计算思维(Computational Thinking）是指通过信息加工媒介解决以各种软件、程序、数字驱动形式存在的问题。在小学阶段一般通过可视化编程对计算思维进行测评，考察学生利用基本编程基本概念知识设计算法解决程序类问题的能力及该过程中所表现出的行为模式和作答策略，前者为计算概念，后者为计算实践。随着国务院、教育部政策支持，我国已有部分地区的小学开展了可视化编程教育，为开展我国义务教育阶段计算思维测验工具的开发和优化带来了机遇。

尽管计算思维逐渐在小学教育阶段得到重视，但其测验模式及思维特点研究仍存在诸多待探究完善之处。在测验方式上，以往基于可视化编程的计算思维测验工具大致分为非插电和编程操作两种环境。非插电测验工具多采用选择题形式，通过信效度等检验，可实现大规模施测，但无法考察学生作答过程中的行为模式及作答策略。后者可深度剖析学习者的思维方式和行为过程，但时间和人力成本较高，对计算概念的考察完全依赖学生编写程序时所使用的代码。在学生计算思维特点上，计算概念的特点研究大多停留理解层面，没有深入考察学生的应用能力；对计算实践的研究集中于抽象与模型、再利用等个别指标，缺乏对纠错策略及算法设计效率的研究。

因此，能开发一款弥补先前工具不足的测验模式对深入探索和科学评估小学生计算思维特点研究有重要意义。鉴于此，本论文将聚焦于编程操作环境小学生计算思维在线测验工具的开发，在大规模施测的同时系统探索和评估小学高年级学生计算思维的特点，主要研究思路如下：首先，构建符合我国可视化编程教育的计算思维理论模型，在该框架指导下搭建在线测验系统，从CTT角度完成测验的信效度分析；其次，尝试利用结果数据对学生计算概念及其各指标的特点进行分析；最后，利用过程性数据，基于自上而下理论指导完成与尝试次数、纠错策略及再利用有关的计算实践特点分析；利用自下而上数据驱动完成对学生计算实践中算法设计的特点分析。研究者共设计出5个研究，具体如下：

研究一构建了符合我国小学编程教育现状的计算思维测验模型。首先以可视化编程计算思维理论为基础，汲取问题解决全过程及《义务教育课程方案和课程标准》（以下简称课标）要求，构建二级初始模型。而后采用德尔菲法，邀请十五位长期从事编程教育的一线教师和领域内的专家，经过两轮意见征询，结合专家意见及统计指标修订优化模型。最终的三级模型包含计算概念和计算实践两个内容，计算概念包含顺序、循环（重复循环、条件循环）、条件句（if-then和if-else）；计算实践包括调试与纠错（尝试次数、纠错探究行为）、再利用和算法设计效率，各指标的专家意见集中程度和离散趋势符合测量学检验标准。

研究二根据计算思维测验模型编制任务，利用可视化编程系列技术开发了适用于小学3 ~ 6年级学生的在线测验平台，完成工具质量检验。首先，通过多元且适合小学生群体特点的任务及平台设计触发受测者的认知能力：在任务背景和情境呈现上，契合被测群体的认知特征，选择符合年龄阶段的经典任务场景和丰富的动画操作，通过趣味性的交互式过程获取潜在信息。在任务设置上，设计符合认知发展的提示环节，极大程度探知学生最近发展区，融合编程开源思想的同时充分开发该群体学生潜在提升空间。在编程块数量上，限制目标程序编程块的数量反映学生对算法执行时间效率的意识。在动画操作上，设置调整动画速度滑块，降低动画运行速度，更好匹配动画及程序运行的对应关系。其次，工具的开发遵循严格的标准化测验开发流程，包括蓝图设计、专家讨论、预实验、任务修订及正式测验等一系列流程。基于Scratch-gui、Scratch自定义功能及逻辑判断实现测验系统开发和自动评分，保证平台平稳流畅运行。最后，基于学生作答结果，从经典测验理论(Classical Test Theory, CTT)角度对在线测验工具质量进行检验和修订，结果表明该工具有良好的信度、内部结构效度和效标关联效度，题目难度及区分度分布也较为合理。

研究三利用学生作答结果数据，探讨了小学5 ~ 6年级学生计算概念的年级及性别特点。主要结果如下：1）男生的计算概念整体表现高于女生，6年级高于5年级。2）在顺序、重复循环维度，男女表现相同，在条件循环、if-then、if-else维度，男生表现高于女生；除了顺序维度，6年级在其余四个指标表现高于5年级。3）依据学生在五个维度的得分，将学生分为“入门”“熟练”“精通”三类，学生熟练程度越高，在计算概念整体及分维度的表现越好。

