Toward a Unified Theory of Reliability for Social Measurement.
Two major models have dominated the assessment of reliability in social measurement: internal consistency and stability over time. With internal consistency models, there is no repeated measurement of the same items or instruments over time; with stability models, idiosyncratic change over time (or certain classes of change) is treated as error even though it is "true" change. These two models are combined into a single and more general "time-item" model that combines their advantages and tempers their weaknesses. Using an analysis of variance approach, the model results in a new coefficient, delta, that can be calculated as a simple function of mean squares from an ANOVA table. The time-item model also leads to a more complete estimation of change score reliability, also computable from an ANOVA table. Empirical examples from two different longitudinal studies show that delta provides a more stable and more realistic estimate of reliability in over-time studies. (For the American Sociological Review.)