Un confronto tra metodi di identificazione di osservazioni outlier in indagini longitudinali finalizzate alla stima di una variazione: proposte teoriche e verifiche empiriche

Abstract

If the main purpose of a sampling survey consists in the estimation of a change concerning a certain quantitative variable, the presence of outliers could lead to significant biases in the estimation process. As a consequence, problems occurring are: a) how identifying outliers; b) how treating them from a statistical point of view. In this paper we tried to evaluate the technique for identifying outliers, originally proposed by Hidiroglou and Berthelot, presenting a new technique as well, that generally leads both to a significantly lower amount of units identified as outliers and of micro-data alterations. Moreover, we proposed and compared some criteria for defining the tolerance intervals. Finally, we carried out an empirical study, in which we compared 14 different ways for dealing with outliers.