Modelos de regressão normal independente com erros de medida e dados censurados

Abstract

Measurement error models (MEM) are useful for describing different phenomena in several areas of knowledge. They are used to compare measuring devices that vary in cost, time and efficiency. Although several models consider the existence of poorly measured covariates, many of them do not consider censored observations for the response variable. On the other hand, this is fundamental since in several studies the observed response is subject to maximum and/or minimum detection limits. In this context, we extend the work of Matos et al. (2016), who developed the estimation of parameters of the model with a multivariate measurement error by using the Student-t distribution with censored observations, to a more general class of independent normal distributions (multivariate Student-t and multivariate slash). In addition to developing robust estimation and inference procedures in order to use a distribution that more efficiently accommodates outliers observations than the normal distribution, we also carry out a diagnostic study of global influence and local influence using the methodology proposed by Zhu e Lee (2001).