研究四利用学生作答过程数据，采用自上而下特征抽取方式，探讨了5 ~ 6年级学生计算实践维度中调试与纠错、再利用的特点。主要结果如下：1）基于理论建构提取了冒险行为、调节动画速度观察、直接观察及改变结构四种纠错探究行为，其中前两种探究行为出现频率较低，后两种探究行为频率较高。2）学生在尝试次数、四种探究行为（冒险行为、调节动画速度观察、直接观察及改变结构）、再利用指标的年级及性别差异不同。3）尝试次数、冒险行为、调节动画速度观察、再利用与计算概念作答结果显著正相关，直接观察与计算概念能力显著负相关。冒险行为、改变结构和再利用是影响计算概念潜在类别分布的主要因素。

研究五利用学生作答过程数据，采用自下而上特征抽取方式，利用社会网络分析方法(SNA)探讨了5 ~ 6年级学生计算实践维度中算法设计效率的特点。主要结果如下：1）SNA的全局指标（密度和中心势）、二元模型（互惠性）及部分三元模型能有效判别学生的算法设计效率水平。2）不同年级、性别及水平学生在各指标上的表现不同，分别体现了不同群体在作答序列长度、动作节点均衡性、动作反复频率、作答系统性等方面的差异。3）动作网络转换图可视化了动作节点及动作转换的频率，依据节点颜色深浅判断，无论哪个群体学生，在利用动画速度观察及开源信息上的意识均较差。

本研究立足本土，优化了计算思维测验模型，并在可视化编程操作环境中搭建了计算思维在线测验平台，为评估我国小学生计算思维领域研究提供高质量的测验工具。基于结果数据的分析，为全方位揭示学生在计算概念具体指标上的能力表现提供依据，为实践教育教学具有重要的诊断和指导价值。基于过程数据综合理论驱动及数据驱动的指标抽取方式，为日后计算实践中策略类及算法设计效率类研究提供新的研究范式。

Computational Thinking refers to solving problems in various software, program, and digitally-driven forms through information processing mediums. In primary school, computational thinking is typically assessed through visual programming, examining students' ability to use fundamental programming concepts to design algorithms for solving programming-related problems, besides the behavioral patterns and answering strategies displayed during this process. The former pertains to computational concepts, while the latter relates to computational practices. With the support of policies from the State Council and the Ministry of Education, some regions in China have introduced visual programming education in primary schools, creating opportunities for the development and improvement of computational thinking assessment tools in compulsory education in China.

Despite increasing attention given to computational thinking in primary education, there remain numerous areas that require exploration and improvement concerning its assessment methods and cognitive characteristics. Historically, the test tools for computational thinking based on visual programming can be broadly applied both in the unplugged and programmed environment for the test type. While unplugged test tools are capable of large-scale reliability and validity tests, they fail to examine students' behavior patterns and response strategies. On the other hand, programmed test tools offer an in-depth analysis of learners' thinking modes and behavioral processes; however, they incur high time and labor costs while relying solely on the code used by students when writing programs to investigate computing concepts. Given these circumstances, it is highly significant to develop a test tool that combines both advantages in order to effectively explore and systematically evaluate the characteristics of computational thinking in primary school students. Regarding students' computational thinking attributes, research primarily focuses on conceptual understanding without delving into their practical application abilities. Studies on computing practices emphasize individual indicators such as abstractions and model reuse but lack the study on error correction strategies and algorithm design efficiency.

Therefore, it is of great significance to develop a test model that explores deeply and evaluates scientifically the characteristics of computational thinking in primary school students to remedy the shortage of the previous test tool. In light of this, this paper focuses on the development of online testing tools for primary school students' computational thinking in a programming operating environment, aiming to systematically explore and evaluate the characteristics of computational thinking among senior primary school students through large-scale testing. The main research thoughts are as follows: Firstly, constructing a theoretical model of computational thinking that aligns with China's visualization programming education, and building an online test system guided by this framework, conducting reliability and validity analysis from the perspective of CTT. Secondly, utilizing the result data to analyze students' characteristics for conceptual understanding and its dimension indicators; Finally, analyze students' computing practices using process data along with top-down theoretical guidance and bottom-up data-driven approaches. Five studies are designed to address the questions above, and they are presented as follows:

Study 1 presents a computational thinking test model that aligns with the current state of programming education patterns for primary schools in China. Firstly, drawing upon visual programming theory, we construct a two-level initial model based on the entire problem-solving process and the requirements of the latest curriculum standards. Subsequently, we employ the Delphi method to engage 15 experienced frontline teachers and experts in programming education for extensive consultation. Through two rounds of revision and optimized by expert opinions and statistical indicators, we finalize a three-level model comprising two first-level dimensions: computational concept and computational practice. The computational concept dimension encompasses sequence, loop (repeated loop, conditional loop), and conditional statement (if-then and if-else), while the computational practice dimension includes bugging and error modeling (number of attempts, behavior pattern of error correction), reuse, and algorithm efficiency. The concentration level of expert opinions as well as the trend of dispersion for each indicator are consistent with statistical testing indices.

