Generalizability theory: a unified approach to assessing the dependability (reliability) of measurements in the health sciences

Abstract

The reliability of health promotion program evaluation measures, behavioral and attitudinal measures, and clinical measures is a concern to many health educators. Classical reliability coefficients, such as Cronbach's alpha, apply to narrowly defined, prespecified measurement situations. Classical theory does not provide adequate reliability assessments for criterion-referenced measures, for measurement situations having multiple sources of error, or for aggregate-level variables. Generalizability theory can be used to assess the reliability of measures in these situations that are not adequately modeled by Classical theory. Additionally, Generalizability theory affords a broader view and a deeper understanding of the dependability of measurements and the role of different sources of error in the variability of measures.