基于共建代码仓库的编程学习参与度与学业成效的实证研究

An Empirical Study of Programming Learning Engagement and Academic Achievement Based on a Co-Built Code Repository

摘要: 针对程序设计课程中长期存在的看得懂却写不出、代码阅读能力薄弱与成绩分布偏低等问题，本文报告一项以全班共建代码仓库为载体的教学改革实践，并对学生的学习参与度与学业成效之间的关系进行实证分析。改革的核心在于让学生主动参与仓库的构建：每名学生在以学号命名的独立目录中提交代码与学习笔记，仓库通过持续集成工作流校验目录权限，教学范例库则覆盖从基础语法到区块链密码学的完整示例。本文以本校《Go语言程序设计》课程同一学期51名学生为对象，将其仓库参与度指标(代码行数、文件数、提交次数、笔记字数)与期末卷面成绩进行配对分析。结果显示，四项参与度指标与期末卷面成绩均呈正相关(代码行数r = 0.39，p < 0.01；文件数r = 0.37，p < 0.01；Spearman秩相关结论一致)；以文件数划分的高参与组(n = 16)期末卷面均分为75.88，显著高于低参与组(n = 35)的66.77 (Welch t = 2.64，p = 0.012，Cohen’s d = 0.73，中等偏大效应)；分组阈值的敏感性检验显示结论不依赖于特定切点。两届课程的趋势比较显示整体成绩明显上移，不及格比例由37.5%下降至17.6%，降幅接近50%，及格率由62.5%提升至82.4%。研究表明，“参与即贡献”的协作机制与编程学习成效之间存在稳定的正向关联。基于上述发现，本文在讨论部分进一步提出“同伴语料锚定的AI辅助”这一面向未来的教学设计构想，为AI时代的编程教育提供新的研究方向。作为观察性研究，参与度与成绩之间的因果关系仍需通过引入对照组的准实验设计加以检验。

Abstract: This paper reports a teaching reform practice in a programming course that uses a class-wide collaboratively constructed code repository to address long-standing problems, including students’ ability to understand code but difficulty in writing it independently, weak code-reading competence, and a generally low distribution of academic performance. It further conducts an empirical analysis of the relationship between students’ learning engagement and academic achievement. The core of the reform is to engage students as active contributors to the construction of the repository. Each student submits code and learning notes to an individual directory named after their student ID; the repository uses a continuous-integration workflow to verify directory permissions; and the instructional example library provides complete examples ranging from basic syntax to blockchain cryptography. Using data from 51 students enrolled in the “Go Language Programming” course at the university in the same semester, this study pairs repository-engagement indicators, including lines of code, number of files, number of commits, and word count of notes, with final-examination scores for analysis. The results show that all four engagement indicators are positively correlated with final-examination scores, including lines of code, r = 0.39, p < 0.01, and number of files, r = 0.37, p < 0.01; the Spearman rank-correlation results are consistent with these findings. The high-engagement group, divided by number of files (n = 16) achieved an average final-examination score of 75.88, significantly higher than the low-engagement group (n = 35), whose average score was 66.77 (Welch’s t = 2.64, p = 0.012, Cohen’s d = 0.73), indicating a moderate-to-large effect. A sensitivity test of the grouping threshold further shows that the conclusion does not depend on a specific cutoff point. A trend comparison across two cohorts shows a clear upward shift in overall academic performance: the failure rate decreased from 37.5% to 17.6%, a reduction of nearly 50%, while the pass rate increased from 62.5% to 82.4%. The findings indicate a stable positive association between the collaborative mechanism of “participation as contribution” and programming learning outcomes. Based on these findings, this paper further proposes, in the discussion section, the future-oriented teaching-design concept of “peer-corpus-anchored AI assistance,” offering a new research direction for programming education in the AI era. As an observational study, the causal relationship between engagement and achievement still needs to be examined through a quasi-experimental design with a control group.