CAT常用能力估计方法比较及其优化：能力综合估计方法开发

发布时间：2018-04-20 03:09

本文选题：计算机化自适应测验 + 能力估计　；参考：《江西师范大学》2014年硕士论文

【摘要】：近年来，随着测量理论和计算机技术的发展，计算机化自适应测验（Computerized Adaptive Testing，CAT）受到人们越来越多的关注。能力估计技术在CAT中一直扮演重要角色，其估计的准确与否不仅影响选题策略的自适应，还会由此持续的影响CAT最关注的能力估计的准确性。 CAT的能力估计方法至今仍沿用IRT时代的几种主要方法，常见的包括MLE，MAP，EAP，WLE等。本文就CAT中能力估计方法的比较与开发开展了两项研究：研究一对四种常用CAT估计方法采用计算机蒙特卡洛模拟程序，分别从偏差，均方根误差，题库调用均匀性，测验效率等方面，进行了系统性的比较。研究二则是以研究一为基础，根据不同估计方法特点及优劣，开发了一种新的CAT能力估计方法——能力综合估计法，即强调在CAT能力估计的不同阶段，综合运用恰当的CAT能力综合估计方法，以期取长补短，发挥现有能力估计方法的优势，达到同时提高CAT能力估计的准确度及测验效率。研究结果表明： 1） MLE的偏差小但均方根误差大，曝光率相对其他方法更好，但测验效率最差，且对特殊作答模式无法给出有效的估计。 2） WLE的偏差最小，均方根误差多数情况下优于MLE，，在a分层选题且b均匀时曝光率最好，且最大信息量选题时的测验效率最高。 3） MAP的偏差最大，均方根误差较小，曝光率在大多数条件下与WLE，EAP并无区别，且a分层选题策略下的测验效率最高。 4） EAP的偏差仅次于MAP，但均方根误差最小，测验效率略低于MAP。 5）本研究提出的前期和中期用EAP，后期用WLE的能力综合估计法可以有效提高EAP的偏差并基本维持EAP的均方根误差。 6）综合法主要可以在控制均方根误差的基础上有效改善EAP的偏差。对EAP偏差的改善率可达到30%~40%，而均方根误差仅相比EAP差了不到5%。 7）综合法在不同长度的测验中均能有效改善EAP的偏差，其中短测验中改善的效果更好。
[Abstract]:In recent years, with the development of measurement theory and computer technology, more and more attention has been paid to computerized Adaptive testing. Capability estimation technology has always played an important role in CAT. The accuracy of the estimation not only affects the adaptive selection strategy, but also affects the accuracy of the capability estimation that CAT pays most attention to. The capability estimation methods of CAT are still used in the era of IRT, such as MLEMP-MAPE / WLE and so on. In this paper, two studies have been carried out on the comparison and development of capability estimation methods in CAT. A pair of four commonly used CAT estimation methods are studied by using Monte Carlo simulation program, respectively, from deviation, root mean square error, homogeneity of item bank, etc. The efficiency of the test is compared systematically. On the basis of research one, according to the characteristics of different estimation methods and their advantages and disadvantages, a new CAT capability estimation method-capability comprehensive estimation method is developed, which emphasizes the different stages of CAT capability estimation. In order to make use of the advantages of the existing ability estimation methods, we can improve the accuracy of CAT capability estimation and test efficiency by using the appropriate comprehensive estimation method of CAT capability. The results show that: 1) the deviation of MLE is small but the root mean square error is large, the exposure rate is better than other methods, but the test efficiency is the worst, and the estimation of the special response mode can not be given effectively. 2) the deviation of WLE is the smallest, the root mean square error is better than MLEs in most cases, the exposure is the best when a stratified topic is selected and b is uniform, and the test efficiency is the highest when the maximum amount of information is selected. 3) the deviation of MAP is the biggest, the root mean square error is small, the exposure rate is not different from that of WLEN EAP under most conditions, and the test efficiency is the highest under the strategy of a stratified selection. 4) the deviation of EAP is second to that of EAP, but the root mean square error is the least, and the test efficiency is slightly lower than that of EAP. 5) in this study, the error of EAP can be effectively improved and the root mean square error of EAP can be basically maintained by using the capability comprehensive estimation method of WLE in the early and middle stages, and in the latter stage. 6) the synthetic method can effectively improve the deviation of EAP on the basis of controlling root mean square error. The improvement rate of EAP deviation can reach 30% and 40%, but the root mean square error is less than 5% less than that of EAP. 7) the synthetic method can improve the deviation of EAP effectively in the test of different length, and the effect of improvement in the short test is better.
【学位授予单位】：江西师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：B841

【参考文献】