Improving Graphical Response Scoring Accuracy in TIMSS 2023 with AI-Human Expert Integration

• In Event: Models for International Assessments

Abstract

Students sometimes respond to graphical response items in unusual ways where it may be unclear what score the response should receive, leading to inconsistent scoring between human raters. This presentation proposes an AI-based validation method to identify and select these “borderline” responses for expert validation and scoring in TIMSS.

Authors

• Lillian Tyack, Boston College

• Matthias von Davier, BOSTON COLLEGE