Item Selection Methods in Polytomous Computerized Adaptive Testing

Given the rapid advancement of computer technology, the importance of administeringadaptive tests with polytomous items is in great need. With regard to the applicability ofadaptive testing using polytomous IRT models, adaptive testing can use polytomous items of either rating scales, or in some testing situations of multiple choice. Additionally, theavailability of computerized polytomous scoring of open-ended items enhances suchapplicability. This need promotes the research in polytomous adaptive testing (PAT).This dissertation is an e ort to focus on item selection methods, as a major component, inpolytomous computerized adaptive testing. So, it consists of ve chapters that cover thefollowing:Chapter 1 focuses on a thorough introduction to the item response theory (IRT) models and adaptive testing related to polytomous items. Such an important overviewand introduction to basic concepts in test theory and mathematical models for polytomous items is needed for the ow of consequent chapters. Chapter 2 is devoted to the development of a central location index (LI) to uniquely represent the polytomous item with a scale value parameter using most commonly used polytomous models. The motivation and rationale to search for a central or an overall location parameter is twofold:a) the confusion of multiple and di erent parameterizations for a polytomous item even for the same model, and b) the unavailability of such single location parameter block the usage of certain item selection methods in adaptive testing. Two approaches are used toderive the proposed LIs, one is based on the item category response functions (ICRFs) and the other is based on the polytomous item response function (IRF). As a result, four LIs are proposed. Chapter 3 is particularly assigned to development of an item selectionmethod based on the developed location index and primarily assess its performance in thePAT context relative to existing methods. This method belongs to the non-information based item selection methods and we referred it as Matching-LI method. The results support that this proposed method is promising and is capable to produce accurate abilityestimates and successfully manage the item pool usage. Chapter 4 introduces new itemselection methods taking in consideration the previous chapter's results. The newmethods are the hybrid, stage-based information, polytomous a-strati cation methods. The first two methods try to merge more than one criterion for selecting items of each PAT (e.g., the hybrid method merges both the Matching-LI and maximum information(MI) methods). The last method uses Matching-LI method within each stratum. Chapter 5 provides discussion, conclusions, and limitations and future research directions withrespect to important components of an adaptive testing program (i.e., item selection methods, item response models, item banks, and trait versus attribute estimation).