不同資料遺失樣態對於差異試題功能偵測效果之影響

Abstract

本研究旨在探討不同遺失樣態下偵測差異試題功能（DIF）的影響，其中的遺失樣態是指不同的遺失機制與遺失比率。因此，以模擬研究的方式來探究四種遺失機制（MCAR遺失與三種不同形式的MAR遺失）與三種遺失比率（0%、10%、30 %）下，並操弄三種DIF試題比率（0%、10%、20%）與三種DIF程度（0、0.5、0.8），進一步討論兩種遺失值處理方式（有無進行單一插補）與四種DIF偵測方法（有無加入純化程序的Mantel-Haenszel statistic與Lord’s chi-square）對於DIF偵測效果（型一錯誤率與正確偵測率）的影響。 研究結果顯示，遺失樣態對於DIF偵測效果有影響，但僅在以MH法進行DIF分析的情況下。經單一插補處理遺失值後，多數DIF試題的正確偵測率與型一誤判率會增加。無論是以MH法或Lord法作為DIF偵測方法，加入純化程序都能有效改善DIF偵測效果。

Differential item functioning (DIF) is an area of continuous interest within the community of measurement researchers. Recently, there is some interest in the detection of DIF items when missing data are present in the test. Under such circumstances, different treatments on the missing data may have different effects on the detection of DIF items. This article describes the results of a simulation study to investigate the impact of missing data pattern on the detection of uniformly DIF items. In the study, missing data pattern is defined by means of various missing mechanisms and missing rates. The purpose of this study is to investigate how two missing data treatments (utilizing single imputation or not) interact with four methods of DIF detection (Mantel-Haenszel statistic and Lord’s chi-square test with and without purification) under four missing mechanisms (MCAR and three versions of MAR) and three missing rates (0%, 10%, 30%) with three DIF magnitude (0, 0.5, 0.8) by means of examining the type I error rates as well as the statistical power of DIF detection. Results show that missing data pattern has impact on the detection of DIF, but only with respect to MH. After missing data treatment by SI, most type I error rates and statistical power increase. With respect to both MH and Lord’s approaches, purification procedure could improve on their DIF detection performances.