Investigating Repeater Effects on Small Sample Equating: Include or Exclude?. Issue 1 (2nd January 2020)

• Record Type: Journal Article
• Authors: Diao, Hongyu; Keller, Lisa
• Is Part Of: Applied measurement in education. Volume 33:Issue 1(2020)
• Page Start: 54
• Page End: 66
• Publication Date: 2020-01-02
• Subjects: Educational tests and measurements -- United States -- Periodicals; Educational tests and measurements -- Periodicals; Education -- Standards -- Periodicals; 371.26
• DOI: 10.1080/08957347.2019.1674302
• Languages: English
• ISSNs: 0895-7347

Abstract

ABSTRACT: Examinees who attempt the same test multiple times are often referred to as "repeaters." Previous studies suggested that repeaters should be excluded from the total sample before equating because repeater groups are distinguishable from non-repeater groups. In addition, repeaters might memorize anchor items, causing item drift under a non-equivalent anchor test (NEAT) design. However, under small sample equating conditions, removing repeaters might lead to smaller sample size, which increases sampling errors. Therefore, three solutions were investigated in the current study: 1) excluding repeaters, 2) excluding drifted anchor items, and 3) applying Rasch true score equating to maintain the population invariance if repeaters were removed. The results suggested excluding repeaters if the anchor were exposed to them. Circle arc equating can be applied if it is impossible to exclude all drifted anchor items. Applying Rasch true did outperform solutions.