Paper Summary
Share...

Direct link:

Validation of Automated Scoring of Science Assessments

Fri, April 17, 4:05 to 5:35pm, Marriott, Floor: Fourth Level, Belmont

Abstract

Science researchers and educators have called for the assessment community to apply constructed-response items that measure in-depth learning and understanding. However, the use of constructed-response items has been limited by the cost and complexity in its scoring. Automated scoring, when applied appropriately, has great potential to mitigate the scoring limitations for constructed-response items and therefore enhance scaled-up implementation of this item type. This study explored the use of c-rater-ML, an automated scoring engine developed by the Educational Testing Service, in scoring the content of eight complex science inquiry items. In general automated scoring showed satisfactory agreement with human scoring. In addition, examinations of the scoring across subgroups confirmed the consistency between automated and human scoring.

Authors