| 李强.学生成长百分等级(SGP)模型的实践改进研究——基于“天花板/地板”效应和群体前测影响[J].唐山学院学报,2026,39(3):101-108 |
| 学生成长百分等级(SGP)模型的实践改进研究——基于“天花板/地板”效应和群体前测影响 |
| Research on Practical Improvement of the Student Growth Percentile (SGP) Model: Based on “Ceiling/Floor” Effects and Group Pretest Influences |
| 投稿时间：2025-11-03 |
| DOI：10.16160/j.cnki.tsxyxb.2026.03.014 |
| 中文关键词: 学业评价 增值评价 学生成长百分等级模型 R语言 |
| 英文关键词: academic assessment value-added assessment student growth percentile model R language |
| 基金项目:唐山市教育科学研究“十四五”规划2024年度课题(2024LX044) |
|
| 摘要点击次数: 368 |
| 全文下载次数: 224 |
| 中文摘要: |
| 文章针对学生成长百分等级(SGP)模型在教育增值评价实践中易出现评价失真及系统性偏差的现实问题开展实证研究。基于R语言自建SGP模型,以三万余名学生两次考试追踪数据为基础展开系统分析,厘清模型应用中的两大关键问题:一是极端高分、低分区间受"天花板/地板"效应影响,造成个体增值评价结果失真;二是群体增值评价易受前测整体学业水平干扰,进而产生系统性偏差。针对上述问题,提出校正优化方案:在数据密集区间采用三次B样条函数进行拟合,对极端区间作强制水平化处理,有效化解"天花板/地板"效应引发的评价失真问题;通过多水平模型实证发现,学校前测均值对学校整体增值的影响占比达50%~90%,据此提出分组评价策略,以减小群体前测水平带来的系统性偏差。 |
| 英文摘要: |
| This paper conducts empirical research on the practical issues of evaluation distortion and systematic bias in the application of the student growth percentile (SGP) model in educational value-added assessment. Based on self-built SGP models using R language and systematic analysis of tracking data from over 30 000 students across two examinations, two key issues in the model’s application are clarified: First, extreme high and low score intervals are affected by "ceiling/floor" effects, leading to distortion in individual value-added evaluation results; Second, group value-added evaluations are easily influenced by the overall academic level of the pretest, resulting in systematic bias. To address the above issues, this paper proposes optimized correction solutions: Using cubic B-spline function fitting in dense data intervals, apply forced horizontalization in extreme intervals to effectively mitigate evaluation distortions caused by "ceiling/floor" effects; Through empirical analysis with multilevel models, it is found that the school pretest mean accounts for 50%~90% of the impact on the school’s overall value-added. Accordingly, this paper proposes a grouped evaluation strategy to alleviate systematic bias caused by group pretest levels. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|