Several tasks were designed based on the computational thinking test model, and an online test platform using visual programming series technology suitable for primary school students in grades 3 to 6 was developed, to conduct a thorough quality inspection of the tool, in Study 2. First, triggering participants' cognitive abilities through diversified task patterns designed suitable for primary school students' characteristics: In terms of task background and situation presentation, classic task scenes and incorporating rich animation operations that match the cognitive characteristics of the participants at their age stage were selected while obtaining potential information through engaging interactions. Regarding task setting, designing prompt links aligned with cognitive development to probe the zone of proximal development of students extensively while fully developing their potential improvement space by integrating open-source programming concepts. Limiting the number of programming blocks required to complete tasks reflects students' awareness regarding algorithm execution time efficiency. For the animation operation, the slider for adjusting the animation model is set, which can reduce the animation running speed, to better match the corresponding relationship between the animation and the program running. Secondly, tool development followed standardized processes including blueprint design, expert discussions, pre-experiments, task revisions, and formal testing. By utilizing Scratch-gui along with custom functions and logic judgments from Scratch software, we successfully developed a test system that enables automatic scoring while ensuring the smooth operation of the platform. Finally, based on student responses obtained through this online tool evaluation from a CTT perspective; results indicate good reliability and internal structure validity as well as criterion association validity; furthermore, demonstrating reasonable difficulty distribution and differentiation levels.

Based on the data obtained from students' response results, Study 3 investigates the grade and gender characteristics of primary school students' understanding of computing concepts in grades 5 to 6. The key findings are as follows: 1) Boys outperform girls significantly in overall computing concept performance, with grade 6 scoring significantly higher than grade 5; 2) There is no significant gender difference observed in the dimensions of sequential and repetitive cycles. However, boys score significantly higher than girls in the conditional cycle dimensions of if-then and if-else. Additionally, except for the order dimension, grade 6 scores were significantly higher than grade 5 across all other indicators; 3) Based on students' scores across five dimensions, students were categorized into three groups: "novice," "proficient," and "advanced." The more proficient students, the better performance on the computing concept and its dimensions.

Study 4 utilizes students' answering process data and employs the top-down feature extraction method to investigate the characteristics of computing practice among fifth and sixth-grade students. The main findings are: 1) Four procedural inquiry behaviors, namely risk-taking behavior, animation observation, direct observation, and structure change, were extracted based on theoretical construction. Among these behaviors, the first two behaviors had a lower frequency while the latter two had a higher frequency; 2) There were significant grade and gender differences in the number of attempts, four inquiry behaviors (risk-taking behavior, animation observation, direct observation, and structure change), and reuse indicators; 3) The number of attempts, risk-taking behavior, animation observation, and reuse showed significant positive correlations with computational concept response results; whereas direct observation exhibited a significant negative correlation with computational concept ability. Risk-taking behavior, structure change, and reuse emerged as the key factors influencing the potential category distribution of computing concepts.

In Study 5, the bottom-up feature extraction method along with social network analysis (SNA) was employed using students' answering process data to explore the characteristics of answering efficiency among fifth and sixth-grade students. The main findings are as follows: 1) SNA's global indicators (density and centrality degree), binary model(reciprocity), and part of ternary model can effectively assess students' answering efficiency; 2) Students from different grades, genders, and levels demonstrated varying performance across various indicators which respectively reflected differences in response sequence length, balance of action nodes, frequency of action repetition, and system of action; 3) The action network transformation diagram visualizes the action nodes and the frequency of their transformations. Regardless of the student group, it is evident from the varying color depths of the nodes that there is a lack of awareness of utilizing animation observation and open-source information.

This research is based on the local context, optimizing the computational thinking assessment model and establishing a computational thinking online assessment platform within a visual programming environment, providing a high-quality evaluation tool for researching primary school students' computational thinking abilities in China. By employing process data synthesis theory-driven and data-driven indicators extraction methods, this research introduces a novel paradigm for studying strategy formulation and algorithm design efficiency in future computing practices.