Scholarlink

In large-scale assessments, examinees commonly do not provide answers to all items administered. Such (unplanned) missing responses can occur either due to item omissions or due to not reaching the end of the assessment. If not appropriately accounted for, item and person parameter estimates as well as group statistics can be biased and, for example, lead to distorted performance rankings. Considering additional information on test-taking behavior, such as response times, may critically improve the understanding of the occurrence of missing responses and assist a more nuanced handling. After a short review of approaches for handling missing responses that employ information solely from item responses, we review four current model-based approaches that incorporate response time information. Each approach is focused on different missingness mechanisms. In an empirical example, using data from PISA 2018, the approaches are illustrated and compared in terms of the insight on test-taking behavior they provide, as well as their assumptions on the occurrence of missing responses.

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

Using Response Times for Improving the Understanding and Handling of Missing Responses in Low-Stakes Assessments

  • 章节

  • 发表日期：

    2025年

查看全文
  • Esther Ulitzsch,
  • Tobias Alfers,
  • Jing Lu,
  • Chun Wang,
  • Steffi Pohl

摘要

In large-scale assessments, examinees commonly do not provide answers to all items administered. Such (unplanned) missing responses can occur either due to item omissions or due to not reaching the end of the assessment. If not appropriately accounted for, item and person parameter estimates as well as group statistics can be biased and, for example, lead to distorted performance rankings. Considering additional information on test-taking behavior, such as response times, may critically improve the understanding of the occurrence of missing responses and assist a more nuanced handling. After a short review of approaches for handling missing responses that employ information solely from item responses, we review four current model-based approaches that incorporate response time information. Each approach is focused on different missingness mechanisms. In an empirical example, using data from PISA 2018, the approaches are illustrated and compared in terms of the insight on test-taking behavior they provide, as well as their assumptions on the occurrence of missing responses.