ON THE ROBUSTNESS OF MULTINORMAL ELIPTICAL ESTIMATORS UNDER CERTAIN CONDITIONS OF SKEWNESS, SAMPLE SIZE AND COMPLEXITY OF THE COVARIANCE STRUCTURE MODELS

Abstract

The present work focuses on elliptical estimation methods other than multinormal ones, in particular complex models such as panel longitudinal models with non-recursive latent variables. These are especially difficult in view of the complexity of the effects of this type of design and the high number of degrees of freedom. This is consistent with the majority of applied field investigations. The present work analyses accuracy in estimating standard errors and parameters with respect to the three fundamental effects encountered in the longitudinal models investigated (self-correlation or stability, non-recursiveness and cross-lagged or transversals), under certain conditions of skewness, sample size and complexity of the models.