ALGUNOS PROBLEMAS EN LA DETERMINACION DE LA CONFIABILIDAD DE INDICADORES CUALITATIVOS VIA ANALISIS DE CLASES LATENTES

Abstract

The methodology for reliability studies of quantitative variables is well known. Software is available in many statistical packages. The reality is different for qualitative variables. In 1996, Clogg and Manning proposed using latent class models for this purpose and defined some reliability indices for this framework. In this paper, we present some experiences with this new approach and we make three observations concerning (1) some negative effects of sparseness, (2) empirical nonidentifiability, and (3) the local maximum problem.