Research On Mixed Item Type CAT ISS

There are many researches congcerning Computerized Adaptive Testing(CAT) item selection strategy(ISS). In recent years this topic has attraceted many Chinese scholers and the applications of Polytomous items in the test have been wildly used. However, mixed type selection strategy conbined dichotomous items and Polytomous items have barely been studied.As the form of mixed type selection strategy is easyly accepted, and can also provide abundant subjects' level information, thus is very common in our country.In this paper, two studies were carried out by Monte Carlo simulation. Study 1, combination of three classic questions for hybrid computer adaptive test, compared this 9 selection strategies. Study 2, based on Study 1, put forward a new type of hybrid suitable for computerized adaptive test selection strategy.The evaluation of the selection strategy is mainly from the accuracy, robustness and security of the test. In this paper, we choose five indicators, they are namely, the mean deviation Bias, the absolute mean deviation ABS, the average standard deviation RMSE, mean exposure rate of all the items De and chi square value. ABS, Bias is used to evaluate the accuracy of ability estimation, RMSE is used to evaluate the robustness of the rating ability, De and chi square value to evaluate the item call uniformity and item exposure rate. The five indicators are the smaller the better.Simulation study results show that the performance of commonly used selection strategy is poor while directly been used in the mixed types of computerized adaptive test. The maximum amount of information selection strategy for the mixed types of computerized adaptive test will get a high test precision, but is at the expense of great test safety. Higher level of the item can get a larger amount of item information, this is because it repeatedly select a higher level of the item, which will threat the safety of high level items. Risk still exist in class difficulty matching topic selection strategy. Greater number of higher grade items will also be selected in the mixed type of computerized adaptive test, ignoring the other difficulty parameters b and the discrimination parameters a of the item. All these will threaten the safety of the high grade item, which leads to the low accuracy of the test. Therefore, the level of the item must be considered in the mixed type selection strategy. The new selection strategy proposed in this paper can solve this problem very well. The new selection strategy can not only guarantee the accuracy of the estimation of the test ability level, but also dynamically control the distribution of the various grades of the item, so as to better control the item exposure rate of each item and mean exposure rate of all the items, which will make the test more secure.