ALGUNAS CONSIDERACIONES PRÁCTICAS ACERCA DE LA ESTIMACIÓN DE PARÁMETROS EN EL MODELO CLÁSICO DE CLASES LATENTES

Abstract

In this paper the effect of some factors on the parameter estimation in the latent class model is studied by means of a simulation study, in which a classical two latent class model is used and four factors are taken into account: (1) sample size, (2) strength of relation between the manifest variables and the latent one, (3) the relative size of the latent classes, and (4) the level of noise. The response variables studied are the amount of indeterminate results, the amount of frontier solutions, the bias and the mean
square error two of the model parameter estimation. The results suggest that it is possible to find satisfactory estimations of the model parameters of latent classes, even with the samples sizes being relatively small, when all the manifest variables are strongly related to the latent one, and when there is no relatively small latent class. In addition, for each simulated condition, some suggestions related with the necessary sample size are given.