目的:本研究利用評分者差異函數(differential rater functioning, DRF)估計評分者的評分差異,偵測評分者間的偏差。方法:受試者18 名選手,採5 位專家評定13 項標槍投擲技術。研究依據Rasch 模式,以Winsteps 程式估計評分者和項目的測量參數,以及DRF 探查評分者在項目的差異。結果:一、受試者和項目信度良好,但都有模式適合度不良者。二、評分者在項目5 的評分嚴謹度有顯著差異。三、A 評分者在項目5 和項目8 的評分較為寬鬆。四、配對比較評分者在項目的差異函數考驗,皆為A 評分者與其他四位評分者的評分嚴謹度有顯著差異。結論:評分者差異函數可以偵測主觀評分之評分者偏差。
Purpose: the study was using differential rater functioning (DRF) to estimate the rating difference and detect rating bias between raters. Methods: The subjects were 18 javelin players. Using five raters assessed thirteen throwing skills in javelin. The research was based on Rasch model and used Winsteps to estimate measurement parameters of item and rater, and detected the rating bias within items between raters. Results: (1) The reliabilities of items and subjects were good, but both of them had some Misfit. (2) The raters' rating severity in the fifth item had significant differences. (3) The A rater had leniency rating in 5th and 8th items. (4) Comparison with paired raters in the differential item function test, we found out that the A rater's rating severity had significant difference with other four raters' rating. Conclusion: DRF can detect subjective rating bias between raters.