Rouquette Alexandra, Hardouin Jean-Benoit, Coste Joel
Alexandra Rouquette, Hotel-Dieu Hospital, Biostatistics and Epidemiology Department, 1 place du parvis Notre-Dame, 75181 Paris cedex 04, France,
J Appl Meas. 2016;17(3):312-334.
To determine the conditions in which the estimation of a difference between groups for a construct evaluated using a composite measurement scale is biased if the presence of Differential Item Functioning (DIF) is not taken into account.
Datasets were generated using the Partial Credit Model to simulate 642 realistic scenarios. The effect of seven factors on the bias on the estimated difference between groups was evaluated using ANOVA: sample size, true difference between groups, number of items in the scale, proportion of items showing DIF, DIF-size for these items, position of these items location parameters along the latent trait, and uniform/non-uniform DIF.
For uniform DIF, only the DIF-size and the proportion of items showing DIF (and their interaction term) had meaningful effects. The effect of non-uniform DIF was negligible.
The measurement bias resulting from DIF was quantified in various realistic conditions of composite measurement scale use.
确定在使用综合测量量表评估一个结构时，如果不考虑项目功能差异（DIF）的存在，那么对组间差异估计产生偏差的条件。
使用部分计分模型生成数据集，以模拟642种现实情景。使用方差分析评估七个因素对组间估计差异偏差的影响：样本量、组间真实差异、量表中的项目数量、显示DIF的项目比例、这些项目的DIF大小、这些项目位置参数沿潜在特质的位置，以及一致/非一致DIF。
对于一致DIF，只有DIF大小和显示DIF的项目比例（及其交互项）有显著影响。非一致DIF的影响可忽略不计。
在综合测量量表使用的各种现实条件下，对由DIF导致的测量偏差进行了量化。
