On the Inappropriateness of Using Items to Calculate Total Scale Score Reliability via Coefficient Alpha for Multidimensional Scales
On the Inappropriateness of Using Items to Calculate Total Scale Score Reliability via Coefficient Alpha for Multidimensional Scales
复制标题
DOI：
10.1027/1015-5759/a000181
复制
发表时间：
2014-01-01
影响因子：
2.5
通讯作者：
Gignac, Gilles E.
中科院分区：
文献类型：
--
作者：
Gignac, Gilles E.
研究者可以隐式选择基于项目间相关矩阵（项目单位级）或子量表间相关矩阵（子量表单位级）计算多维量表总得分的内部一致性信度（系数a）。研究表明，项目单元级和子量表单元级的信度估计在实际应用中往往存在较大的差异。具体来说，项目单位水平的信度估计往往大于相应的子量表单位水平估计。如果研究人员在项目单元水平上计算总量表得分的信度，那么当底层模型是多维的时候，应该采用基于模型的内部一致性信度估计方法（即omega分层法）。如果欧米茄层次不能应用任何特殊原因，建议总量表得分可靠性计算在分析的子量表单位水平，而不是项目单位水平。
Researchers have the implicit option of calculating internal consistency reliability (coefficient a) for total scale scores derived from multidimensional inventories based on either the inter-item correlation matrix (item unit- level) or the inter-subscale correlation matrix (subscale unit-level). It is demonstrated that item unit- level and subscale unit- level reliability estimates often diverge substantially in practice. Specifically, the item unit- level reliability estimation is often larger than the corresponding subscale unit- level estimate. It is recommended that if researchers calculate total scale score reliability at the item unit- level, then a model- based approach to the estimation of internal consistency reliability (i.e., omega hierarchical) should be applied, when the underlying model is multidimensional. If omega hierarchical cannot be applied for any particular reason, it is recommended that total scale score reliabilities be calculated at the subscale unit- level of analysis, not the item unit-level.