School of Education, University of Delaware.
Perelman School of Medicine, Department of Medicine, University of Pennsylvania.
Psychol Assess. 2016 Feb;28(2):125-33. doi: 10.1037/pas0000146. Epub 2015 Jun 1.
The Mini-Mental State Examination (MMSE) is a 30-item, dichotomously scored test of general cognition. A number of benefits could be gained by modeling the MMSE in an item response theory (IRT) framework, as opposed to the currently used classical additive approach. However, the test, which is built from groups of items related to separate cognitive subdomains, may violate a key assumption of IRT: local item independence. This study aimed to identify the most appropriate measurement model for the MMSE: a unidimensional IRT model, a testlet response theory model, or a bifactor model. Local dependence analysis using nationally representative data showed a meaningful violation of the local item independence assumption, indicating multidimensionality. In addition, the testlet and bifactor models displayed superior fit indices over a unidimensional IRT model. Statistical comparisons showed that the bifactor model fit MMSE respondent data significantly better than the other models considered. These results suggest that application of a traditional unidimensional IRT model is inappropriate in this context. Instead, a bifactor model is suggested for future modeling of MMSE data as it more accurately represents the multidimensional nature of the scale. (PsycINFO Database Record
简易精神状态检查（MMSE）是一个包含 30 个项目的二分测试，用于评估一般认知能力。与目前使用的经典加性方法相比，在项目反应理论（IRT）框架中对 MMSE 进行建模可以带来许多好处。然而，该测试由与不同认知子领域相关的项目组构成，这可能违反了 IRT 的一个关键假设：局部项目独立性。本研究旨在确定 MMSE 的最合适测量模型：单维 IRT 模型、测验分组反应理论模型还是双因子模型。使用全国代表性数据进行的局部依赖分析显示，局部项目独立性假设存在有意义的违反，表明存在多维性。此外，测验分组和双因子模型的拟合指标均优于单维 IRT 模型。统计比较表明，双因子模型比其他考虑的模型更能拟合 MMSE 应答者数据。这些结果表明，在这种情况下，应用传统的单维 IRT 模型是不合适的。相反，建议在未来对 MMSE 数据进行建模时使用双因子模型，因为它更能准确地反映量表的多维性质。
