在教育测量理论蓬勃发展的背景下，如何结合学科特点，将合适的教育测量理论运用到地理测评的分析中，是当前地理教育测评的重要研究议题。目前，常用的教育测量理论有经典测量理论（CTT）和项目反应理论（IRT）。项目反应理论因具有不依赖被试样本、测量精度更高等优势，在教育测评领域被广泛使用，其中又以Rasch模型的应用为主。本文对经典测量理论、项目反应理论和Rasch模型进行了系统的概念界定和文献综述，发现在地理教育领域很少有研究应用Rasch模型进行测评分析。针对这一研究领域的空白，本文将研究目标确定为：基于Rasch模型对北京三校高一上学期地理学科的学期总测进行质量分析，总结试卷中出现的问题并提出改进的方案。
本文的研究样本来自北京市房山区、延庆区、海淀区三所中学的学生对试题的作答表现，三校生源水平差别较大，基本涵盖了北京不同能力的学生群体。本文选取Winsteps（3.72.0）软件进行Rasch模型的Logistic回归分析，根据软件输出的信效度、拟合度、怀特图等指标进行试题分析，得出了三所中学试题质量的反馈信息。基于Rasch模型的分析结果表明：三校试题整体信效度较高，满足Rasch模型的假设，大部分题目和Rasch模型的拟合较好，试卷整体信息量较大。
Rasch模型的输出结果也反映出了三校试题中存在的主要问题，包括：部分题目间存在局部相关，部分试题与Rasch模型拟合不佳，部分试题过于简单和试题信息量不足。针对这些问题，本文通过质性分析和量化分析相结合的混合分析方法提出了针对性的改进方案，并凝练出了基于Rasch模型反馈信息的试题解决方案：对于存在相关性的题目，需要结合试题内容进行分析和改进；对于欠拟合的题目，可以通过残差分析法，结合学生作答切片进行改进；对于低难度的、信息量不足的题目，主要通过情境、知识内容与学科能力、干扰项设置三个方面进行修改。

In the context of the flourishing development of educational measurement theories, it is an important research topic in current geography educational assessment that using appropriate educational measurement theories to analysis geography assessment. At present, the commonly used educational measurement theories are Classical Measurement Theory (CTT) and Item Response Theory (IRT). IRT is widely used in the field of educational assessment due to its advantages of independence of the sample and higher measurement accuracy, among which the application of Rasch model is the main one. After a systematic conceptual definition and literature review of CTT, IRT and the Rasch model, we found that using Rasch model to analyze geography education is a gap in the current researches. In response to this gap, this study aims to analyze the first semester geography general test in three Beijing senior high schools based on the Rasch model, summarizes the problems in those tests and then proposes the strategies of improvement.
In this study, we selected the first semester geography general test of three Beijing senior high school in Fangshan District, Yanqing District and Haidian District. Using Winsteps (3.72.0) software to conduct logistic regression analysis of Rasch model, we carried out the quality analysis of those tests according to the feedback information from Winsteps such as validity, reliability, fitness and Wright maps. The results of the analysis based on the Rasch model showed that the overall reliability of those tests was high, the assumptions of the Rasch model were satisfied, most items fitted well with the Rasch model, and the information of those tests was adequate.
The output of the Winsteps also indicated some main problems of some items, including partial correlation between some items, poor fitness, low difficulty and insufficient information. To address these problems, this paper proposes some strategies of improvement through a mixed analysis method combining qualitative analysis and quantitative analysis.
For items with partial correlation, judgement needs to be made based on the content of those items. For under-fitted items, modifications can be made through residual analysis and students’ responses. For items with low difficulty or insufficient information, improvements can made through those aspects: situation, key concepts & geographical abilities, and Distractors.