Similarity of the cut score in test sets with different item amounts using the modified Angoff, modified Ebel, and Hofstee standard-setting methods for the Korean Medical Licensing Examination

• Published in: Journal of educational evaluation for health professions Vol. 17; pp. 28 - 10
• Main Authors: Park, Janghee, Yim, Mi Kyoung, Kim, Na Jin, Ahn, Duck Sun, Kim, Young-Min
• Format: Journal Article
• Language: English
• Published: Korea (South) Korea Health Personnel Licensing Examination Institute 2020; 한국보건의료인국가시험원
• Subjects: educational measurement; medical education; medical licensure; reproducibility of results; republic of korea; 교육학; Medical licensing examination; Standard setting; Ebel; Modified Angoff; Hofstee
• ISSN: 1975-5937; 1975-5937
• DOI: 10.3352/jeehp.2020.17.28

Purpose: The Korea Medical Licensing Exam (KMLE) typically contains a large number of items. The purpose of this study was to investigate whether there is a difference in the cut score between evaluating all items of the exam and evaluating only some items when conducting standard-setting.Methods: We divided the item sets that appeared on 3 recent KMLEs for the past 3 years into 4 subsets of each year of 25% each based on their item content categories, discrimination index, and difficulty index. The entire panel of 15 members assessed all the items (360 items, 100%) of the year 2017. In split-half set 1, each item set contained 184 (51%) items of year 2018 and each set from split-half set 2 contained 182 (51%) items of the year 2019 using the same method. We used the modified Angoff, modified Ebel, and Hofstee methods in the standard-setting process.Results: Less than a 1% cut score difference was observed when the same method was used to stratify item subsets containing 25%, 51%, or 100% of the entire set. When rating fewer items, higher rater reliability was observed.Conclusion: When the entire item set was divided into equivalent subsets, assessing the exam using a portion of the item set (90 out of 360 items) yielded similar cut scores to those derived using the entire item set. There was a higher correlation between panelists’ individual assessments and the overall assessments.