多维计算机化自适应测验选题策略的研究与应用

Multidimensional computerized adaptive testing (MCAT) features a combination of tailored testing and multitrait estimation which shows great potential to improve test efficiency and support formative assessments than traditional CAT. So, MCAT has become an interesting and frontier research fields in psychological and educational measurement and has been addressed much concern especially in recent years..　　　Up to now, research reports of MCAT are not ample and most researches are focus on item selection methods and termination rule for tests containing dichotomous items and measuring two dimensional abilities. There are hardly any studies of MCAT about polytomous items. .　　　The objectives of this project are exploring theories and methods for item selection and examination system, and testing the validity of assessments made by MCAT in practice through application of psychological, mathematics and computer science. Firstly, following three sub-studies are executed for the aims of increasing measurement precision, controlling item exposure and satisfying content constraints in item selection. More specifically, they are: (a) proposing simple and valid item selection methods for tests contained with dichotomous items and measuring at least three dimensions; (b) putting forward item selection methods for multidimensional tests which were included polytomous items and (c) exploring item selection methods suitable for mix-formed tests which were consisted with both dichotomous and polytomous items. Then, algorithms of online automatic scoring for polytomous item will be developed before constructing the examination system of MCAT. Last, synthesized mathematics tests of the fifth grade in primary school will be took as an example and an empirical study of MCAT will be done with the cooperation of Institute of Educational Science in Sichuan province in order to examining the practical validity of MCAT. Moreover, new problems will be continuously found and solved in practices.

多维项目反应理论计算机化自适应测验（MCAT）与传统CAT相比，在提高测验效率的同时丰富了测评信息，是心理与教育测量的前沿研究领域。.　　迄今为止，国内、外关于MCAT的研究不多，且主要以二维为例研究二级评分项目的选题策略和终止规则，几乎没有多级评分项目MCAT的相关研究。.　　本项目结合心理学、数学和计算机科学，探索MCAT中项目选择和系统开发的相关理论与方法，并在实践中检验MCAT的诊断效度。首先，针对测量精度、曝光控制和内容约束问题开展以下研究：对高维度、二级评分项目，探索计算简单、有效的项目选择方法；研究MCAT中多级评分项目的选题策略；探索包括二级和多级评分项目测验的选题策略。其次，开发多级评分项目在线自动评分算法和MCAT系统。最后，与四川省教育科学研究所合作，以小学五年级数学综合测试为例开展实证研究，以考察理论研究的实践效能，并且在实践中发现新问题、解决新问题。

项目组紧密围绕多维计算机化自适应测验（MCAT）选题策略的理论与实践开展以下研究：文献研读、MCAT选题策略研究、MCAT系统开发、题库建设以及多维项目反应理论(MIRT)的实践应用。研究发现一些重要结果，完成预期目标，对MCAT理论和实践具备一定的指导意义。.首先，针对项目的核心内容“MIRT”、“MCAT选题策略”和“多级评分项目诊断测验”，通过文献研读发现双因子多维模型(bifactor model, BFM)和MCAT在国内没有得到足够认识，缺乏相关的理论和应用研究。项目组完成三篇（发表两篇、录用一篇）文献综述，以期为心理、医学和教育领域相关工作者概览BFM与多级评分诊断模型的特征、技术与应用；了解MCAT的研究问题、现状和进展提供参考。.其次，围绕MCAT“项目曝光控制”、“二级、多级评分项目选题策略”和“混合测验设计”开展理论研究和模拟实验，完成四篇（录用一篇，退修两篇，评审一篇）研究论文。研究证明BFM下维度缩减方法将多维能力积分化简为多个二维积分，极大地降低了计算量；研究发现维度缩减方法还适用于题组效应模型。通过模拟实验，研究提出了MCAT中最优项目曝光控制方法、二级和多级评分项目测验中最优项目选择方法以及混合测验中多级评分项目的最优施测比例和施测顺序。研究解决了MCAT计算复杂，耗时长的实际困难，拓广题组项目测验数据的分析方法，并为MCAT实践中项目曝光控制、选题策略和混合测验设计提供模拟结果的证据支持和建议。.再次，课题组开发了融合认知诊断理论和MIRT的在线CAT考试系统；基于MIRT构建了小学五年级数学综合测试题库；并开展多项MIRT实践研究。完成六篇（刊出3篇，录用1篇，评审1篇，撰写1篇）有关MIRT在“数学素养测评”、“数学成绩和心理特质的关系”以及“模型拟合”方面的研究报告。研究发现MIRT在模型数据拟合和实践应用方面具有许多优势，为教育工作者提供建议，并为MIRT的实践应用提供了借鉴和参考。.基于MCAT模型基础、选题策略和实践的研究为今后指明了模型研究的新方向、题组项目测验研究的新问题以及MCAT应用的新领域。