Research On Item Response Theory Based Computerized Adaptive Testing System

Computerized Adaptive Testing has many advantages that classical testing in papers and pens can't provide: the ability to provide dynamic testing service personaly individually based on different testing status of each testee. Compared with Classical Testing Theory, Item Response Theory has such advantages like the independence of test; repeatable test and so on while also has such disadvantages like the complexity of the algorithim and the difficulty of the appliance in the real testing project and so on. However Item Response Theory based Computerized Adaptive Testing achieves the great complementary, exert the power of both: efficiency and agility.At the moment, Item Response Theory based Computerized Adaptive Testing is widely used in various kinds of testing applications. For example, GRE Testing is one of the most famous. But there are still some difficulties and challenges when applying Item Response Theory based Computerized Adaptive Testing shown as below: the selection of IRT models; the choice of estimating algorithm based on the selected Item Response Theory model; the strategy of the Computerized Adaptive Testing. To solve those problems listed above, it's usually necessary to design and apply the Item Response Theory based Computerized Adaptive Testing based on the actual testing project itself.This paper's main purpose is to represent the research, analysis and design of the Item Response Theory based Computerized Adaptive Testing system for the application of the testing project named"Autism and Retarded Children Testing Project"leadered by Shanghai Mental Health Center and Waterville Inc. Based on the comparison and analysis on the existed Item Response Theory models, estimating algorithms and strategies of Computerized Adaptive Testing with the consideration of the actual requirement of the project, one executable designing proposal for the Item Response Theory based Computerized Adaptive Testing system combining with Logistic 3PM model, Maxium Likelihood Estimating, Max Information Method is raised in the paper. Also the enhancement of Maxium Likelihood Estimating is proposed as the bright point in the paper. The core idea in this creative enhancement is that Interval-Iteration is used instead of Initial Value Estimation on the purpose of rootly avoiding the situation that Maxium Likelihood Estimating will fail in the iteration process or will output a bad result when the estimated initial value is not so good.