Scholarlink

<p>Recent research has expanded beyond investigating differential item functioning (DIF) in response score data to examine DIF in response times. However, studies on response time-based DIF have predominantly focused on uniform DIF, with non-uniform DIF receiving limited attention. This study applies the log-normal response time model to directly model non-uniform DIF in response times. Linear regression is then utilized to identify items exhibiting DIF in response times. A simulation study evaluated the impact of examinee sample sizes, focal-to-reference group ratios, the number of DIF items in a test, and DIF effect sizes on the method’s performance. Results indicate that the proposed approach maintains superior performance, demonstrating robust detection power even under small sample sizes and small-magnitude DIF effects while effectively controlling Type I error rates within acceptable limits. An empirical study using the PISA 2018 dataset assessed the presence of response time DIF between Chinese and American students across five dimensions of science items, confirming the method’s practical utility and potential for real-world testing applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detecting non-uniform differential item functioning in response time through linear regression

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月26日

查看全文
  • Zhicheng Liu,
  • Chunying Qin,
  • Zhaosheng Luo,
  • Xiaofeng Yu

摘要

Recent research has expanded beyond investigating differential item functioning (DIF) in response score data to examine DIF in response times. However, studies on response time-based DIF have predominantly focused on uniform DIF, with non-uniform DIF receiving limited attention. This study applies the log-normal response time model to directly model non-uniform DIF in response times. Linear regression is then utilized to identify items exhibiting DIF in response times. A simulation study evaluated the impact of examinee sample sizes, focal-to-reference group ratios, the number of DIF items in a test, and DIF effect sizes on the method’s performance. Results indicate that the proposed approach maintains superior performance, demonstrating robust detection power even under small sample sizes and small-magnitude DIF effects while effectively controlling Type I error rates within acceptable limits. An empirical study using the PISA 2018 dataset assessed the presence of response time DIF between Chinese and American students across five dimensions of science items, confirming the method’s practical utility and potential for real-world testing applications.