlet model and the three-level one-parameter testlet model) were proposed to account for local item dependence; multilevel models were proposed to account for local person dependence. However, there are limited studies to deal with local item dependence and local person dependence simultaneously. The major purpose of this study is that proposed a multilevel testlet model to deal with the violation of dual local independence, such as the examinees are selected from different schools or classes in a teselet-based assessment.
Questions & Comments
1. How to calculate the effect size (f) which is used in the results part?
2. For demonstrating the classification accuracy difference between the proposed model and other compared models, presented a table does more understand than that by words.
3. The result shows that the item difficulty parameter estimates were not greatly affected by estimation models but the ability parameter estimates were. Whether it implies that when the dual local dependence exists, it is probably to use a multilevel model instead of a four-level IRT model for dual local dependence for avoiding large number of parameters?