Generalización de la fiabilidad: un enfoque metaanalítico aplicado a la fiabilidad

Reliability is not a property inherent to the test, so that sentences such as “the test reliability is 0.80” are wrong. That is because reliability is a property of scores obtained in a given application o a test. Reliability generalization (RG) is a new kind of meta-analysis which enables to empirically examine the variability of the reliability estimates across different applications of a test. The RG studies are evidencing how unadvisable is the usual practice of researchers of inducing reliability from previous estimates. In this article an overview of the RG approach is presented, describing the required steps. Moreover, some of the most important statistical issues concerning RG studies are discussed, such as: (a) transforming procedures of the reliability coefficients, (b) weighting methods of the coefficients, and (c) statistical models that can be assumed.