ABSTRACT
The Generalizability approach is a comprehensive model that can conceptually accommodate the coexistence of numerous reliability estimates. This approach departs from the conventional methods in two respects. Statistically, it approaches the task of estimating reliabilities through an alternative technique known as the intraclass correlation. Conceptually, it systematically defines for each reliability assessment situation its unique frame of reference. Essentially, with the intraclass correlation approach, one attempts to estimate the true variance and error variance components directly through a two-way analysis of variance procedure. The total variance in the scores across all observers can be attributed to three sources: systematic variance across subjects, systematic variance across observers, and random error variance. The major uniqueness of the Generalizability Theory is in its fundamental idea of reliability and error. Data gathered within the sample facets are used for the statistical analyses of the Generalizability (G) study. Classical reliability indices estimate the degree of data consistency within an undefined unidimensional universe.
