Investigation of optimal design and scoring for adaptive multii-stage testing: A tree-based regression approach
Recently, Multi-stage testing (MST) has received a lot of attention. Similar to computer adaptive testing (CAT), MST has the efficiency of testing and its applications also rely heavily on item response theory (IRT) currently. However, it is unrealistic to suppose that standard IRT models will be appropriate for all MST applications as it is for CAT applications (Yan, Lewis, and Stocking, 2004). This research introduces a nonparametric, tree-based algorithm for adaptive multi-stage testing with modules, and explores various designs and scoring for tree-based multi-stage testing. This new approach has several advantages over the traditional approaches to multi-stage testing, including simplicity, lack of restrictive assumptions, and the possibility of implementation based on small samples. The results of the study demonstrated the feasibility of the new approach and its ability to produce reliable scores with fewer items than a fixed linear test. The results also identified the optimal multi-stage design among the alternative designs considered. It is an extension of the tree-based CAT by Yan, Lewis and Stocking (2004).
Yan, Duanli, "Investigation of optimal design and scoring for adaptive multii-stage testing: A tree-based regression approach" (2010). ETD Collection for Fordham University. AAI3452799.