Developing Three New Strategies Of Item Selection Of Computerized Adaptive Testing

In the last decade, one of the most significant advances in psychometric practice is computerized adaptive testing (CAT). In this way, computer can select items from item pools to examinees according to performance on the items answered previously. Then, examinees can obtain quicker and more reliable measures of the trait levels of themselves than with conventional paper‐and‐pencil tests.Compared with the present, in the earlier, most simple computerized test let computer takes of a random paper and make examinees answer. This method is still the traditional paper‐r‐and‐pencil tests methods. The role of the computer is just a simple media tools, almost no effect on the formation of test preparation and low intelligence. Besides, there are many drawbacks: first, simultaneity, which need to be tested at the same time or even in the same place and add the difficulty to arrange the time and the place. Second, the value of ability is unknown. Examinees are likely to tested too simple or too difficult items. So, not only the ability can’t be accurately tested, but waste time and energy. However, with the introduction of computerized adaptive testing, the computer is no longer a mere media tools and has the functions of decision‐makers, expanding the new world of psychological measurement and opened up new prospects.Computerized adaptive testing usually consists of six major components, the response patterns of the project, composition of banks, Item selection strategy, trait estimation and test termination rules. Item selection strategy is a key part of computerized adaptive testing and will have a direct impact on accuracy, safety, efficiency and reliability of the testing. The stratification of the item selection strategy is a kind of common strategy. In this paper, stratification strategy is improved and three new Item selection strategies are proposed: First, new item selection Strategy based on Sampling principle, improving item selection strategies between Maximum Information Stratification and A‐Stratification with exposure‐control factor and similar distributed stratification method for computerized adaptive testing. In addition, comparing item selection strategies on different proportions partitioned item banks is also a part of this paper. According to Monte Carlo simulation, the results show that the new approaches have more preferably performance comparing with the original method on several indexes. The effect of different proportions partitioned item banks presents quite different on most indicators.