Generalizability theory applied to observational designs

Abstract

The application of generalizabitity theory to the general research designs is reviewed. The theory assumes that different sources of variance in a global structure permits particular applications of sampling statistical theory. Generalizability theory assumes multiple sources of measurement error in an observational design. It is possible to assess each of this sources of error and their diferent interactions. The measurement error is the effect of the fluctuations determined by the randomized choice of subjects, observers, categories, and sessions. To optimize such measure means to adapt one's design in order to decrease the sample variance determined by such facets. One example is presented where eight subjects were evaluated by two observers, in an interactive play situation, with a seven categories catalogue. Under some circuntstances, differentiating individuals, is not as important as differentiating observers, categories and sessions, in order to assume the sinzetry principle: the successive objects of measurement may be evaluated within the same design. Under this principle, each facet of the research design may be selected as an object of study and in each generalizability analysis, this facet could be considerated as an measurement instrument or evaluative condition in the study of the remaining facets