Accurate Ability Measurement in Constructed-Response Tests through the Application of Automated Scoring Techniques and Item Response Theory

There is a growing need to assess students’ higher-order abilities, such as expressive skills and logical thinking.Constructed-response tests with short-answer and essay items have become widely adopted to meet this demand. Apersistent limitation of these tests is the substantial manual grading, making them both labor-intensive and costly. Itemresponse theory (IRT) offers a promising remedy by enabling ability estimation from incomplete score data, in whichhuman raters grade only a subset of answers across multiple test items. However, the accuracy of ability estimationdeclines as the proportion of missing scores increases. Although data-augmentation techniques for imputing missingscores have been explored to mitigate this issue, they often struggle to remain accurate under sparse or heterogeneousdata. To address these challenges, this study proposes a method that imputes missing scores by leveraging automatedscoring technologies to support accurate IRT-based ability estimation. The proposed approach attains high estimationaccuracy while markedly reducing the manual grading workload.