多维IRT与单维IRT在多维量表中应用的差异

The Difference between Unidimensional IRT and Multidimensional IRT in the Application of Multidimensional Scale

目的探讨单维条目反应理论与多维条目反应理论在多维量表分析中的差异,并从中找出较优的分析方法。方法用单维分部评分模型(PCM)和多维分部评分模型分别对世界卫生组织生存质量研究小组提供的来自世界20个研究中心的WHOQOL-OLD量表数据进行条目和量表结构的分析。结果 "感觉能力"领域中的条目OLD_10"您的感觉功能的问题影响您和他人交往的能力吗?"同时不拟合两种模型,单维IRT得到Infit和Outfit均方拟合指数、6个维度的信度系数及潜在特质间的相关系数均低于多维IRT。结论多维IRT更适合于条目数较少的多维量表的分析和评价。

Objective To explore the difference between unidimensional IRT and multidimensional IRT in the application of scale which includes several subscales and find out the better method.Methods The data of WHOQOL-OLD came from the field study of 20 national study centers of WHOQOL Group which was conducted.The Unidimensional Partial Credit Model and Multidimensional Partial Credit Model were used to analyze the property of items and construct of scale.Results Item OLD_10（Problems with sensory functioning affect ability to interact） in the ＂Sensory Abilities＂ domain showed poorer fit to two models.And the Infit and Outfit Mnsq,reliability coefficients and correlation coefficients of latent ability of six domains from unidimensional IRT analysis all were lower than that of multidimensional IRT analysis.Conclusion Multidimensional IRT is more appropriate than unidimensional IRT for the analysis and evaluation of multidimensional and short scale.