Reliability

Separating signal from noise is the primary challenge of measurement and is the fundamental goal of all approaches to reliability theory. Reliability is the ability to generalize about individual differences across alternative sources of variation. Generalizations within a domain of items use internal consistency estimates. This chapter examines reliability to estimate the true score given an observed score, and to establish confidence intervals around this estimate based upon the standard error of the observed scores. The concept that observed covariances reflect true covariances is the basis for structural equation modeling, in which relationships between observed scores are expressed in terms of relationships between latent scores and the reliability of the measurement of the latent variables. Reliability estimates can be found based upon variations in the overall test, variations over time, variation over items in a test, and variability associated with who is giving the test.