Using Confirmatory Factor Analysis of Correlated Uniquenesses to Estimate Method Variance in Multitrait-Multimethod Matrices

Conway used a special case and several supporting examples to argue that the average proportion of method variance in a multitrait-multimethod (MTMM) matrix can be estimated by averaging the correlated uniquenesses (cus). This article builds on Conway in two ways. First, it generalizes the logic, showing that the average cu is a lower bound estimate of the correct value for the average proportion of method variance. The generalization also allows examination of the method's applicability to a much broader range of MTMM matrices. Second, this article presents a new method based on confirmatory factor analysis of the covariance matrix of cus for measuring method variance. The new method provides more precise and unbiased estimates of the average proportion of method variance associated with each measurement method, and is the first method that uses the correlated uniquenesses model to estimate the proportion of method variance in individual measures.