【作者】 周骏；

【导师】 戴海琦；

【作者基本信息】 江西师范大学 ， 基础心理学， 2008， 博士

【摘要】 传统测验重视统计技术,不重视测量结构的心理学意义,使得它的功能局限于筛选,不能提供更多的信息。近年来,心理学理论的发展和心理计量技术的革新对测验发展的方向影响巨大。认知心理学不再将人类的内心世界视为“黑箱”,也不再将被试作答的过程抛开,认知心理学中的认知成分分析对测验的发展有关键性的影响。Sternberg(1991)指出未来测验发展的方向应结合认知心理学理论、心理计量学理论以及教学,使测验的发展具有认知心理学的基础,测验的结果能提供有关对被试的诊断信息,测验的分数能反映出答题的心理过程等。由于认知心理学理论、心理计量学以及计算机技术的发展,使得基于认知理论指导下的项目生成技术逐渐成熟。在诸多关于智力的传统测验中,矩阵完成问题以其适用范围广、使用方便以及结果解释直观、简单等优点得到了广泛的使用。智力的PASS理论以神经心理学为基础,整合信息加工心理学、心理测量学,突破以往智力理论停留在智力概念争论上的缺陷,使智力测验拥有了具体、系统的理论基础,为智力测验开辟了崭新的研究领域。本研究以PASS理论中同时性加工编码为研究基础,以矩阵完成问题为考察智力的形式,使用项目生成技术中的认知设计系统方法,分别对矩阵问题的认知模型、矩阵完成问题的项目生成算法,以及矩阵完成问题自适应项目生成的系统编制等进行研究。研究结果表明:1、自建认知模型在对矩阵项目难度等项目属性的预测方面优于其它认知模型。自建的认知模型包括矩阵问题加工理论、矩阵问题知觉变量两部分;矩阵问题加工理论是Carpenter et al.基于对瑞文测验中矩阵问题分析的基础之上得到的;矩阵问题知觉变量是基于对刘声涛等(2004)提出的矩阵问题绘图原则基础上归纳合并得到的;这两部分内容整合在一起的认知模型能够很好的预测矩阵问题的项目参数。2、项目生成算法编制的矩阵完成测验有较好的信度和效度。研究者设计项目结构编制矩阵测验,通过认知模型与编制矩阵测验项目的难度建立的数学模型,来验证认知模型为矩阵问题的自适应项目生成服务。研究中设计了27种项目结构,生成的矩阵测验与瑞文测验相比,难度相近且区分度稍好,矩阵测验与瑞文测验的相关达到0.937(p＜0.01),且内部一致性也相近,矩阵测验的项目比瑞文测验的项目要少。3、自编矩阵问题自适应项目生成测验(M-AIG)系统,对被试的抽象推理能力有较好的区分。对临时项目库的分析发现生成的项目具有较合适的心理计量属性,M-AIG系统的重测信度达到0.83(P＜0.01)。更多还原

【Abstract】 The traditional test paid attention to statistical techniques, rather than psychological meaning of the structure, which made its function confine to screening and couldn’t offer more information. In recent years, the development of psychological theory and the innovation of measuring technology of psychology have had great influence on the direction of development of test. Cognitive psychology is not regarded as "Black Case" again in human inner world, and examinee’ answering process is not set aside any longer. The analysis of cognitive composition in cognitive psychology and the theory of cognitive development by Vygotsky (1978) have key influence on the development of test. Sternberg (1991) pointed that the direction of test development in the future should combine cognitive psychology theory, psychological metrology theory and teaching, which made test development have foundation of cognitive psychology, made test result offer diagnostic messages of examinees, and made test mark reflect mental process of answering, etc. Because of the development of cognitive psychology theory, psychological metrology theory and computer technology, item generation technology based on cognitive theory has become mature gradually.Within many of the traditional intelligence tests, Matrix Completed Problem has been widely used with a wide range of applications, the advantages of easy-to-use as well as explaining intuitively and simply, etc. PASS theory of intelligence ,which is based on neural psychology and combines information processing psychology and psychological measurement, breaks through the defects of staying in the intelligence concept on past intelligence theory, makes the intelligence test have concrete, systematic theoretical foundations, and has opened up the brand-new research field for intelligence test.The research is on the basis of the simultaneous processing code of PASS theory, and in the form of Matrix completed problem to investigate intelligence. The cognitive model, the item generation algorithm and the adaptive item generation system of matrix completed problem are researched separately by using cognitive design system method of item generation technology. The results show that:1. The self-build cognitive models are superior to other ones on predicting item attributes such as the item difficulty of the matrix, and so on. The self-build cognitive models include two parts: the processing theories and the perceptual variables of matrix problem. The former are on the Carpenter et al’ analysis of matrix problems in Raven’s Progressive Matrices. The latter are on the summarization of mapping principles of matrix problems proposed by Shengtao Liu, et al (2004). The cognitive models integrating these two parts can predict the item parameters of matrix problems better.2. The matrix completed tests developed by item generation algorithms have a better reliability and validity. The researcher has designed item structures to develop matrices, and has built mathematical models with the cognitive models and the item difficult to verify the cognitive models for adaptive item generation. 27 kinds of item structure have been designed in the research. Compared with Raven’s test, the matrix test generated in the research has similar difficulties, but slightly better discriminations. Between the two tests, the coefficient reaches 0.937(p<0.01), the internal consistencyis similar, but items in the matrix test are fewer than in Raven’s test.3. It has better distinguish on examinees’ abstract reasoning ability by creating adaptive item generation system (M-AIG) of matrix problems. From the analysis of the temporary item bank, it is found that the generating item has more suitable psychological psychometric attributes and the test-retest reliability of M-AIG system reaches 0.83(p<0.01).更多还原