Diagnóstico de Influência Local para a obtenção de dados mascarados influentes em modelos de regressão com erros nas variáveis e propriedades assintóticas do modelo de calibração ultraestrutural

Abstract

Proficiency testing determines the performance of individual laboratories with respect to specific measurement tests and it is used to monitor the reliability of laboratory measurements. Considering a multivariate measurement error model with replicates, the ultrastructural calibration model with replicates, we will develop the asymptotic properties associated with this model in order to test the equivalence between the measurements obtained by different laboratories in proficiency testing. After adjusting a model, the diagnostic analysis is an important step in the analysis of a data set. Cook (1986) introduced a general methodology to assess the local influence of minor perturbations in the statistical model, using different types of perturbations. As a complement to the techniques for detecting discrepant observations, it is proposed the forward search by Atkinson and Riani (2000), which is a methodology to detect atypical masked observations. We propose a methodology based on the local influence of Cook (1986) and Poon and Poon (1999) with the forward search of Atkinson and Riani (2000) considering the measurement error regression models to obtain masked influential observations.