This paper considers a test score model in which the true score and the measurement error are autocorrelated. After some preliminary remarks on corrections for attenuation, the paper focuses on the three-wave ease (t = 1, 2, 3), demonstrating identifiability, showing an estimation algorithm, and providing a numerical illustration. [ABSTRACT FROM AUTHOR]
The article provides an overview of the theoretical problems posed by multiple-item scales and also guide through few cautions that could be taken while applying procedures for determining the same. George W. Bohrnstedt provided a quick method for determining the reliability and validity of multiple-item scales. The reliabilities are easily found using the path analysis algorithm since there are exactly three known and three unknown correlations. With more than three items, the knowns will be more numerous than the unknowns, that is, over-determination. The authors consider reliability as the agreement between several efforts to measure the same trait through maximally similar methods, and convergent validity as the agreement when the methods are maximally different. The concept of discriminant validity involves the introduction of a second trait factor, which itself may be fallible with possibly nonindependent errors of measurement, both among the items measuring the same trait and/or between items measuring different traits.