應用MIMIC模式評估方法以檢定試題差異性之研究

Abstract

本文主要探討以MIMIC模式(multiple-indicators, multiple-causes model)檢查題組型的試題群DIF(Differential Item Functioning)時，在多種不同的模擬實驗設計下，檢測的錯誤率、正確率及樣本數間的關係。MIMIC模式在1975年由Jöreskog與Goldberger提出，該模型可用來檢測依變數、獨立變數及潛在變數的間接或直接關係。另一延伸的應用可將MIMIC模式用來進行不同樣本間的測驗(measurement)或結構(structural)比較分析。本研究即利用強韌性卡方差異檢定(robust chi-square difference test)於評估MIMIC模式的適配度時，此檢定方法之整體品質。模擬研究結果顯示，當使用MIMIC模式進行試題差異檢定時，強韌性卡方差異檢定法之檢測正確率會隨著試題數、試題難度差異量、試題差異比例增加而上升。 The main aims of this paper are to study effects of inaccuracy, accuracy, and sample sizes when the multiple-indicators, multiple-causes (MIMIC) model was used for performing difference item functioning (DIF) in some designed item groups of simulation study. The MTMTC model, proposed by Jöreskog and Goldberger (1975), can be utilized to evaluate direct/indirect relationships among dependent, independent, and latent variables as well as to compare measurement or structural difference between samples. To achieve the evaluation purpose, robust chi-square difference test was used to address overall qualities of the goodness-of-fit index of MIMIC models. Results by the simulation study show when MIMIC models are used to detect DIF between groups, accuracy of DIF identification can be increased as the test length, DIF magnitudes, percentages of DIF items in the test are increased.