The generalizability study as a method of assessing intra- and interobserver reliability in observational research

Abstract

Most researchers use interobserver agreement percentages to express the quality of their observational data. A better method is the generalizability study, which allows the variance in a set of scores to be partitioned among several sources, such as observers, occasions, subjects, and error. In this paper the practical application of generalizability theory is described and illustrated, using the frequency and duration of two behaviors observed during the mother-infant interaction. This method allows the user to assess intra- and interobserver differences and to examine the relationship between these differences and subject characteristics. © 1979 Psychonomic Society, Inc.