Kelly P A, Kallen M A, Suárez-Almazor M E
Michael E. DeBakey VA Medical Center, and Department of Medicine, Baylor College of Medicine, Houston, TX 77030, USA.
J Clin Epidemiol. 2007 May;60(5):440-7. doi: 10.1016/j.jclinepi.2006.08.005. Epub 2006 Dec 11.
The Multidimensional Health Locus of Control (MHLC) scales are widely used to measure beliefs about determinants of persons' health. We evaluated the scales over the largest-ever disease-specific sample of subjects using a combined-method psychometric approach.
We performed a secondary analysis of data from 1,206 subjects from three osteoarthritis studies, using Rasch analysis and confirmatory factor analysis simultaneously. Differential item functioning (DIF) by gender and data source, scale dimensionality, and item fit were examined. The Rasch model fit the data if Rasch residual principal components analysis (PCA) corroborated three distinct dimensions and item fit statistics fell between 0.80 and 1.20. The confirmatory factor (CFA) model fit the data if factor loadings exceeded 0.50 for all items.
DIF by gender or data source was not materially evident for any items. PCA supported existence of three dimensions in the data. Both Rasch and CFA models fit the data for 16 items; two items were detected as misperforming. When these items were removed, fit of both models improved.
Results of this large-sample evaluation of the MHLC scales corroborated earlier findings that removal of certain items improves the scales. The combined Rasch-CFA approach provided better insight to scale performance problems than either method alone provided.
多维健康控制点（MHLC）量表被广泛用于衡量人们对健康决定因素的信念。我们使用综合方法的心理测量方法，在有史以来最大的特定疾病样本受试者中对这些量表进行了评估。
我们对来自三项骨关节炎研究的1206名受试者的数据进行了二次分析，同时使用了拉施分析和验证性因素分析。检查了按性别和数据源划分的项目功能差异（DIF）、量表维度和项目拟合情况。如果拉施残差主成分分析（PCA）证实存在三个不同维度且项目拟合统计量在0.80至1.20之间，则拉施模型适合该数据。如果所有项目的因子载荷超过0.50，则验证性因子（CFA）模型适合该数据。
任何项目按性别或数据源划分均未明显存在项目功能差异。PCA支持数据中存在三个维度。拉施模型和CFA模型对16个项目的数据均拟合良好；检测到两个项目表现不佳。去除这些项目后，两个模型的拟合度均有所提高。
本次对MHLC量表的大样本评估结果证实了早期的研究发现，即去除某些项目可改善量表。与单独使用任何一种方法相比，拉施分析与CFA相结合的方法能更好地洞察量表性能问题。
