Yuan-chin I. Chang

The testlet is defined as a group of correlated test items, which is longer than an individual item and shorter than an entire test. The items within a testlet are usually assumed to be correlated due to, for example, sharing the same content. The concept of testlet was first introduced by Wainer and Kiely (1987) (see Wainer, et. al. (2007)). In this talk, we study the statistical model for the testlet-based testing; i.e. to study the estimation and testing problems of computerized adaptive testing which uses the testlet as the unit of construction. Instead of the models suggested in the literature, we apply the modern statistical regression techniques of correlated data to analyze the testlet response data such that the correlations among items can be correctly estimated from the observations. The estimate of latent trait level of examinees under such a model are studied and its statistical properties are reported. The results are compared to that of some models available in the literature. Besides the latent trait estimation problem, we also study the the testlet based mastery testing, in which a group sequential method is considered. Due to the correlated structure among test items within each testlet, the classical sequential theory does not usually applied to such kinds of the testlet based testings. The theoretical justification of applying sequential methods to the testlet based testing is given here, and numerical studies are presented for illustrating the performance of the proposed methods.

**Reference:**

Wainer, Bradlow and Wang (2007). Testlet response theory and its applications, Cambridge University Press.

**Keywords:** Sequential method; Testlet response theory; Generalized estimating equation; Adaptive testing

**Biography:** Yuan-chin Ivan Chang obtained PhD degree in Statistics from University of Illinois, Urbana/Champaign., and is now a research fellow of Institute of Statistical Science, Academia Sinica, Taipei, Taiwan. His research interest includes sequential analysis, generalized linear models, longitudinal data analysis, classification method, bioinformatics and machine learning.